HPE Reference Architecture for MapR Converged …...The configurations described in this Reference Architecture are based on the 5.2 release of the MapR Converged Data Platform and

HPE Reference Architecture for MapR Converged Data Platform on HPE Elastic Platform for Big Data Analytics (EPA) HPE Converged Infrastructure with MapR Converged Data Platform 5.2

Reference Architecture

Reference Architecture

Contents Executive summary ................................................................................................................................................................................................................................................................................................................................ 3 Solution overview ..................................................................................................................................................................................................................................................................................................................................... 3

HPE Workload and Density Optimized (WDO) solution ............................................................................................................................................................................................................................... 4 HPE Balanced and Density Optimized (BDO) solution .................................................................................................................................................................................................................................. 4 HPE EPA control block, network block .......................................................................................................................................................................................................................................................................... 5

HPE Pointnext Services ..................................................................................................................................................................................................................................................................................................................... 5 MapR Converged Data Platform overview ....................................................................................................................................................................................................................................................................... 6 Solution components ............................................................................................................................................................................................................................................................................................................................ 6

HPE Workload and Density Optimized solution components for single and multi-rack ................................................................................................................................................. 6 HPE Balanced and Density Optimized solution components for single and multi-rack .................................................................................................................................................. 9 Pre-deployment considerations ....................................................................................................................................................................................................................................................................................... 12 High Availability considerations....................................................................................................................................................................................................................................................................................... 14

Configuration guide for the solution .................................................................................................................................................................................................................................................................................. 14 Management/Head/Edge blocks for BDO and WDO workloads ....................................................................................................................................................................................................... 14 Workload and Density Optimized blocks for WDO ......................................................................................................................................................................................................................................... 17 Balanced and Density Optimized blocks for BDO ........................................................................................................................................................................................................................................... 19 Networking/switch selection block BDO and WDO workloads ............................................................................................................................................................................................................ 22 MapR Best Practices and Yarn configuration ...................................................................................................................................................................................................................................................... 23 MapR Best Practice ..................................................................................................................................................................................................................................................................................................................... 24

EPA WDO and BDO Reference Architecture details ........................................................................................................................................................................................................................................... 25 Single-rack Reference Architecture for WDO and BDO ............................................................................................................................................................................................................................. 25 Multi-rack Reference Architecture for WDO and BDO ................................................................................................................................................................................................................................ 28

Capacity and sizing ............................................................................................................................................................................................................................................................................................................................ 31 HPE Insight Cluster Management Utility ....................................................................................................................................................................................................................................................................... 36 Summary ...................................................................................................................................................................................................................................................................................................................................................... 38 Implementing a proof-of-concept ......................................................................................................................................................................................................................................................................................... 38 Appendix A: Bill of Materials ...................................................................................................................................................................................................................................................................................................... 38 Appendix B: Alternate parts for storage node ........................................................................................................................................................................................................................................................... 43 Appendix C: Alternate parts for compute node ....................................................................................................................................................................................................................................................... 44 Appendix D: MapR cluster tuning/optimization ...................................................................................................................................................................................................................................................... 45 Appendix E: HPE Pointnext value-added services and support .............................................................................................................................................................................................................. 47 Resources and additional links ................................................................................................................................................................................................................................................................................................ 49

Reference Architecture Page 3

Executive summary Hewlett Packard Enterprise and MapR allow you to derive new business insights from all of your data by providing a platform to store, manage, and process data at scale. This Reference Architecture provides several performance optimized configurations for deploying the MapR Converged Data Platform clusters on Hewlett Packard Enterprise infrastructure that provide a significant reduction in complexity and a recognized increase in value and performance.

The configurations described in this Reference Architecture are based on the 5.2 release of the MapR Converged Data Platform and the HPE Elastic Platform for Big Data Analytics (EPA) infrastructure; and they highlight solutions based on Workload and Density Optimized (WDO) as well as Balanced and Density Optimized (BDO) systems. These configurations have been designed and developed by Hewlett Packard Enterprise to provide the highest levels of computational performance for MapR.

HPE Elastic Platform for Big Data Analytics (EPA) is designed as a modular infrastructure foundation to address the need for a scalable multi-tenant platform, by enabling independent scaling of compute and storage through infrastructure building blocks that are optimized for density and workloads. Hewlett Packard Enterprise supports two different deployment models under this platform:

• HPE Workload and Density Optimized (WDO) system – Harnesses the power of faster Ethernet networks that enable a building block approach to independently scale compute and storage and lets you consolidate your data and workloads growing at different rates. The HPE WDO system is based on the HPE Apollo 4200 storage block and the HPE Apollo 2000 compute block.

• HPE Balanced and Density Optimized (BDO) system – Supports MapR deployments that scale compute and storage together, with some flexibility in choice of memory, processor, and storage capacity. This is primarily based on the HPE ProLiant DL380 server platform, with density optimized variants using HPE Apollo 4200 servers. In this paper we have tested with two different BDO systems, one on the HPE ProLiant DL380 server platform and another on the HPE Apollo 4200 server platform.

This Reference Architecture (RA) describes deployment options for the MapR Converged Data Platform 5.2 using the HPE Elastic Platform for Big Data Analytics - modular building blocks of compute and storage optimized for modern workloads white paper, http://h20195.www2.hpe.com/V2/GetDocument.aspx?docname=4AA6-8931ENW. This RA also provides suggested configurations that highlight the benefits of a building block approach to address the diverse processing and storage requirements typical of modern Big Data platforms.

The Hewlett Packard Enterprise software, HPE ProLiant DL380 Gen9 servers, HPE Apollo 4200 Gen9 servers, HPE Apollo 2000 Gen9 servers, and the HPE networking switches, and all of their respective configurations, that are recommended in this RA have been carefully tested with a variety of I/O, CPU, network, and memory bound workloads. The configurations included provide optimum MapReduce, YARN, Spark, Hive, and HBase computational performance, resulting in a significant performance increase at an optimal cost. The HPE EPA solutions provide excellent performance and availability, with integrated software, services, infrastructure, and management – all delivered as one proven configuration, described in more detail at hpe.com/info/hadoop . The HPE Reference Library provides a comprehensive list of technical articles on Big Data, http://h17007.www1.hpe.com/us/en/enterprise/reference-architecture/info-library/index.aspx?workload=big_data

Target audience: This document is intended for decision makers, system and solution architects, system administrators and experienced users who are interested in reducing the time to design and purchase an HPE and MapR Converged Data Platform solution. An intermediate knowledge of Apache Hadoop and scale out infrastructure is recommended. Those already possessing expert knowledge about these topics may proceed directly to the Solution components section.

Document purpose: The purpose of this document is to describe a Reference Architecture, highlighting recognizable benefits to technical audiences and providing guidance for end users on selecting the right configuration for building their MapR cluster needs.

This white paper describes testing performed in April 2017.

Solution overview HPE and MapR allow you to derive new business insights from Big Data by providing a platform to store, manage and process data at scale. This Reference Architecture provides several performance optimized configurations for deploying MapR Converged Data Platform clusters on HPE EPA configurations.

The configurations are based on the MapR Converged Data Platform edition, specifically Converged Data Platform 5.2 version and several BDO and WDO systems based on varying underlying compute infrastructure including the HPE ProLiant DL380 Gen9 server platform, HPE Apollo 4200 Gen9, and HPE Apollo 2000 system (HPE ProLiant XL170r Gen9 servers).

http://h20195.www2.hpe.com/V2/GetDocument.aspx?docname=4AA6-8931ENW

http://www.hpe.com/info/hadoop

http://h17007.www1.hpe.com/us/en/enterprise/reference-architecture/info-library/index.aspx?workload=big_data


HPE Workload and Density Optimized (WDO) solution The HPE WDO solution infrastructure blueprints are composed of five blocks: compute blocks, storage blocks, control blocks, network blocks, and rack blocks. Listed below are the blocks in a WDO solution:

• The standard compute block is one HPE Apollo 2000 chassis consisting of four HPE ProLiant XL170r Gen9 servers. In this RA we have used 3 compute blocks.

• The standard storage block is one HPE Apollo 4200 Gen9, consisting of 28x LFF HDDs or SSDs. In this we have used 4 storage blocks.

• The control block is made up of three HPE ProLiant DL360 Gen9 servers, with an optional fourth server acting as an edge or gateway node depending on the customer enterprise network requirements.

• The HPE WDO network block consists of two HPE FlexFabric 5950 48SFP28+ 10/25GbE switches and one HPE FlexFabric 5900AF-48G-4XG-2QSFP+ 1GbE switch.

• The aggregation network block consists of two HPE FlexFabric 5950-32QSFP28 (1U) switches, used when adding a third rack.

• The HPE WDO rack block consists of either a 1200mm or 1075mm rack and its accessories.

HPE Apollo 2000 compute block The Apollo 2000 provides a density optimized compute platform that doubles the density of traditional 1U servers, providing up to four 2P servers in a 2U form factor. The HPE ProLiant XL170r Gen9 server is an ideal node for compute-intensive Spark, Hive/Tez, MapReduce, YARN, and Vertica SQL on MapR and/or in-memory analytics in real-time, interactive, and/or batch workloads.

HPE Apollo 4200 storage block HPE Apollo 4200 servers are cost-effective industry–standard 2U storage-optimized servers, purpose built for Big Data with converged infrastructure and ProLiant technology that offers high density energy-efficient storage at up to 280TB per node, or 5.6PB per rack. Ideal for ETL offload workloads, the HPE Apollo 4200 is the foundation storage block for workload consolidation and is a key platform in the migration from a BDO to WDO system.

HPE Balanced and Density Optimized (BDO) solution The HPE BDO solution infrastructure blueprints are composed of four blocks: compute blocks, control blocks, network blocks and rack blocks. Listed below are the blocks in a BDO solution:

• The balanced compute block is one HPE ProLiant DL380 Gen9 server. In this RA we have used 9 balanced compute blocks.

• The density optimized compute block is one Apollo 4200 Gen9, consisting of 28x LFF HDDs. In this RA we have used 8 density optimized compute blocks.

• The control block is made up of three HPE ProLiant DL360 Gen9 servers, with an optional fourth server acting as an edge or gateway node depending on the customer enterprise network requirements.

• The HPE WDO network block consists of two HPE FlexFabric 5950 48SFP28+ 10/25GbE switches and one HPE FlexFabric 5900AF-48G-4XG-2QSFP+ 1GbE switch.

• The aggregation network block consists of two HPE FlexFabric 5950-32QSFP28 switch (1U) switches. Used when adding a third rack.

• The HPE BDO rack block consists of either a 1200mm or 1075mm rack and its accessories.

HPE ProLiant DL380 - balanced block The HPE ProLiant DL380 Gen9 (2U) is the most widely deployed MapR worker node in conventional deployments. HPE ProLiant DL380 Gen9 servers deliver the best performance and expandability in the HPE 2P rack portfolio. The ProLiant DL380 is known for its reliability, serviceability, near continuous availability and comprehensive warranty, making it the ideal node for most standard deployments. They have long been the server of choice for conventional Big Data deployments worldwide as they provide the preferred CPU core-to-spindle ratio that has typically defined symmetric Hadoop worker node configurations.

HPE Apollo 4200 – density optimized block HPE Apollo 4200 servers are cost-effective industry–standard 2U storage-optimized servers, purpose built for Big Data with converged infrastructure and ProLiant technology that offers high density energy-efficient storage at up to 280TB per node, or 5.6PB per rack. Ideal for


ETL offload workloads, the HPE Apollo 4200 is the foundation storage block for workload consolidation and is a key platform in the migration from a BDO to WDO system.

HPE EPA control block, network block HPE ProLiant DL360 Gen9 – HPE WDO/BDO control block HPE ProLiant DL360 servers include two sockets using the Intel® Xeon® E5-2600 v4 product family to provide the high performance required for management or master services on MapR clusters. The HPE EPA control block contains three HPE ProLiant DL360 Gen9 servers for use in both WDO and BDO architectures.

A fourth HPE ProLiant DL360 Gen9 server may be configured with the control block to act as an edge node. An edge node acts as a gateway between the clusters private VLAN and the external routable network. Any application that requires both external network access and cluster private network access can run on this server. When significant storage bandwidth is required, use additional HPE ProLiant DL360 Gen9 servers as edge nodes.

HPE WDO/BDO network block The HPE WDO/BDO network block consists of two (2) IRF-bonded HPE FlexFabric 5950 48SFP28+ switch for high performance and redundancy. Each provides eight 100GbE uplinks that can be used to connect to the desired network, or, in a multi-rack configuration another pair of HPE FlexFabric 5950-32QSFP28 10/25GbE switches that are used for aggregation.

The HPE WDO/BDO network block also includes a single HPE FlexFabric 5900AF-48G-4XG-2QSFP+ switch, used exclusively to provide connectivity to HPE Integrated Lights-Out (iLO) management ports, which run at or below 1GbE. The iLO network is used for system provisioning and maintenance.

Table 1. HPE Elastic Platform for Big Data Analytics – WDO/BDO network block

Component Recommended configuration

Model (2) HPE FlexFabric 5950 48SFP28+ switch

Model HPE 5900AF-48G-4XG-2QSFP+ switch

HPE offers an aggregation network block consisting of two HPE FlexFabric 5950-32QSFP28 aggregation switches. The aggregation network block can be used with either WDO or BDO solutions. Depending on the number of ports required for customer connectivity and the bandwidth needed between racks it is possible for a single pair of aggregation switches to support up to 16 racks.

Table 2. HPE Elastic Platform for Big Data Analytics – WDO/BDO aggregation network block


Model (2) HPE FlexFabric 5950-32QSFP28 switch

HPE Pointnext Services In order to simplify the build for customers, HPE provides a bill of materials in this document to allow customers to purchase this complete solution. HPE recommends that customers purchase the option of services from HPE Pointnext, as detailed in Appendix E: HPE Pointnext value-added services and support, to install and configure the operating system, verify if all firmware and versions are installed correctly, and run a suite of tests that verify that the configuration is performing optimally. Once this has been done, the customer can perform a standard MapR installation using the recommended guidelines in this document.


MapR Converged Data Platform overview The MapR Converged Data Platform delivers distributed processing, real-time analytics, and enterprise-grade requirements across cloud and on-premises environments, while leveraging the significant ongoing development in open source technologies including Spark and Hadoop. Converge-X Data Fabric powers the shared services of the Converged Data Platform including High Availability, Unified Security, Multi-tenancy, Disaster Recovery, Global Namespace, Management, Automation and real-time data access.

Figure 1. MapR Converged Data Platform

MAPR-XD MapR-XD is an exabyte-scale, reliable, globally distributed data store, delivering an organization’s data fabric for managing files, objects, and containers. MapR-XD supports the most stringent speed, scale, and reliability requirements within and across multiple edge, on-premises, and cloud environments.

MapR-XD supports storage pools for striping data write operations. Azure Virtual Machines (VMs) support multiple data disks. The maximum number of attached disks is determined by the VM size. Striping is a common practice for optimizing I/O on Azure VM disks. VM data disk configuration deserves a closer look, when deploying to Microsoft Azure. Please refer to the VM Disks section for the details.

MAPR-DB MapR-DB is an enterprise-grade, high performance, multi-model NoSQL database management system that supports real-time, operational, and analytical processing. Customers use MapR-DB to manage multiple NoSQL data models, including key-value tables, wide columns, and JSON documents, which enables faster, more efficient processing of data. MapR-DB has great scale and the strong consistency needed to deploy real-time operational apps in a globally distributed environment.

MAPR-ES MapR-ES supports processing of event-based data, including real-time data streams. MapR Streams is a publish/subscribe framework that can support the interaction of millions of producing and consuming applications at a rate of billions of events per seconds.

For detailed information on MapR Converged Data Platform, visit https://mapr.com/products/mapr-converged-data-platform. For detailed information on MapR-XD, visit https://mapr.com/products/mapr-xd/.

Solution components HPE Workload and Density Optimized solution components for single and multi-rack The HPE Workload and Density Optimized (WDO) system harnesses the power of faster Ethernet networks to independently scale compute and storage using a building block approach, and lets you consolidate your data and workloads growing at different rates. The base HPE WDO system uses the HPE Apollo 4200 as a storage block and the HPE Apollo 2000 as a compute block.

Figures 2 and 3 show a MapR network with workload and density optimized systems, with separate storage (MapR-XD) and compute (YARN, Spark and others) on a different platform. The Apollo 4200 density optimized Gen9 servers provides additional storage density (28 LFF disk drives). It is important to have dual high bandwidth network connectivity (25GbE/40GbE/100GbE) to eliminate network bottlenecks. Apollo 2000 is used for compute.

https://mapr.com/products/mapr-converged-data-platform

https://mapr.com/products/mapr-xd/


Figure 2 provides a basic conceptual diagram of HPE MapR Worker and Density Optimized single rack with HPE Apollo 4200 storage block, HPE Apollo 2000 compute block and HPE FlexFabric 5950 48SFP28+ switch network block.

Note Separate head nodes (control blocks) are not required for running CLDB, Resource Manager, Zookeeper and Job History services as these services can run on storage nodes or compute nodes in the cluster. For large clusters (100 nodes or more) using separate head nodes will improve performance and reliability of the services.

Figure 2. Basic conceptual diagram of an HPE Apollo 4200 storage block and HPE 2000 compute block WDO MapR Single-rack Reference Architecture


Figure 3 provides a basic conceptual diagram of HPE MapR RA for Workload and Density Optimized multi-rack with HPE Apollo 4200 storage blocks, HPE Apollo 2000 compute blocks and HPE FlexFabric 5950 48SFP28+ switch network block. The diagram shows multi-rack configuration in which aggregation switch on the expansion rack as part of the cluster. The HPE FlexFabric 5950 48SFP28 8QSFP28 switch has 8 QSFP28 ports for spine/ToR uplinks with 100GbE/40GbE/25GbE/10GbE connectivity support. The 32 port HPE FlexFabric 5950 32QSFP28 switch is used as aggregation switch. In a multi-rack environment, it is recommended to spread the head nodes to different racks configured in HA environment to improve high availability of the cluster.

Note It is recommended to spread the HPE FlexFabric 5950 32QSFP28 aggregation switches to different racks in multi-rack architecture in order avoid a single point of failure.

Figure 3. Basic conceptual diagram of an HPE Apollo 4200 storage block and HPE 2000 compute block WDO MapR multi-rack Reference Architecture


HPE Balanced and Density Optimized solution components for single and multi-rack The following two building blocks are recommended for HPE BDO solution:

• HPE ProLiant DL380 - balanced block

• HPE Apollo 4200 – density optimized block

Figure 4 provides a basic conceptual diagram of HPE MapR RA for balanced block single-rack with HPE ProLiant DL380 Gen9 balanced blocks and HPE FlexFabric 5950 48SFP28+ switch network block. In a multi-rack environment, it’s recommended to spread the head nodes to different racks configured in HA environment to improve high availability of the cluster.

Note Separate head nodes (control block) are not required for running CLDB, Resource Manager, Zookeeper and Job History services as these services can run on storage nodes or compute nodes in the cluster. For large clusters (100 nodes or more) using separate head nodes will improve performance and reliability of the services.

Figure 4. Basic conceptual diagram of an HPE ProLiant DL380 balanced block MapR single-rack Reference Architecture


Figure 5 provides a basic conceptual diagram of HPE MapR RA for multi-rack with HPE ProLiant DL380 Gen9 balanced blocks and HPE FlexFabric 5950 48SFP28+ switch network block. The diagram shows multi-rack configuration in which aggregation switch on the expansion rack as part of the cluster. If the cluster grows beyond 2 racks an aggregation switch will be required as shown in Figure 2. The HPE FlexFabric 5950 48SFP28 8QSFP28 switch has 8 QSFP28 ports for spine/ToR uplinks with 100GbE/40GbE/25GbE/10GbE connectivity support. The 32 port HPE FlexFabric 5950 32QSFP28 switch is used as aggregation switch. In a multi-rack environment, it’s recommended to spread the head nodes to different racks configured in HA environment to improve high availability of the cluster.

Note It is recommended to spread the HPE FlexFabric 5950 32QSFP28 aggregation switches to different racks in multi-rack architecture in order avoid a single point of failure.

Figure 5. Basic conceptual diagram of an HPE ProLiant DL380 balanced block MapR multi-rack Reference Architecture

For a full BOM listing on the products selected, refer to the Bill of Materials section of this white paper.


Figure 6 provides a basic conceptual diagram of HPE MapR RA for single-rack with HPE Apollo 4200 Gen9 density optimized blocks and HPE FlexFabric 5950 48SFP28+ switch network block. The diagram shows a rack configuration in which no aggregation switch is part of the cluster. The switch HPE FlexFabric 5950 48SFP28 8QSFP28 switch has 8 QSFP28 ports for spine/ToR uplinks with 100GbE/40GbE/25GbE/10GbE connectivity support. In a multi-rack environment, it’s recommended to spread the head nodes to different racks configured in HA environment to improve high availability of the cluster.

Figure 6. Basic conceptual diagram of an HPE ProLiant Apollo 4200 density optimized block MapR single-rack Reference Architecture


Figure 7 provides a basic conceptual diagram of HPE MapR RA for multi-rack with HPE Apollo 4200 Gen9 density optimized blocks and HPE FlexFabric 5950 48SFP28+ switch network block. The diagram shows multi-rack configuration in which aggregation switch on the expansion rack as part of the cluster. Once the cluster grows beyond 2 racks an aggregation switch will be required as shown in Figure 2. The HPE FlexFabric 5950 48SFP28 8QSFP28 switch has 8 QSFP28 ports for spine/ToR uplinks with 100GbE/40GbE/25GbE/10GbE connectivity support. The 32 port HPE FlexFabric 5950 32QSFP28 switch is used as aggregation switch. In a multi-rack environment, it is recommended to spread the head nodes to different racks configured in HA environment to improve high availability of the cluster.

Figure 7. Basic conceptual diagram of an HPE Apollo 4200 density optimized MapR multi-rack Reference Architecture

Pre-deployment considerations The operating system and the network are key factors you need to consider prior to designing and deploying a MapR cluster. The following subsections articulate the design decisions in creating the baseline configurations for the Reference Architectures.

Operating system MapR 5.2 supports 64-bit operating systems, visit http://maprdocs.mapr.com/home/InteropMatrix/r_os_matrix.html for the minimum requirements. In this RA, we have tested with Red Hat Enterprise Linux® (RHEL) 7.2.

http://maprdocs.mapr.com/home/InteropMatrix/r_os_matrix.html


Key point Hewlett Packard Enterprise recommends all HPE ProLiant systems be upgraded to the latest BIOS and firmware versions before installing the OS. HPE Service Pack for ProLiant (SPP) is a comprehensive systems software and firmware update solution, which is delivered as a single ISO image. The minimum SPP version recommended is 2015.10.0 (B). The latest version of SPP is available at: http://h17007.www1.hpe.com/us/en/enterprise/servers/products/service_pack/spp/index.aspx

Computations Unlike MR1, under YARN/MR2, there is no distinction between resources available for maps, and resources available for reduces – all MR2 resources are available for both in the form of containers. Employing Hyper-Threading increases effective core count, potentially allowing the ResourceManager to assign more cores as needed. There are several factors to consider and balance when determining the number of disks a MapR worker node requires.

Storage capacity The number of disks and their corresponding storage capacity determines the total amount of the storage capacity for your cluster.

Redundancy MapR ensures that a certain number of block copies are consistently available. This number is configurable in the block replication factor setting, which is typically set to three. If a MapR worker node goes down, MapR will replicate the blocks that had been on that server onto other servers in the cluster to maintain the consistency of the number of block copies. For example, if the NIC (Network Interface Card) on a server with 16TB of block data fails, 16TB of block data will be replicated between other servers in the cluster to ensure the appropriate number of replicas exist. Furthermore, the failure of a non-redundant ToR (Top of Rack) switch will generate even more replication traffic. MapR provides data throttling capability in the event of a node/disk failure so as to not overload the network.

I/O performance The more disks you have, the less likely it is that you will have multiple tasks accessing a given disk at the same time. This avoids queued I/O requests and incurring the resulting I/O performance degradation.

Disk configuration For management nodes, storage reliability is important and SAS drives are recommended. For worker nodes, one has the choice of SAS or SATA and as with any component there is a cost/performance tradeoff. Specific details around disk and RAID configurations will be provided in the next section.

Network Configuring a single ToR switch per rack introduces a single point of failure for each rack. In a multi-rack system such a failure will result in a very long replication recovery time as MapR rebalances storage; and, in a single-rack system such a failure could bring down the whole cluster. Consequently, configuring two ToR switches per rack is recommended for all production configurations as it provides an additional measure of redundancy. This can be further improved by configuring link aggregation between the switches. The most desirable way to configure link aggregation is by bonding the two physical NICs on each server. Port1 wired to the first ToR switch and Port2 wired to the second ToR switch, with the two switches IRF bonded. When done properly, this allows the bandwidth of both links to be used. If either of the switches fail, the servers will still have full network functionality, but with the performance of only a single link. Not all switches have the ability to do link aggregation from individual servers to multiple switches; however, the HPE FlexFabric 5950 48XGT 6QSFP28+ switch supports this through HPE Intelligent Resilient Fabric (IRF) technology. In addition, switch failures can be further mitigated by incorporating dual power supplies for the switches.

The MapR platform is rack-aware and tries to limit the amount of network traffic between racks. The bandwidth and latency provided by two bonded 25/10 Gigabit Ethernet (GbE) connections from the worker nodes to the ToR switch is more than adequate for most MapR configurations.

A more detailed white paper for Hadoop Networking best practices is available at, http://h20195.www2.hpe.com/V2/GetDocument.aspx?docname=a00004216enw.

For sizing the cluster use the HPE EPA Sizing tool, available at https://www.hpe.com/h20195/v2/GetDocument.aspx?docname=a00005868enw.

http://h17007.www1.hpe.com/us/en/enterprise/servers/products/service_pack/spp/index.aspx

http://h20195.www2.hpe.com/V2/GetDocument.aspx?docname=a00004216enw

https://www.hpe.com/h20195/v2/GetDocument.aspx?docname=a00005868enw


High Availability considerations The following are some of the High Availability (HA) features considered in this Reference Architecture configuration:

• Container Location Database (CLDB) HA – The configurations in this white paper utilize quorum-based journaling high-availability feature. For this feature, servers should have similar I/O subsystems and server profiles so that each CLDB server can potentially take the role of another. Another reason to have similar configurations is to ensure that ZooKeeper’s quorum algorithm is not affected by a machine in the quorum that cannot make a decision as fast as its quorum peers.

• ResourceManager HA – To make a YARN cluster highly available (similar to JobTracker HA in MR1), the underlying architecture of an Active/Standby pair is configured, hence the completed tasks of in-flight MapReduce jobs are not re-run on recovery after the ResourceManager is restarted or failed over. One ResourceManager is Active and one or more ResourceManagers are in standby mode waiting to take over should anything happen to the Active ResourceManager. MCS provides a simple wizard to enable HA for YARN ResourceManager.

• OS availability and reliability – For the reliability of the server, the OS disk is configured in a RAID1+0 configuration thus preventing failure of the system from OS hard disk failures.

• Network reliability – The reference architecture configuration uses the standard HPE WDO/BDO network block with two HPE FlexFabric 5950 48SFP28 8QSFP28 switches for redundancy, resiliency and scalability through using Intelligent Resilient Fabric (IRF) bonding. We recommend using redundant power supplies.

• Power supply – To ensure the servers and racks have adequate power redundancy we recommend that each server have a backup power supply, and each rack have at least two Power Distribution Units (PDUs).

Configuration guide for the solution This section will provide topologies for the deployment of Management and worker nodes for single and multi-rack clusters. Depending on the size of the cluster, a MapR deployment consists of one or more nodes running management services and a quantity of worker nodes. This section specifies which server to use and the rationale behind it.

Management/Head/Edge blocks for BDO and WDO workloads The control block is made up of three HPE ProLiant DL360 Gen9 servers, with an optional fourth server acting as an edge or gateway node depending on the customer enterprise network requirements.

Server platform: HPE ProLiant DL360 Gen9 control block The HPE ProLiant DL360 Gen9 (1U) server platform shown in Figure 8 below, is an excellent choice as the server platform for the management nodes and head nodes.

Figure 8. HPE ProLiant DL360 Gen9 server

Processor configuration for HPE ProLiant DL360 Gen9 control block The configuration features two sockets with 10-core processors of the Intel E5-2640 v4 processor family, which provide 16 physical cores and 32 Hyper-Threaded cores per server. We recommend that Hyper-Threading be turned on.

The Reference Architecture was tested using Intel Xeon E5-2640 v4 processors for the Management servers with the Web Server, ZooKeeper and head nodes with CLDB, Zookeeper, Resource Manager, History Sever services.


Drive configuration for HPE ProLiant DL360 Gen9 control block The HPE Smart Array P440ar controller is specified to drive eight 1TB 2.5” SAS disks on the management node and head node servers. Hot pluggable drives are specified so that drives can be replaced without restarting the server. Due to this design, one should configure the HPE Smart Array P440ar controller to apply the following RAID schemes:

• Management node: eight disks, with OS on (2) RAID1 set, one JBOD or RAID0 for ZooKeeper, one JBOD or RAID0 for MCS (Web Server), 4 disks in RAID1+0 for database.

• ResourceManager and CLDB servers: four disks with RAID1+0 for OS and MapR software, one JBOD or RAID0 for ZooKeeper, and two disks are available for other filesystems

• Edge node: A multi-homed edge node can be added with eight disks to provide a sufficient amount of disk capacity for a staging area for ingesting data into the MapR cluster from another subnet.

Best practice For a performance oriented solution, HPE recommends 15K SAS or SSD drives as they offer a significant read and write performance enhancement over SATA disks. The HPE Smart Array P440ar controller provides 2 port connectors per controller with each containing four SAS links. The drive cage for the HPE ProLiant DL360 Gen9 contains eight disk slots and thus each disk slot has a dedicated SAS link which ensures the server provides the maximum throughput that each drive can give. These nodes require highly reliable storage for databases, namespace storage, and edit-log journaling (ZooKeeper) services running on the servers.

Cluster isolation and access configuration It is important to isolate the MapR cluster so that external network traffic does not affect the performance of the cluster. In addition, isolation allows the MapR cluster to be managed independently from its users, ensuring that the cluster administrator is the only person who can make changes to the cluster configuration. Thus, Hewlett Packard Enterprise recommends deploying whole cluster on a private MapR cluster subnet.

Memory configuration for HPE ProLiant DL360 Gen9 control block Servers running management services such as the Hive Master, Resource Manager, CLDB and Web Server should have sufficient memory as they can be memory intensive. When configuring memory, one should always attempt to populate all the memory channels available to ensure optimum performance. The dual Intel Xeon E5-2640 v4 series processors in the HPE ProLiant DL360 Gen9 have 4 memory channels per processor which equates to 8 channels per server. The configurations for the Management and head node servers were tested with 128GB of RAM, which equated to eight 16GB DIMMs.

Management node The management node hosts the applications that submit jobs to the MapR cluster. We recommend that you install with the software components shown in Table 3.

Table 3. Management node basic software components

Software Description

Red Hat Enterprise Linux 7.2 Recommended Operating System

HPE Insight CMU 8.0 Infrastructure Deployment, Management, and Monitoring

Oracle JDK 1.8 Java Development Kit

MCS MapR Control System (Web Server)

ZooKeeper Cluster coordination service

Head nodes The head node servers contain the following software components with HA feature enabled. See the following link for more information on installing and configuring the service, http://maprdocs.mapr.com/home/AdvancedInstallation/InstallationGuide.html

http://maprdocs.mapr.com/home/AdvancedInstallation/InstallationGuide.html


Table 4 shows the head node servers base software components.

Table 4. Head node server base software components




ResourceManager YARN ResourceManager

CLDB Maintains the locations of services, containers and other cluster information

HiveServer2 Hive Service to run SQL-like ad hoc queries

ZooKeeper Cluster coordination service

Job History Server Job History for ResourceManager

Fileserver Disk Storage for MapR-XD

Warden Manage, Monitor and report on the services on each node

Edge node The edge node hosts the client configurations that submit jobs to the MapR cluster, but this optional control block depending on the customer enterprise network requirements. We recommend that you install with the software components shown in Table 5.

Table 5. Edge node basic software components




Gateway Services MapR Gateway Services (MapR-XD, YARN, MapReduce, HBase, and others)

The HPE ProLiant DL360 Gen9 (2U) as configured for the Reference Architecture (as a Master/Head/Edge node) has the following configuration:

Table 6. HPE Elastic Platform for Big Data Analytics – control block – HPE ProLiant DL360


Model HPE ProLiant DL360 Gen9 8SFF server

Processor (2) Intel E5-2640 v4 10-core processors

Memory 128GB RAM

OS disks (2) HPE 900GB 12G SAS 10K 2.5in SC ENT HDD

Controller HPE Smart Array P440ar/2G controller

Data disks (6) HPE 900GB 12G SAS 10K 2.5in SC ENT HDD

Network card HPE Ethernet 10/25Gb 2-port 640FLR-SFP28 Adapter

Note The HPE Apollo 2000 system with HPE ProLiant XL170r Gen9 servers can also serve as a control block for WDO configuration. Three of the four server slots can be used for management and head nodes, the fourth slot in the chassis can be used for an edge node.

A BOM for the management node is available in the Appendix A: Bill of Materials section of this white paper.


Workload and Density Optimized blocks for WDO HPE Apollo 2000 compute block The HPE Apollo 2000 system shown below in Figure 9, is a dense solution with four independent HPE ProLiant XL170r Gen9 hot-pluggable server nodes in a standard 2U chassis. Each HPE ProLiant XL170r Gen9 Server node is serviced individually without impacting the operation of other nodes sharing the same chassis to provide increased server uptime.

Figure 9. HPE Apollo 2000 Gen9 server

Processor selection for HPE Apollo 2000 compute block The configuration features two processors from the Intel Xeon E5-2600 v4 family. The high performance architecture and efficient configuration provides 28 physical or 56 Hyper-Threaded cores per server. For basic configurations, 12-core processors can be selected to provide 24 physical or 48 Hyper-Threaded cores per server. HPE recommends that Hyper-Threading be turned on. For this RA, we chose 2 x E5-2680v4 (14 cores/2.4GHz) CPUs.

Memory selection for HPE Apollo 2000 compute block Servers running the worker node processes should have sufficient memory for the amount of MapReduce slots configured on the server. The dual Intel Xeon E5-2600 v4 series processors in the HPE Apollo 2000 Gen9 have 4 memory channels per processor which equates to 8 channels per server. When configuring memory, one should always attempt to populate all the memory channels available to ensure optimum performance. For this RA, we chose 256GB memory (8x 32GB 2Rx4 PC4-2400T-R Kit).

Drive configuration for HPE Apollo 2000 compute block Redundancy is built into the MapR architecture and thus there is no need for RAID schemes to improve redundancy on the worker nodes as it is all coordinated and managed by MapR. Drives should use a Just a Bunch of Disks (JBOD) configuration, which can be achieved with the HPE Smart Array P840 controller by configuring each individual disk as a separate RAID0 volume. Additionally array acceleration features on the HPE Smart Array P840 should be turned off for the RAID0 data volumes. The 120GB M.2 SSD drive are configured for OS.

Compute block software components Table 7 lists the worker node software components. See the following link for more information on installing and configuring the NodeManager and Fileserver: http://maprdocs.mapr.com/home/AdvancedInstallation/PlanningtheCluster.html?hl=planning,cluster.

Table 7. Compute block base software components




NodeManager The NodeManager process for MR2/YARN


http://maprdocs.mapr.com/home/AdvancedInstallation/PlanningtheCluster.html?hl=planning,cluster


The HPE Apollo 2000 Gen9 (2U) as configured for the Reference Architecture as a worker node has the following configuration.

Table 8. HPE Elastic Platform for Big Data Analytics – WDO compute block – HPE Apollo 2000


Model HPE Apollo r2600 24SFF chassis with 4x HPE ProLiant XL170r Gen9 servers

Processor (2) Intel Xeon E5-2680 v4 14-core processors

Memory 256GB RAM

OS disks (2) HPE 480GB 6Gb SATA 2.5in MU-2 SC SSD

Controller HPE Dynamic Smart Array B140i controller


Note To add a Task Tracker Node to a cluster, follow the modifications outlined: Before you start the Warden Service (which starts the Fileserver service), add the following line to /opt/mapr/conf/mfs.conf: mfs.network.location=/compute-only

Note Customers also have the option of purchasing a second power supply for additional power redundancy. This is especially appropriate for single-rack clusters where the loss of a node represents a noticeable percentage of the cluster.

The BOM for the HPE Apollo 2000 compute block is provided in the Bill of Materials section of this white paper.

HPE Apollo 4200 storage block The HPE Apollo 4200 Gen9 server (2U) shown below in Figure 10, is an excellent choice as the server platform for the storage nodes due to exceptional storage density. The storage nodes run the Fileserver, and thus storage capacity is the important factor.

Figure 10. HPE Apollo 4200 Gen9 Server

Processor configuration for HPE Apollo 4200 storage block The configuration features two processors from the Intel Xeon E5-2600 v4 family. The high performance architecture configuration provides 28 physical or 56 Hyper-Threaded cores per server. For basic configurations, 12-core processors can be selected to provide 24 physical or 48 Hyper-Threaded cores per server. HPE recommends that Hyper-Threading be turned on. For this RA, we chose 2x E5-2680v4 (14 cores/2.4GHz) CPUs.

Memory configuration for HPE Apollo 4200 storage block The dual Intel Xeon E5-2600 v4 series processors in the HPE Apollo 4200 Gen9 have 4 memory channels per processor which equate to 8 channels per server. When configuring memory, one should always attempt to populate all the memory channels available to ensure the optimum performance. For this RA, we chose 256GB memory (8x 32GB 2Rx4 PC4-2400T-R Kit).


Drive configuration for HPE Apollo 4200 storage block When configuring MapR-XD storage disks as RAID0 data volumes on the HPE Smart Array P840ar array controller, HPE recommends turning off the array acceleration features. The first two SSDs in HPE Dual 120GB RI Solid State M.2 Kit are configured in a RAID1 set for OS. The HPE Smart Array P840ar controller provides two port connectors per controller with each containing eight SAS links. For a performance oriented solution, we recommend SAS drives as they offer a significant read and write throughput performance enhancement over SATA disks. For this RA, we chose 3TB LFF SATA drives

Storage block software components Table 9 lists the storage node software components. See the following link for more information on installing and configuring the Fileserver: http://maprdocs.mapr.com/home/AdvancedInstallation/PlanningtheCluster.html?hl=planning,cluster

Table 9. Storage node base software components





The HPE Apollo 4200 Gen9 (2U) as configured for the Reference Architecture as a worker node has the following configuration:

Table 10. HPE Elastic Platform for Big Data Analytics – WDO storage block – HPE Apollo 4200


Model HPE Apollo 4200 Gen9 24LFF server, plus optional LFF rear 4HDD cage


Memory 256GB RAM

OS disks HPE Dual 120GB RI Solid State M.2 kit

Controller HPE Smart Array P840ar/2G FIO controller

Data disks (28) HPE 4TB 6G SATA 7.2k 3.5in MDL LP HDD


A BOM for the management node is available in the Bill of Materials section of this white paper.

Balanced and Density Optimized blocks for BDO The worker nodes run the Fileserver, NodeManager and YARN container processes and thus storage capacity and compute performance are important factors.

HPE ProLiant DL380 balanced block The HPE ProLiant DL380 Gen9 (2U) shown below in Figure 11, is an excellent choice as the server platform for the worker nodes. For ease of management we recommend a homogenous server infrastructure for your worker nodes.

Figure 11. HPE ProLiant DL380 Gen9 server



Processor configuration for HPE ProLiant DL380 balanced block The configuration features 2 processors from the Intel Xeon E5-2680 v4 family. This provides 24 physical or 48 Hyper-Threaded cores per server with 2x E5-2680 v4 (14 cores/2.5 GHz) CPUs.

Memory configuration for HPE ProLiant DL380 balanced block Servers running the worker node processes should have sufficient memory for the number of yarn containers configured on the server. The dual Intel Xeon E5-2680 v4 series processors in the HPE ProLiant DL380 Gen9 have 8 memory channels per processor which equate to 16 channels per server. When configuring memory, one should always attempt to populate all the memory channels available to ensure optimum performance.

With the advent of Spark and YARN, the memory requirement has gone up significantly to support a new generation of Hadoop applications. For this Reference Architecture, we chose 256GB memory (16x HPE 16GB 2Rx4 PC4-2400T-R Kit).

Best practice To ensure optimal memory performance and bandwidth, we recommend using 8x 16GB DIMMs, one in each memory channel. For any applications requiring more than 256GB capacity, we recommend going with 32GB DIMMs, not to populate the third slot on the memory channels to maintain full memory channel speed.

Drive configuration for HPE ProLiant DL380 balanced block Redundancy is built into the MapR Converged Data Platform and thus there is no need for RAID schemes to improve redundancy on the worker nodes as it is all coordinated and managed by MapR. Balanced block drives should use a Just a Bunch of Disks (JBOD) configuration, which can be achieved with the HPE Smart Array P840 controller by configuring each individual disk as a separate RAID0 volume. Additionally, array acceleration features on the HPE Smart Array P840 should be turned off for the RAID0 data volumes. The 340GB M.2 SSD drives are configured for OS.

The HPE Smart Array P840 controller provides two port connectors per controller with each containing 8 SAS links.

Best practice For a performance oriented solution HPE recommends 15K SAS or SSD drives as they offer a significant read and write performance enhancement over SATA disks, which improves the performance on I/O intensive workloads.

Balanced block software components Table 11 lists the worker node software components. See the following link for more information on installing and configuring the NodeManager and Fileserver: http://maprdocs.mapr.com/home/AdvancedInstallation/PlanningtheCluster.html?hl=planning,cluster

Table 11. Worker node base software components

software description



NodeManager The NodeManager process for MR2/YARN




The HPE balanced block is configured for the Reference Architecture as a worker node has the following configuration

Table 12. HPE Elastic Platform for Big Data Analytics – BDO worker node block – HPE ProLiant DL380


Model HPE ProLiant DL380 Gen9 12LFF server, plus optional LFF rear 3HDD cage


Memory 256GB RAM

OS disks HPE Dual 340GB RI-2 Solid State M.2 kit

Controller HPE Smart Array P840/4G controller

Data disks (15) HPE 4TB 6G SATA 7.2k 3.5in SC MDL HDD


The BOM for the HPE ProLiant DL380 balanced block is provided in the Bill of Materials section of this white paper. HPE Apollo 4200 – density pptimized block The HPE Apollo 4200 Gen9 server (2U) shown below in Figure 12, provides exceptional storage density with the compute required to perform symmetric MapReduce functions. The storage nodes run the Fileserver, and thus storage capacity is the important factor.

Figure 12. HPE Apollo 4200 Gen9 Server

Processor configuration for HPE Apollo 4200 density optimized block The configuration features has 2 processors from the Intel Xeon E5-2600 v4 family. The configuration provides 28 physical or 56 Hyper-Threaded cores per server. For basic configurations, 12-core processors can be selected to provide 24 physical or 48 Hyper-Threaded cores per server. HPE recommends that Hyper-Threading be turned on. For this RA, we chose 2x E5-2680v4 (14 cores/2.4GHz) CPUs.

Memory configuration for HPE Apollo 4200 density optimized block Servers running the worker node processes should have sufficient memory for the amount of MapReduce Slots configured on the server. The dual Intel Xeon E5-2600 v4 series processors in the HPE Apollo 4200 Gen9 have 4 memory channels per processor which equate to eight channels per server. When configuring memory, one should always attempt to populate all the memory channels available to ensure the optimum performance. A base configuration of 256GB is recommended and for certain high memory capacity applications 256GB is recommended. For this RA, we chose 256GB memory (8x 32GB 2Rx4 PC4-2400T-R Kit).

Drive configuration for HPE Apollo 4200 density optimized block When configuring MapR-XD storage disks as RAID0 data volumes on the HPE Smart Array P840ar array controller, HPE recommends turning off the array acceleration features. The first 2 SSDs in HPE Dual 120GB RI Solid State M.2 Kit are configured in a RAID1 set for OS. HPE Smart Array P840ar controller provides two port connectors per controller with each containing eight SAS links. For a performance oriented solution, we recommend SAS drives as they offer a significant read and write throughput performance enhancement over SATA disks. For this RA, we chose 3TB LFF SATA drives.


Density optimized block software components Table 13 lists the storage node software components. See the following link for more information on installing and configuring the Fileserver: http://maprdocs.mapr.com/home/AdvancedInstallation/PlanningtheCluster.html?hl=planning,cluster

Table 13. Storage node base software components

software description




The HPE Apollo 4200 Gen9 (2U) as configured for the Reference Architecture as a worker node has the following configuration:

Table 14. HPE Elastic Platform for Big Data Analytics – BDO worker node block – HPE Apollo 4200


Model HPE Apollo 4200 Gen9 24LFF server, plus optional LFF rear 4HDD cage


Memory 256GB RAM

OS disks HPE Dual 120GB RI Solid State M.2 kit

Controller HPE Smart Array P840ar/2G FIO controller

Data disks (28) HPE 4TB 6G SATA 7.2k 3.5in MDL LP HDD


A BOM for the management node is available in the Bill of Materials section of this white paper.

NOTE For disk configuration refer to the MapR Best Practices section of this Reference Architecture.

Networking/switch selection block BDO and WDO workloads MapR clusters contain two types of switches, namely ToR switches and aggregation switches. ToR switches route the traffic between the nodes in each rack and aggregation switches route the traffic between the racks.

HPE FlexFabric 5950 32QSFP28 switch (JH321A)

Figure 13. HPE FlexFabric 5950 switch



The HPE FlexFabric 5950 switch provides advanced features and high performance in a top-of-rack, data center switch architecture. Consisting of a 1U 32-port 100GbE QSFP28 switch, the 5950 brings high density to a small footprint. While 10-unit IRF reduces management complexities by up to 88%, it also delivers <50 msec convergence time. You can rely on the FlexFabric 5950 switch to improve switch utilization and lower TCO, while delivering business resilience, high availability. This switch can be used for high-density 100GbE/40GbE/25GbE/10GbE spine/ToR connectivity. 100GbE ports may be split into four 25GbE ports and can also support 40GbE which can be split into four by 10GbE for a total of 128 25/10GbE ports.

The HPE FlexFabric 5950 48SFP28 8QSFP28 switch is a high density ToR switch available as a 1RU 48-port 25GbE SFP28 form factor. This switch can be used for high-density 10GbE/25GbE ToR with 100GbE/40GbE/25GbE/10GbE spine/ToR connectivity. 100GbE ports may be split into four 25GbE ports and can also support 40GbE which can be split into four by 10GbE for a total of 80 25/10GbE ports.

Table 15. HPE Network switches and configuration details

I/O ports and slots Memory and processor Throughput

Routing/Switching capacity

HPE FlexFabric 5940 48XGT 6QSFP+ switch (JH394A)

48 1/10GBASE-T ports 6 QSFP+ 40GbE ports

1 GB flash; Packet buffer size: 12.2 MB, 4 GB SDRAM up to 1071 Mpps 1440 Gbps

HPE FlexFabric 5950 48SFP28 8QSFP28 switch (JH402A)

48 SFP28 25GbE ports 8 QSFP28 100GbE ports

1 GB flash; Packet buffer size: 16 MB, 4 GB SDRAM up to 2796 Mpps 3200 Gbps

HPE FlexFabric 5900AF 48G 4XG 2QSFP+ switch (JG510A)

48 autosensing 10/100/1000 ports 4 fixed 1000/10000 SFP+ ports 2 QSFP+ 40-GbE ports

512 MB flash; Packet buffer size: 9 MB, 2 GB SDRAM

up to 250 Mpps (64-byte packets) 336 Gbps

Mellanox Spectrum™ SN2100 (2) 16-port 100GbE switch 32 10/25/40/50/56/100GbE 8 GB System; Packet buffer size: 16 MB;

16 GB SSD up to 4760 Mpps 3200 Gbps

MapR Best Practices and Yarn configuration Every MapR installation requires services to manage jobs and applications. JobTracker and TaskTracker manage MapReduce v1 jobs. ResourceManager and NodeManager manage MapReduce v2 and other applications that can run on YARN. In addition, MapR requires the ZooKeeper service to coordinate the cluster, and at least one node must run the CLDB service. The WebServer service is required if the browser-based MapR Control System will be used.

The following are guidelines about which services to separate on large clusters:

• JobTracker and Resource Manager on ZooKeeper nodes: Avoid running the JobTracker and ResourceManager service on nodes that are running the ZooKeeper service. On large clusters, the JobTracker and ResourceManager services can consume significant resources.

• MySQL on CLDB nodes: Avoid running the MySQL server that supports the MapR Metrics service on a CLDB node. Consider running the MySQL server on a machine external to the cluster to prevent the MySQL server’s resource needs from affecting services on the cluster.

• TaskTracker on CLDB or ZooKeeper nodes: When the TaskTracker service is running on a node that is also running the CLDB or ZooKeeper services, consider reducing the number of task slots that this node's instance of the TaskTracker service provides.

• Webserver on CLDB nodes: Avoid running the webserver on CLDB nodes. Queries to the MapR Metrics service can impose a bandwidth load that reduces CLDB performance.

• JobTracker: Run the JobTracker services on dedicated nodes for clusters with over 250 nodes.

• ResourceManager: Run the ResourceManager services on dedicated nodes for clusters with over 250 nodes.

Yarn Configuration For configuring YARN, update the default values of the following attributes with ones that reflect the cores and memory available on a worker node.

• yarn.nodemanager.resource.memory-mb - Defines the memory available to processing Yarn containers on the node in MB.

• yarn.nodemanager.resource.cpu-vcores - Defines the number of CPUs available to process YARN containers on the node.


While configuring YARN for MapReduce jobs make sure that the following attributes have been specified with sufficient vcores and memory. They represent resource allocation attributes for map and reduce containers. Note that the optimum values for these attributes depend on the nature of workload/use case.

• mapreduce.map.memory.mb - Defines the container size for map tasks in MB.

• mapreduce.reduce.memory.mb - Defines the container size for reduce tasks in MB.

In mapred-site.xml, the following properties define the number of disks a MapReduce task requires. On nodes, this value should be 0 so that the number of MapReduce tasks spawned does not depend on the number of disks present on the node.

• mapreduce.mapr.disk - Defines the number of disks a map task requires. Set to 0 to disable checking the value.

• mapreduce.reduce.disk - Defines the number of disks that a reduce task requires. Set to 0 to disable checking the value.

Similarly, specify the appropriate size for map and reduce task heap sizes using the following attributes:

• mapreduce.map.java.opts - Java options for map tasks.

• mapreduce.reduce.java.opts - Java options for reduce tasks.

MapR Best Practice MapR-XD is configured using 4 disks per storage pool so that all 28 disks are utilized for MapR-XD.

Syntax: /opt/mapr/server/disksetup <options><disk list file>

Setting up the Storage Pool specified in the file /tmp/disks.txt: /opt/mapr/server/disksetup -F -W 4 /tmp/disks.txt

Options: -F: Forces formatting of all specified disks

-W: Specifies the number of disks per storage pool.

For more information refer to: http://maprdocs.mapr.com/home/ReferenceGuide/disksetup.html?hl=disksetup

MapR Best Practice Isolating CLDB nodes In a large cluster (100 nodes or more) create CLDB-only nodes to ensure high performance. This configuration also provides additional control over the placement of the CLDB data, for load balancing, fault tolerance, or high availability (HA). Setting up CLDB-only nodes involves restricting the CLDB volume to its own topology and making sure all other volumes are on a separate topology.

Isolating ZooKeeper Nodes For large clusters (100 nodes or more), isolate ZooKeeper on nodes that do not perform any other function. Isolating ZooKeeper enables the node to perform its functions without competing for resources with other processes. Installing a ZooKeeper-only node is similar to any typical node installation, but with a specific subset of packages.

Warning Do not install the Fileserver package on an isolated ZooKeeper node in order to prevent MapR from using this node for data storage.

Refer to MapR best practices, for more information: http://maprdocs.mapr.com/home/ReferenceGuide/BestPractices.html

http://maprdocs.mapr.com/home/ReferenceGuide/disksetup.html?hl=disksetup

http://maprdocs.mapr.com/home/ReferenceGuide/BestPractices.html


EPA WDO and BDO Reference Architecture details The following sections illustrate a reference progression of MapR clusters from a single-rack to a multi-rack configuration. Best practices for each of the components within the configurations specified have been articulated earlier in this document.

Single-rack Reference Architecture for WDO and BDO The single-rack MapR Reference Architecture (RA) is designed to perform well as a single-rack cluster design but also form the basis for a much larger multi-rack design. When moving from the single-rack to multi-rack design, one can simply add racks to the cluster without having to change any components within the single-rack. The RA reflects the following.

Single-rack network block The HPE FlexFabric 5950 48SFP28 8QSFP28 switch is a high density ToR switch available as a 1RU 48-port 25GbE SFP28 form factory. This switch can be used for high-density 10GbE/25GbE ToR with 100GbE/40GbE/25GbE/10GbE spine/ToR connectivity. 100GbE ports may be split into four 25GbE ports and can also support 40GbE which can be split into four by 10GbE for a total of 80 25/10GbE ports. The HPE FlexFabric 5950 48SFP28 8QSFP28 switch includes 8 100GbE uplinks which can be used to connect the switches in the rack into the desired network or to the 100GbE HPE FlexFabric 5950 32QSFP28 aggregation switch. Keep in mind that if IRF bonding is used, it requires 2x 100GbE ports per switch, which would leave 6x 100GbE ports on each HPE FlexFabric 5950 48SFP28 8QSFP28 switch for uplinks.

Staging data In addition, once the MapR cluster is on its own private network one needs to think about how to be able to reach the MapR-XD data store in order to ingest data. The MapR-XD client needs the ability to reach every MapR data node in the cluster in order to stream blocks of data onto MapR-XD. The Reference Architecture provides below options to do this.

One option is to use the already multi-homed edge node, which can be configured to provide a staging area for ingesting data into the MapR cluster from another subnet.

Note The benefit of using dual-homed edge nodes to isolate the in-cluster MapR traffic from the ETL traffic flowing to the cluster is often debated. One benefit of doing so is better security; however, the downside of a dual-homed network architecture is ETL performance/connectivity issues, since a relatively few number of nodes in the cluster are capable of ingesting data.

Power and cooling In planning for large clusters, it is important to properly manage power redundancy and distribution. To ensure the servers and racks have adequate power redundancy we recommend that each server have a backup power supply, and each rack have at least two Power Distribution Units (PDUs). There is an additional cost associated with procuring redundant power supplies.

Best practice For each server, HPE recommends that each power supply is connected to a different PDU than the other power supply on the same server. Furthermore, the PDUs in the rack can each be connected to a separate data center power line to protect the infrastructure from a data center power line failure.

Additionally, distributing the server power supply connections evenly to the in-rack PDUs, as well as distributing the PDU connections evenly to the data center power lines, ensures an even power distribution in the data center and avoids overloading any single data center power line. When designing a cluster, check the maximum power and cooling that the data center can supply to each rack and ensure that the rack does not require more power and cooling than is available.


Reference architecture for WDO with HPE Apollo 2000 compute and HPE Apollo 4200 storage blocks Refer to Figure 14 for a rack-level view of the single-rack MapR Reference Architecture for this solution with HPE Apollo 4200 storage and HPE Apollo 2000 compute blocks.

For more information on configuration for the Workload and Balanced Optimization solution refer to the HPE Workload and Density Optimized configurations section.

Figure 14. Single-rack MapR Reference Architecture – Rack-level view


Reference Architecture for standard BDO with HPE ProLiant DL380 balanced block Refer to Figure 15 for a rack-level view of the single-rack MapR Reference Architecture for this solution with HPE ProLiant DL380 balanced block.

For more information on configuration for the Balanced and Density Optimization solution refer to section HPE Balanced and Density Optimized configurations – Balanced Block



Reference Architecture for density optimized BDO Rack with HPE Apollo 4200 – Density Optimized block Refer to Figure 16 for a rack-level view of the single-rack MapR Reference Architecture for this solution with HPE Apollo 4200 Density Optimized block

For more information on configuration for the Balanced and Density Optimization solution refer to section HPE Balanced and Density Optimized configurations – Density Optimized Block


Multi-rack Reference Architecture for WDO and BDO The multi-rack design assumes the single-rack RA cluster design is already in place and extends its scalability. The single-rack configuration ensures the required amount of management services are in place for large scale out. For multi-rack clusters, one simply adds expansion racks of a similar configuration to the single-rack configuration as shown in Figure 14 to Figure 16. This section reflects the design of those racks, and Figures 17 and 18 show the rack-level view of the multi-rack architecture.

Multi-rack network block The HPE FlexFabric 5950 48SFP28 8QSFP28 switch is a high density ToR switch available as a 1RU 48-port 25GbE SFP28 form factory. This switch can be used for high-density 10GbE/25GbE ToR with 100GbE/40GbE/25GbE/10GbE spine/ToR connectivity. 100GbE ports may be split into four 25GbE ports and can also support 40GbE which can be split into four by 10GbE for a total of 80 25/10GbE ports. The HPE FlexFabric 5950 48SFP28 8QSFP28 switch includes 8 100GbE uplinks which can be used to connect the switches in the rack into the desired network or to the 100GbE HPE FlexFabric 5950 32QSFP28 aggregation switch. Keep in mind that if IRF bonding is used, it requires 2x 100GbE ports per switch, which would leave 6x 100GbE ports on each HPE FlexFabric 5950 48SFP28 8QSFP28 switch for uplinks.


Multi-rack architecture for WDO with HPE Apollo 4200 storage block and HPE Apollo 2000 compute block The expansion rack for the EPA WDO RA contains nine HPE Apollo 4200 storage blocks, eight HPE Apollo 2000 compute blocks and two HPE FlexFabric 5950 48SFP28+ switches network block within a 42U rack; When moving from a single-rack to a multi-rack solution, you simply add racks without having to change any components within the base rack expect the control blocks.

The HPE Apollo 4200 Gen9 in the rack all are configured as storage Nodes and Apollo 2000 Gen9 in the rack all configured as worker nodes in the cluster, as all required management processes are already configured in the base rack. Aside from the OS, each worker node typically runs, NodeManager and Fileserver and each storage node runs datanode (Fileserver) (and HBaseRegionServer if you are using HBase).

Note In a multi-rack configuration we recommend moving the Master nodes onto different racks for better resiliency. Also move one of the aggregation switches to a different rack for better resiliency using IRF.

For more information on configuration for the Workload and Balanced Optimization solution refer to the HPE Workload and Density Optimized configurations section.

Figure 17. Multi-rack Reference Architecture for work and density optimized– Rack-level view


Note There is no need for an aggregation switch when a second rack is added to the base rack. The existing HPE FlexFabric 5900 switches in the racks can be used to network between both (base racks). Figure 3, 5 and 7 shows the network connection, using an additional aggregation switch, when a third rack (expansion rack) is added to the base racks.

Multi-rack architecture for BDO with HPE Apollo 4200 Density Optimized block The expansion rack for the EPA density optimized block RA contains sixteen HPE Apollo 4200 and two HPE FlexFabric 5950 48SFP28+ switches as network block within a 42U rack; When moving from a single-rack to a multi-rack solution, you simply add racks without having to change any components within the base rack expect the control blocks.

For more information on configuration for the Balanced and Density Optimization solution refer to section HPE Balanced and Density Optimized configurations – Density Optimized Block

Figure 18. Multi-rack MapR Reference Architecture for density optimized – Rack-level view


Capacity and sizing MapR cluster storage sizing requires careful planning and identifying the current and future storage and compute needs. Use the following as general guidelines for data inventory:

• Sources of data

• Frequency of data

• Raw storage

• Processed MapR-XD storage

• Replication factor

• Default compression turned on

• Space for intermediate files

Performance Graph The graph from Figure 19 to Figure 21 shows the performance results obtained for TeraSort with 1TB load for different workloads like BDO balanced, BDO density optimized, WDO and WDO multi-rack. The hardware configuration for system is shown in below tables from Table 16 to Table 18. The graph shows incremental gains of WDO over the BDO system, where in WDO we can separate the compute and storage in the cluster. The WDO systems allows for independent scaling of Storage and Compute utilizing the power of faster Ethernet networks while accommodating the independent growth of data and compute workloads. With the separation of storage (MapR-XD) and compute (YARN) functionality with WDO architecture, one can notice with offloading the compute function from the storage nodes. There is a performance advantage of ~10% as the compute functionality (YARN) is offloaded to compute nodes. You can reduce the amount of memory you need on the Data nodes and also install low CPU specs on these nodes as they are mainly used for storage. The compute node can be installed with high CPU specs and large memory to handle the computational load on the server. As you can independently scale both storage and compute, as the capacity and requirements change you can add those nodes to the cluster.

BDO Balanced block TeraSort results are with 9 HPE ProLiant DL380 Gen9 cluster servers with [email protected], 256GB RAM, 15 LFF 4TB SATA drives.

BDO Density Optimized block TeraSort results are with 8 Apollo 4200 cluster servers with [email protected], 256GB RAM, 28 LFF 4TB SATA drives.

BDO Density Optimized block TeraSort results are with 4 Apollo 4200 cluster servers with [email protected], 256GB RAM, 28 LFF 4TB SATA drives.

WDO TeraSort results are with 4 Apollo 4200 storage blocks [email protected], 256GB RAM, 28 LFF 4TB SATA drives, 12 Apollo 2000 compute block [email protected], 256GB RAM, 2 480GB SFF SATA SDD drives.

WDO multi-rack TeraSort results are with two racks of 4 Apollo 4200 storage blocks [email protected], 256GB RAM, 28 LFF 4TB SATA drives, 12 Apollo 2000 compute block [email protected], 256GB RAM, 2 480GB SFF SATA SDD drives in a cluster configuration.

The CPU utilization on BDO Balanced and BDO Density Optimized blocks were 46% and 65% average for the duration of the test. With WDO tests the CPU utilization on the storage nodes was reduced to <10% whereas the CPU utilization on compute nodes was > 45%.

NOTE The results are with replication 3 and might vary depending on the configuration.

mailto:[email protected]








The figure 19 shows the performance difference between BDO balanced and BDO density optimized block configuration while doubling the capacity in the same space you also lower $/TB because you need half the number of nodes which come with their associated CPU/MEM/OS/MapR license/support costs. You can achieve 25% more performance as shown by the graph below and also increase the throughput by 40%. The throughput numbers are calculated by the number of seconds to complete Terasort 1TB run in MB/sec.

Figure 19. Performance Graph for BDO balanced, BDO density optimized

The graph 20 shows the performance between single and multi-rack configuration with near linear scale as the nodes are doubled from one rack to another.

Figure 20. Performance Graph for WDO single-rack and WDO multi-rack

0

20

40

60

80

100

120

140

160

180

200

0.00

20.00

40.00

60.00

80.00

100.00

120.00

140.00

BDO balanced block BDO Density-optimized block

Thro

ughp

ut p

er n

ode

MB/

sec

Perc

enta

ge

MapR BDO Terasort 1TB performance

Perf Diff MB/sec

020406080

100120140160180200

WDO single rack WDO multi-rack

Perc

enta

ge

MapR WDO Multi-rack Terasort 1TB performance

Perf Diff


The graph in figure 21 shows the performance with 4 node BDO density optimized configuration with WDO single-rack configuration. As the 4 node Apollo 4200 BDO is running storage and compute services on the nodes. The system is constrained by the compute. Once you separate the compute with 12 node Apollo 2000 you are able to improve the performance by 2.5 times using the same storage nodes.

Figure 21. Performance Graph for BDO density optimized 4 node and WDO single-rack

HPE Balanced and Density Optimized configurations – Balanced Block Table 16 provides configurations for BDO balanced HPE ProLiant DL380 Gen9 servers, scaling from a POC starter kit up to a multi-rack configuration. Building block quantities are provided in parenthesis in the appropriate row. Usable capacity is determined using a replication level of 3, 1:1 compression ratio and overhead of 25%.

Table 16. BDO configuration – HPE ProLiant DL380 balanced

POC Starter Half Rack Full Rack Multi Rack

Control block (3) HPE ProLiant DL360 (3) HPE ProLiant DL360 (3) HPE ProLiant DL360 (3) HPE ProLiant DL360

Compute block (3) HPE ProLiant DL380 with 4TB HDD

(8) HPE ProLiant DL380 with 4TB HDD



Network block (2) HPE 5950 48SFP28 switch

(1) HPE 5900AF-48G switch

(2) HPE 5950 48SFP28 switch






Aggregation network block

(2) HPE 5950 32QSFP28 switch

Rack block HPE 1075mm rack HPE 1075mm rack HPE 1075mm rack (3) HPE 1075mm rack

Specifications

Raw disk capacity (TB) 180 480 960 1920

Usable disk capacity (TB)

45 120 240 480

Compute Performance (SpecInt per rack)

3870 10320 20640 41280

Compute performance (SpecInt per U)

322.5 469 543 550

Storage performance (MB/s per U)

412.5 600 694 704

0

100

200

300

BDO 4 node Density-optimized block WDO single rack

Perc

enta

ge

MapR 4 node storage Terasort 1TB performance

BDO and WDO

Perf Diff


HPE Balanced and Density Optimized configurations – Density Optimized Block Table 17 provides example blueprints for storage optimized BDO configurations. These configurations leverage the Apollo 4200 density optimized block in a BDO model. The specifications show approximately 85% more storage density per U when leveraging the Apollo 4200 compared to the HPE ProLiant DL380. Usable capacity is determined using a replication level of 3, 1:1 compression ratio and overhead of 25%.

Table 17. BDO configuration – HPE Apollo 4200 density optimized


Control block (3) HPE ProLiant DL360 (3) HPE ProLiant DL360 (3) DL360 (3) HPE ProLiant DL360

Compute block (3) Apollo 4200 with 4TB HDD (8) Apollo 4200 with 4TB HDD (16) Apollo 4200 with 4TB HDD (32) Apollo 4200 with 4TB HDD







(4) HPE 5950 48SFP28 switch (2) HPE 5900AF-48G switch

Aggregation network block (2) HPE 5950 32QSFP28 switch


Specifications


Usable disk capacity (TB) 84 224 398 896

Compute Performance (SpecInt per rack)

3870 10320 20640 41280


322.5 469 543 550


770 1120 1296.8 1314

HPE Workload and Density Optimized configurations Table 18 provides WDO configurations using the standard WDO compute and storage blocks. Since the compute and storage tiers are disaggregated there is more compute capacity and storage density as compared to the standard BDO configurations. This is by design as the WDO model provides a foundation for running multiple disparate services without having to have separate standalone MapR clusters and the resulting duplication of data. Usable capacity is determined using a replication level of 3, 1:1 compression ratio and overhead of 25%.

Table 18. WDO configuration – WDO HPE Apollo 2000 compute and HPE Apollo 4200 storage


Control block (3) HPE ProLiant DL360 (3) HPE ProLiant DL360 (3) HPE ProLiant DL360 (3) HPE ProLiant DL360

Compute block (3) XL170r (16) XL170r (32) XL170r (96) XL170r

Storage block (3) Apollo 4200 with 4TB HDD (5) Apollo 4200 with 4TB HDD (10) Apollo 4200 with 4TB HDD (30) Apollo 4200 with 4TB HDD









Aggregation network block (2) HPE 5950 32QSFP28 switch


Specifications


Usable disk capacity (TB) 85 144 288 863

Compute performance (SpecInt per rack)

7320 26390 52780 158340


542 1099.6 1123 1266


922 1355 1383.8 1561


Processor options For workloads that are CPU intensive it is recommended to choose higher capacity processors with more cores. Tables 19 and 20 show alternate CPUs for the selected BDO and WDO.

Table 19. CPU recommendations for BDO

CPU Description

2x E5-2660 v3 Min configuration (10 cores/2.6GHz)

2x E5-2680 v4 Base configuration (14 cores/2.5GHz)

2x E5-2690 v4 Enhanced (14 cores/2.6GHz)

Table 20. CPU recommendations for WDO

CPU Description

2x E5-2680 v3 Min configuration (12 cores/2.6GHz)

2x E5-2690 v4 Base configuration (14cores/2.5GHz)

2x E5-2697 v4 Enhanced (18 cores/2.6GHz)

See Appendix C: Alternate parts for compute node, Table C and Table D -1 for BOM details on various CPU choices.

Memory options When calculating memory requirements, remember that Java uses up to 10% of memory to manage the virtual machine. HPE recommends to configure MapR to use strict heap size restrictions to avoid memory swapping to disk.

It is important to optimize RAM for the memory channel width. For example, when using dual-channel memory, each machine should be configured with pairs of DIMMs. With triple-channel memory each machine should have triplets of DIMMs. Similarly, quad-channel DIMMs should be in groups of four. Table 21 and 22 shows the recommended memory configurations.

Table 21. Memory recommendations for BDO

Memory Description

128GB 8x HPE 16GB 2Rx4 PC4-2133P-R Kit Base configuration

256GB 16x HPE 16GB 2Rx4 PC4-2133P-R Kit High capacity configuration

Table 22. Memory recommendations for WDO

Memory Description

128GB 8x HPE 16GB 2Rx4 PC4-2133P-R Kit Min configuration

256GB 16x HPE 16GB 2Rx4 PC4-2133P-R Kit Base configuration

512 GB 16x HPE 32GB 2Rx4 PC4-2133P-R Kit High Configuration

See Appendix C: Alternate parts for compute node, Table C-2 for BOM details on alternate memory configurations.


Storage options For workloads such as ETL and similar long running queries where the amount of storage is likely to grow, it is recommended to pick higher capacity and faster drives. Table 23 shows alternate storage options for the selected HPE ProLiant DL380.

Table 23. HDD recommendations

HDD Description

2/3/4/6/8TB LFF SATA 7.2K Base configuration

2/3/4/6/8TB LFF SAS 15K Performance configuration

See Appendix C: Alternate parts for compute node, Table C-3 for BOM details on alternate memory configurations.

HPE Insight Cluster Management Utility HPE Insight Cluster Management Utility (CMU) is an efficient and robust hyper-scale cluster lifecycle management framework and suite of tools for large Linux clusters such as those found in High Performance Computing (HPC) and Big Data environments. A simple graphical interface enables an “at-a-glance” real-time or 3D historical view of the entire cluster for both infrastructure and application (including MapR) metrics, provides frictionless scalable remote management and analysis, and allows rapid provisioning of software to all nodes of the system. HPE Insight CMU makes the management of a cluster more user friendly, efficient, and error free than if it were being managed by scripts, or on a node-by-node basis. HPE Insight CMU offers full support for HPE iLO 2, iLO 3, iLO 4 and LO100i adapters on all HPE ProLiant servers in the cluster.

Best practice HPE recommends using HPE Insight CMU for all MapR clusters. HPE Insight CMU allows one to easily correlate MapR metrics with cluster infrastructure metrics, such as CPU Utilization, Network Transmit/Receive, Memory Utilization and I/O Read/Write. This allows characterization of MapR workloads and optimization of the system thereby improving the performance of the MapR cluster. HPE Insight CMU Time View Metric Visualizations will help you understand, based on your workloads, whether your cluster needs more memory, a faster network or processors with faster clock speeds. In addition, HPE Insight CMU also greatly simplifies the deployment of MapR, with its ability to create a golden Image from a node and then deploy that image to up to 4000 nodes. HPE Insight CMU is able to deploy 800 nodes in 30 minutes.

HPE Insight CMU is highly flexible and customizable, offers both GUI and CLI interfaces supports for Ansible, and can be used to deploy a range of software environments, from simple compute farms to highly customized, application-specific configurations. HPE Insight CMU is available for HPE ProLiant and HPE BladeSystem servers, and is supported on a variety of Linux operating systems, including Red Hat Enterprise Linux, SUSE Linux Enterprise Server, CentOS, and Ubuntu. HPE Insight CMU also includes options for monitoring graphical processing units (GPUs) and for installing GPU drivers and software. Figures 22 and 23 show views of the HPE Insight CMU.

HPE Insight CMU can be configured to support High Availability with an active-passive cluster. For more information, see hpe.com/info/cmu

http://www.hpe.com/info/cmu


For the HPE Insight Cluster Management Utility BOM, see the Bill of Materials section of this paper.

Figure 22. HPE Insight CMU Interface – real-time view

Figure 23. HPE Insight CMU Interface – Time view


Summary HPE and MapR allow one to derive new business insights from Big Data by providing a platform to store, manage and process data at scale. However, designing and ordering MapR clusters can be both complex and time consuming. This white paper provides several Reference Architecture configurations for deploying clusters of varying sizes with the MapR Converged Data Platform (MapR) 5.2 on HPE infrastructure and management software. These configurations leverage HPE balanced building blocks of servers, storage and networking, along with integrated management software and bundled support. In addition, this white paper has been created to assist in the rapid design and deployment of MapR Converged Data Platform software on HPE infrastructure for clusters of various sizes.

Implementing a proof-of-concept As a matter of best practice for all deployments, HPE recommends implementing a proof-of-concept using a test environment that matches as closely as possible the planned production environment. In this way, appropriate performance and scalability characterizations can be obtained. For help with a proof-of-concept, contact an HPE Services representative (https://www.hpe.com/us/en/services/consulting/big-data.html) or your HPE partner.

Appendix A: Bill of Materials The BOMs outlined in this section are based on a tested configuration for a single-rack Reference Architecture with 1 management node, 2 head nodes, 8 worker nodes for BDO and 1 management node, 2 head nodes, 4 storage nodes and 3 worker nodes for WDO and 2 ToR switches. The following tables show the Bill of Materials for nodes and switches.

The following BOMs contain electronic license to use (E-LTU) parts. Electronic software license delivery is now available in most countries. HPE recommends purchasing electronic products over physical products (when available) for faster delivery and for the convenience of not tracking and managing confidential paper licenses. For more information, contact your reseller or an HPE representative.

Note Part numbers are at time of publication and subject to change. The bill of Materials does not include complete support options or other rack and power requirements. If you have questions regarding ordering, please consult with your HPE Reseller or HPE Sales Representative for more details. hpe.com/us/en/services/consulting.html

Management node and head node BOM for BDO and WDO Table A-1. BOM for the HPE ProLiant DL360 Gen9 server configuration

Qty Part Number Description

Management and head nodes

1 755258-B21 HPE DL360 Gen9 8-SFF CTO Chassis

1 818176-L21 HPE DL360 Gen9 E5-2640v4 FIO Kit

1 818176-B21 HPE DL360 Gen9 E5-2640v4 Kit

8 836220-B21 HPE 16GB 2Rx4 PC4-2400T-R Kit

24 785069-B21 HPE 900GB 12G SAS 10K 2.5in SC ENT HDD

1 817749-B21 HPE Ethernet 10/25Gb 2-port 640FLR-SFP28 Adapter

1 749974-B21 HPE Smart Array P440ar/2G FIO Controller

2 720478-B21 HPE 500W FS Plat Hot Plug Power Supply Kit

1 663201-B21 HPE 1U SFF Ball Bearing Rail Kit

1 C6N36ABE HPE Insight Control ML/DL/BL Bundle E-LTU

1 G3J28AAE RHEL Svr 2 Sckt/2 Gst 1yr 24x7 E-LTU

https://www.hpe.com/us/en/services/consulting.html


Edge node BOM Table A-2. BOM for the HPE ProLiant DL360 Gen9 server configuration


1 Edge nodes

1 755258-B21 HPE DL360 Gen9 8-SFF CTO Chassis


1 755836-B21 HPE DL380 Gen9 E5-2680v4 Kit


24 785069-B21 HPE 900GB 12G SAS 10K 2.5in SC ENT HDD


1 749974-B21 HPE Smart Array P440ar/2G FIO Controller


1 663201-B21 HPE 1U SFF Ball Bearing Rail Kit


C6N36A HPE Insight Control ML/DL/BL FIO Bndl Lic (optional if E-LTU is not available)


Control Block BOM for WDO with Apollo 2000

Table A-3. HPE Elastic Platform for Big Data Analytics – WDO control block – Apollo 2000


Control block

1 798153-B21 HPE Apollo r2600 24SFF CTO Chassis

2 800059-B21 HPE Apollo 2000 FAN-module Kit

4 798155-B21 HPE ProLiant XL170r Gen9 CTO Svr

4 850300-L21 HPE XL1x0r Gen9 E5-2640v4 FIO Kit

4 850300-B21 HPE XL1x0r Gen9 E5-2640v4 Kit



8 832414-B21 HPE 480GB 6G SATA MU-2 SFF SC SSD

4 844540-B21 HPE XL1x0r Gen9 NGFF 120Gx2 Riser Drive

4 798178-B21 HPE XL170r/190r LP PCIex16 L Riser Kit

4 798180-B21 HPE XL170r FLOM x8 R Riser Kit

4 798192-B21 HPE XL170r/190r Dedicated NIC IM Board Ki

4 800060-B21 HPE XL170r Mini-SAS B140 Cable Kit

2 720620-B21 HPE 1400W FS Plat Pl Hot Plug PS Kit

1 740713-B21 HPE t2500 Strap Shipping Bracket

1 822731-B21 HPE 2U Shelf-Mount Adjustable Rail Kit



1 G3J28AAE RHEL Server 2 Socket/2 Gst 1yr 24x7 E-LTU


Worker node BOM for BDO HPE ProLiant DL380 Table A-4. BOM for the HPE ProLiant DL380 Gen9 server configuration


1 Worker node

1 719061-B21 HPE DL380 Gen9 12-LFF CTO Server


1 762766-B21 HPE DL380 Gen9 E5-2680v3 Kit

8 726719-B21 HPE 16GB 2Rx4 PC4-2133P-R Kit

1 835565-B21 HPE Dual 340GB Read Intensive-2 Solid State M.2 Enablement Kit

15 628061-B21 HPE 3TB 6G SATA 7.2k 3.5in SC MDL HDD


1 761874-B21 HPE Smart Array P840/4G FIO Controller

1 768856-B21 HPE DL380 Gen9 3LFF Rear SAS/SATA Kit

1 727250-B21 HPE 12Gb DL380 Gen9 SAS Expander Card

1 720864-B21 HPE 2U LFF BB Gen9 Rail Kit





Storage node BOM for BDO Apollo 4200 Table A-5. BOM for the HPE Apollo 4200 server configuration


Worker node

4 808027-B21 HPE Apollo 4200 Gen9 24LFF CTO Server

4 830742-L21 HPE Apollo 4200 Gen9 E5-2680v4 FIO Kit

4 830742-B21 HPE Apollo 4200 Gen9 E5-2680v4 Kit


4 806563-B21 HPE Apollo 4200 Gen9 LFF Rear HDD Cage Kit

112 861688-B21 HPE 3TB 6G SATA 7.2K LFF MDL LP HDD

4 813546-B21 HPE 2nd Cage FIO Controller Mode for Rear Strg

4 777894-B21 HPE Dual 120GB RI Solid State M.2 Kit


4 764285-B21 HPE IB FDR/EN 40Gb 2P 544+ FLR-QSFP Adapter


4 806562-B21 HPE Apollo 4200 Gen9 Redundant Fan Kit

4 806565-B21 HPE Apollo 4200 Gen9 iLO Mgmt Prt Kit


4 822640-B21 HPE Apollo 4200 Gen9 FIO Strap Ship Brkt





Worker node BOM for WDO Apollo 2000

Table A-6. BOM for the HPE Apollo 2000 server configuration


Worker node

3 798153-B21 HPE Apollo r2600 24SFF CTO Chassis

6 800059-B21 HPE Apollo 2000 FAN-module Kit

12 798155-B21 HPE ProLiant XL170r Gen9 CTO Svr

12 850314-L21 HPE XL1x0r Gen9 E5-2680v4 FIO Kit




24 832414-B21 HPE 480GB 6G SATA MU-2 SFF SC SSD

12 844540-B21 HPE XL1x0r Gen9 NGFF 120Gx2 Riser Drive

12 798178-B21 HPE XL170r/190r LP PCIex16 L Riser Kit

12 798180-B21 HPE XL170r FLOM x8 R Riser Kit

12 798192-B21 HPE XL170r/190r Dedicated NIC IM Board Ki

12 800060-B21 HPE XL170r Mini-SAS B140 Cable Kit

6 720620-B21 HPE 1400W FS Plat Pl Hot Plug PS Kit

3 740713-B21 HPE t2500 Strap Shipping Bracket




1 G3J28AAE RHEL Server 2 Socket/2 Gst 1yr 24x7 E-LTU

Network BOMs Table A-7. BOM for HPE FlexFabric 5950 32Q28 switch


2 JH321A HPE FlexFabric 5950 32Q28 switch

4 JC680A HPE A58x0AF 650W AC power supply

12 JH389A HPE X712 Bck(pwr)-Frt(prt) HV2 Fan Tray

4 JL271A HPE X240 100G QSFP28 1m DAC Cable

Table A-8. BOM for HPE FlexFabric 5950 48SFP28 8QSFP28 switch


2 JH402A HPE FlexFabric 5950 48SFP28 8QSFP28 switch


10 JH389A HPE X712 Bck(pwr)-Frt(prt) HV2 Fan Tray

4 JL294A HPE X240 25G SFP28 to SFP28 1m DAC


Table A-9. Network – Top of Rack switch for separate iLO and PXE network


1 JG510A HPE 5900AF-48G-4XG-2QSFP+ switch


2 JC682A HPE A58x0AF Back (power side) to front (port side) airflow fan tray

1 806565-B21 HPE Apollo 4200 Gen9 iLO Mgmt Prt kit (one kit per HPE Apollo 4200 Gen9 server)

Other hardware and software BOM Table A-10. Hardware – Rack and PDU


Rack infrastructure

4 AF520A HPE Intelligent 4.9kVA/L6-30P/NA/J PDU

8 AF547A HPE 5xC13 Intelligent PDU Extension Bar G2 Kit

1 BW946A HPE 42U Location Discovery Kit

1 BW904A HPE 642 1075mm Shock Intelligent Series Rack

1 BW932A HPE 600mm Rack Stabilizer Kit

1 BW930A HPE Air Flow Optimization Kit

1 BW906A HPE 42U 1075mm Side Panel Kit

1 BW891A HPE Rack Grounding Kit

Note The quantity specified below is for a single node.

Table A-11. Software – HPE Insight Cluster Management Utility (CMU) options


CMU-options

1 QL803B HPE Insight CMU 1yr 24x7 Flex Lic

1 QL803BAE HPE Insight CMU 1yr 24x7 Flex E-LTU

1 BD476A HPE Insight CMU 3yr 24x7 Flex Lic

1 BD476AAE HPE Insight CMU 3yr 24x7 Flex E-LTU

1 BD477A HPE Insight CMU Media

Table A-12. Software – Red Hat Enterprise Linux




Appendix B: Alternate parts for storage node Table B-1. Alternate processors – HPE ProLiant DL380


1

1

762764-L21

762764-B21

HPE DL380 Gen9 E5 2660v3 FIO Kit (10 cores/2.6GHz)

HPE DL380 Gen9 E5 2660v3 Kit

1

1

719054-L21

719054-B21

HPE DL380 Gen9 E5-2697v3 FIO Kit (14 cores/2.3GHz)


1

1

817945-L21

817945-B21



1

1

817959-L21

817959-B21



Table B-2. Alternate memory – HPE ProLiant DL380


8 726719-B21 HPE 16GB 2Rx4 PC4-2133P-R Kit for 128GB of Memory






Table B-3. Alternate disk drives – HPE ProLiant DL380


15 652757-B21 HPE 2TB 6G SAS 7.2K 3.5in SC MDL HDD




15 793703-B21 HPE 8TB 12G SAS 7.2K 3.5in 512e SC HDD





15 793695-B21 HPE 8TB 6G SATA 7.2K 3.5in 512e SC HDD

Table B-4. Alternate processors – HPE Apollo 4200


1 830746-L21 HPE Apollo 4200 Gen9 E5-2690v4 FIO Processor Kit (14C,2.6GHz)


1 803308-L21 HPE Apollo 4200 Gen9 E5-2660v3 FIO Processor Kit(12C,2.4GHz)



Table B-5. Alternate memory – HPE Apollo 4200 (Quantity per node specific to customer deployment)

Qty per node Part Number Description

Deployment-specific 805351-B21 HPE 32GB 2Rx4 PC4-2400T-R Kit



Appendix C: Alternate parts for compute node Table C-1. Alternate processors – HPE Apollo 2000 – HPE ProLiant XL170r


1 850322-L21 HPE XL1x0r Gen9 E5-2697v4 FIO Kit (18C, 2.3 GHz)


1 850318-L21 HPE XL1x0r Gen9 E5-2690v4 FIO Kit (14C, 2.6 GHz)

1 850318-B21 HPE XL1x0r Gen9 E5-2690v4Kit

1 793028-B21 HPE XL1x0r Gen9 E5-2680v3 FIO Kit (12C, 2.6 GHz)

1 793028-L21 HPE XL1x0r Gen9 E5-2680v4Kit

1 793024 -L21 HPE XL1x0r Gen9 E5-2660v3 FIO Kit (10C, 2.6 GHz)

1 793024 -B21 HPE XL1x0r Gen9 E5-2660v3 Kit

Table C-2. Alternate memory – HPE Apollo 2000 – HPE ProLiant XL170r

Qty per node Part Number Description




Table C-3. Alternate disk drives – HPE Apollo 2000 – HPE ProLiant XL170r

Qty per Node Part Number Description

2/4/6 804625-B21 HPE 800GB 6G SATA MU-2 SFF SC SSD

2/4/6 817011-B21 HPE 1.92TB 6G SATA MU-3 SFF SC SSD

2/4/6 804631-B21 HPE 1.6TB 6G SATA MU-2 SFF SC SSD

1 798190-B21 HPE XL1x0r Gen9 NGFF Riser w/ 2x64G Drive (Optional for Operating System)


Appendix D: MapR cluster tuning/optimization Server tuning Below are some general guidelines for tuning the server OS and the storage controller for a typical MapR proof-of-concept (POC). Please note that these parameters are recommended for MapReduce workloads which are most prevalent in MapR environments. Please note that there is no silver bullet performance tuning. Modifications will be needed for other types of workloads. For additional MapR performance tuning, visit http://maprdocs.mapr.com/home/AdministratorGuide/Resource-Allocation-Jobs-Apps.html?hl=resource,allocation,jobs,applications

• OS tuning

As a general recommendation, update to the latest patch level available to improve stability and optimize performance. The recommended Linux file system is ext4, 64 bit OS:

– Enable defaults, nodiratime,noatime (/etc/fstab)

– Do not use logical volume management (LVM)

– Tune OS block readahead to 8K (/etc/rc/local):

blockdev --setra 8192 <storage device>

– Decrease disk swappiness to minimum 1:

Set sysctl vm.swappiness=1 in /etc/sysctl.conf

– Tune ulimits for number of open files to a high number:

Example: in /etc/security/limits.conf:

soft nofile 65536

hard nofile 65536

Set nproc = 65536

Add it to end of (/etc/security/limits.conf)

– Set IO scheduler policy to deadline on all the data drives:

echo deadline > /sys/block/<device>/queue/scheduler

– For persistency across boot, append the following to kernel boot line in /etc/grub.conf:

elevator=deadline

– Configure network bonding on two 10GbE server ports, for 20GbE throughput.

– Ensure forward and reverse DNS is working properly.

– Install and configure ntp to ensure clocks on each node are in sync to the management node.

– For good performance improvements, disable transparent huge page compaction:

echo never > /sys/kernel/mm/transparent_hugepage/enabled

• Storage controller tuning

– Tune array controller stripe size to 1024MB:

hpssacli ctrl slot=<slot number> ld <ld number> modify ss=1024

– Disable array accelerator(caching) (aa=disable):

hpssacli ctrl slot=<slot number> ld <ld number> modify aa=disable

http://maprdocs.mapr.com/home/AdministratorGuide/Resource-Allocation-Jobs-Apps.html?hl=resource,allocation,jobs,applications


• Power settings

Please note for a performance driven POC, we recommend using settings that help boost performance but could have negative impact on power consumption measurement:

– HPE Power Profile Maximum Performance

– HPE Power Regulator Static High Performance mode

– Intel_QPI_Link_Mgt Disabled

– Min_Proc_Idle_power_Core_State No C-states

– Mem_Power_Saving Max Perf

– Thermal Configuration increased cooling

– Min_Proc_Idle Power Package state No Package state

– Energy/Performance Bios Disabled

– Collaborative Power Control Disabled

– Dynamic Power Capping Functionality Disabled

DIMM Voltage Preference Optimized for Performance

• CPU tuning

The default BIOS settings for CPU should be adequate for most MapR workloads. Make sure that Hyper-Threading is turned on as it will help with additional performance gain.

• HPE ProLiant BIOS

– SPP version >= 2015.04.0 (B)

– Update System BIOS version to be >= P89

– Update HPE Integrated Lights-Out (iLO) version to be >= 2.03

– Intel Virtualization Technology Disabled

Intel VT-d Disabled

• Embedded HPE Dynamic Smart Array B140i controller

– The embedded HPE Dynamic Smart Array B140i controller used for m.2 OS drives will operate in UEFI only mode. The HPE Dynamic Smart Array B140i defaults to AHCI off the chipset. As shown below, the HPE Smart Array needs to be enabled in BIOS on the SATA-only models, if required.

Embedded SATA Configuration SATA_RAID_ENABLED

– By default, the HPE Dynamic Smart Array B140i will not operate in Legacy mode. For legacy support, additional controllers will be needed; and, for CTO orders, select the Legacy mode settings part, 758959-B22.

– Enable caching on the OS logical drive

ctrl slot=<slot number> create type=ld drives=<drives> caching=enable


• HPE Smart Array P440ar / P840

– Update controller firmware to be >= v1.34

– Configure each MapR data drive as a separate RAID0 array with stripe size of 1024KB

– Turn Off “Array Acceleration” / ”Caching” for all data drives

Example:

ctrl slot=<slot number> ld all modify caching=disable disable caching on all logical drives on 1st ctrlr

ctrl slot=<slot number> ld 1 modify caching=enable enable caching on the OS logical drive on 1st ctrlr

– Tune array controller stripe size to 1024MB:

hpssacli ctrl slot=<slot number> ld <ld number> modify ss=1024

• Network cards

– Ethernet driver ixgbe, version >= 3.23.0.79 and firmware version >= 0x800005ac, 1.949.0

• Oracle Java

java.net.preferIPv4Stack set to true

• Patch common security vulnerabilities

– Check Red Hat Enterprise Linux and SUSE security bulletins for more information.

Appendix E: HPE Pointnext value-added services and support In order to help customers jump-start their Big Data solution development, HPE Pointnext offers flexible, value-added services, including Factory Express and Big Data Consulting services which can accommodate and end-to-end customer experience.

HPE Pointnext Factory Express Services Factory-integration services are available for customers seeking a streamlined deployment experience. With the purchase of Factory Express services, your cluster will arrive racked and cabled, with software installed and configured per an agreed upon custom statement of work, for the easiest deployment possible. HPE Factory Express Level 4 Service (HA454A1) is the recommended Factory Integration service for Big Data covering hardware and software integration, as well as end-to-end delivery project management. Please engage HPE Pointnext Factory Express for details and quoting assistance. For more information and assistance on Factory Integration services, you can go to: https://www.hpe.com/us/en/services/factory-express.html

Or contact:

• AMS: [email protected]

• APJ: [email protected]

• EMEA: [email protected]

HPE Pointnext Big Data Consulting – Reference Architecture Implementation Service for Hadoop With the HPE Reference Architecture Implementation Service for Hadoop, experienced HPE Big Data consultants install, configure, deploy, and test your MapR environment based on the HPE Reference Architecture for MapR. HPE will implement a MapR design: naming, hardware, networking, software, administration, backup and operating procedures and work with you to configure the environment according to your goals and needs. HPE will also conduct an acceptance test to validate and prove that the system is operating as defined in the Reference Architecture.

https://www.hpe.com/us/en/services/factory-express.html





HPE Pointnext Advisory, Transform and Manage - Big Data Consulting Services HPE Pointnext Big Data Consulting Services cover the spectrum of services to advise, transform, and manage your MapR environment, helping you to reshape your IT infrastructure to corral increasing volumes of bytes – from e-mails, social media, and website downloads – and convert them into beneficial information. Our Big Data solutions encompass strategy, design, implementation, protection and compliance. We deliver these solutions in three steps.

1. Big Data Architecture Strategy and Roadmap: We’ll define the functionalities and capabilities needed to align your IT with your Big Data initiatives. Through transformation workshops and roadmap services, you’ll learn to capture, consolidate, manage and protect business-aligned information, including structured, semi-structured and unstructured data.

2. Big Data System Infrastructure: HPE experts will design and implement a high-performance, integrated platform to support a strategic architecture for Big Data. Choose from design and implementation services, Reference Architecture implementations and integration services. Your flexible, scalable infrastructure will support Big Data variety, consolidation, analysis, share and search on HPE platforms.

3. Big Data Protection: Ensure availability, security and compliance of Big Data systems. Our consultants can help you safeguard your data, achieve regulatory compliance and lifecycle protection across your Big Data landscape, as well as improve your backup and continuity measures.

For additional information, visit: hpe.com/us/en/services/consulting/big-data.html

Hewlett Packard Enterprise Support options HPE offers a variety of support levels to meet your needs:

• HPE Datacenter Care - HPE Datacenter Care provides a more personalized, customized approach for large, complex environments, with one solution for reactive, proactive, and multi-vendor support needs.

• HPE Support Plus 24 - For a higher return on your server and storage technology, our combined reactive support service delivers integrated onsite hardware/software support services available 24x7x365, including access to HPE technical resources, 4-hour response onsite hardware support and software updates.

• HPE Proactive Care - HPE Proactive Care begins with providing all of the benefits of proactive monitoring and reporting along with rapid reactive care. You also receive enhanced reactive support, through access to HPE’s expert reactive support specialists. You can customize your reactive support level by selecting either 6 hour call-to-repair or 24x7 with 4 hour onsite response. You may also choose DMR (Defective Media Retention) option.

• HPE Proactive Care with the HPE Personalized Support Option - Adding the Personalized Support Option for HPE Proactive Care is highly recommended. The Personalized Support option builds on the benefits of HPE Proactive Care Service, providing you an assigned Account Support Manager who knows your environment and delivers support planning, regular reviews, and technical and operational advice specific to your environment. These proactive services will be coordinated with Microsoft's proactive services that come with Microsoft® Premier Mission Critical, if applicable.

• HPE Proactive Select - And to address your ongoing/changing needs, HPE recommends adding Proactive Select credits to provide tailored support options from a wide menu of services, designed to help you optimize capacity, performance, and management of your environment. These credits may also be used for assistance in implementing updates for the solution. As your needs change over time you flexibly choose the specific services best suited to address your current IT challenges.

• Other offerings - In addition, Hewlett Packard Enterprise highly recommends HPE Education Services (for customer training and education) and additional Pointnext, as well as in-depth installation or implementation services as may be needed.

https://www.hpe.com/us/en/services/consulting/big-data.html


Sign up for updates

© Copyright 2017 Hewlett Packard Enterprise Development LP. The information contained herein is subject to change without notice. The only warranties for Hewlett Packard Enterprise products and services are set forth in the express warranty statements accompanying such products and services. Nothing herein should be construed as constituting an additional warranty. HPE shall not be liable for technical or editorial errors or omissions contained herein.

Intel and Xeon are trademarks of the Intel Corporation in the U.S. and other countries. Linux is a registered trademark of Linus Torvalds in the U.S. and other countries. Red Hat is a registered trademark of Red Hat, Inc. in the United States and other countries. Microsoft and Windows are either registered trademarks or trademarks of Microsoft Corporation in the United States and/or other countries. Oracle and Java are registered trademarks of Oracle and/or its affiliates.

a00020201enw, July 2017

Resources and additional links MapR, https://www.mapr.com/

HPE Solutions for Apache Hadoop, hpe.com/info/hadoop

Hadoop and Vertica, hpe.com/info/vertica

HPE Insight Cluster Management Utility (CMU), hpe.com/info/cmu

HPE FlexFabric 5900 switch series, hpe.com/networking/5900

HPE FlexFabric 5940 switch series, hpe.com/us/en/product-catalog/networking/networking-switches/pip.hpe-flexfabric-5940-switch-series.1009148840.html

HPE FlexFabric 5950 switch series, hpe.com/us/en/product-catalog/networking/networking-switches/pip.hpe-flexfabric-5950-switch-series.1008901775.html

HPE ProLiant servers, hpe.com/info/proliant

HPE Networking, hpe.com/networking

HPE Services, hpe.com/services

Red Hat, redhat.com

HPE EPA Sizing tool: HPE EPA Sizing Tool

HPE Education Services: http://h10076.www1.hpe.com/ww/en/training/portfolio/bigdata.html

To help us improve our documents, please provide feedback at hpe.com/contact/feedback.

About MapR Headquartered in San Jose, Calif., MapR provides the industry’s only Converged Data Platform that enables customers to harness the power of Big Data by combining analytics in real-time to operational applications to improve business outcomes. With MapR, enterprises have an unparalleled data management platform for undertaking digital transformation initiatives to achieve competitive edge. World-class companies have realized more than five times their return on investment using MapR. Amazon, Cisco, Google, Microsoft, SAP and other leading businesses are part of the global MapR partner ecosystem. For more information, visit MapR.com.

http://www.hpe.com/info/getupdated

http://www.facebook.com/sharer.php?u=https://www.hpe.com/h20195/V2/GetDocument.aspx?docname=a00020201enw

http://twitter.com/home/?status=HPE%20Reference%20Architecture%20for%20MapR%20Converged%20Data%20Platform%20on%20HPE%20Elastic%20Platform%20for%20Big%20Data%20Analytics%20(EPA)+@+https://www.hpe.com/h20195/V2/GetDocument.aspx?docname=a00020201enw

http://www.linkedin.com/shareArticle?mini=true&ro=true&url=https://www.hpe.com/h20195/V2/GetDocument.aspx?docname=a00020201enw&title=HPE%20Reference%20Architecture%20for%20MapR%20Converged%20Data%20Platform%20on%20HPE%20Elastic%20Platform%20for%20Big%20Data%20Analytics%20(EPA)+&armin=armin

https://www.mapr.com/

http://www.hpe.com/info/hadoop

http://www.hpe.com/info/vertica

http://www.hpe.com/info/cmu

http://www.hpe.com/networking/5900

https://www.hpe.com/us/en/product-catalog/networking/networking-switches/pip.hpe-flexfabric-5940-switch-series.1009148840.html




http://www.hpe.com/info/proliant

http://www.hpe.com/networking

http://www.hpe.com/services

http://www.redhat.com/

https://www.hpe.com/h20195/v2/GetDocument.aspx?docname=a00005868enw&doctype=Technical%20Brief&doclang=EN_US&searchquery=&cc=us&lc=en

http://h10076.www1.hpe.com/ww/en/training/portfolio/bigdata.html

http://www.hpe.com/contact/feedback

http://www.mapr.com/

Documents

HPE Reference Architecture for MapR Converged …...The configurations described in this Reference Architecture are based on the 5.2 release of the MapR Converged Data Platform and