Hyperscale Infrastructure for All

FUNGIBLE DATA CENTERS Product Brief

1

THE INDUSTRIALIZATION OF CLOUD INFRASTRUCTURE IS HAPPENING NOW. BE PART OF THE MOVEMENT. The adoption of cloud computing is no longer a question of “if” but a question of “when” and “where.” Organizations worldwide are accelerating the shift toward high-performance, scalable cloud computing architectures that are software-defined and programmable. The need for agility, performance, and simplicity are driving this new wave of digital transformation.

Hyperscale companies have been at the forefront of providing public cloud infrastructure and services to organizations, leveraging economies of scale and optimizing every aspect of the technology stack, from silicon to software. Hyperscale technologies, while desirable, are currently unavailable to anyone beyond a handful of providers.

While many organizations have fully embraced a public cloud strategy, many others have maintained a hybrid approach, choosing to retain the most critical information and intellectual property on-premises. Needless to say, there is no one-size-fits-all solution as organizations’ needs continue to evolve to compete in a shifting business landscape.

However, whether cloud infrastructure is public or private, organizations expect requirements to be duly met: high performance, cost-effectiveness, power, and footprint efficiency without compromising reliability, security, and scalability. Organizations are seeking alternatives to rigid infrastructures that are immutable and slow to respond to change, managing costs more prudently by minimizing the over-provisioning infrastructure to meet peak utilization that occurs infrequently and ensuring expensive resources are not underutilized. And while organizations have in the past, traded performance for simplicity, this can no longer be tolerated with modern applications such as Big Data, AI/ML workloads, and large container farm deployments demanding bare-metal performance. Lastly, organizations are also keenly wary of security threats that are only increasing in sophistication and frequency.

FUNGIBLE DATA CENTERS (FDC) - HYPER DISAGGREGATED, COMPOSABLE INFRASTRUCTUREWhat if you could transform data centers from rigid, complex, and expensive silos to simple, highly efficient, and on-demand powerhouses?

The FDC accomplishes this and more through a hyperdisaggregated, composable infrastructure paradigm, where the data center is comprised of disaggregated resources - servers, storage, networks, GPUs, FPGA accelerators – that can be composed to the exact infrastructure you need in just a few minutes, with a few clicks in software. The FDC also enables bare-metal performance while fully utilizing all the available resources and preserving security postures across tenants.

FUNGIBLE DATA CENTERSHyperscale Infrastructure for All

TARGET CUSTOMERSNext wave service providers (Cloud, Edge, Managed, Telco)

Enterprises and SaaS companies

Research labs and universities

Simplified IT: Management of multi-tenant, secure data centers via a single pane of glass management

Agile IT: Deployment and redeployment of resources to handle workload changes and hot spots in minutes

Bare-Metal Performance: Market-leading performance for modern applications enabled by fully offloading data-centric I/O processing to the Fungible DPUTM

Massive Scalability: Scales from half a rack to hundreds of racks to handle the most complex data center deployments

Unrivaled Economics: Just-in-time composition of independent compute, storage, network, and GPU resources optimizes consumption to exactly meet workload requirements with no resource wastage. Composable, hyperdisaggregated architecture reduces server SKUs to a minimal set, gaining economies of scale

Iron-Clad Security: Independent hardware-accelerated security domains, fine-grained segmentation, robust QoS, and line-rate encryption ensures robust security

BENEFITS


2

The DCC supports role-based access control, presenting two levels of administrative control to FDC users.

THE SOLUTIONThe Fungible Data Center is based on a fundamental rethink of the modern data center and is designed to address the most demanding challenges of hyperscale environments. It is an on-premises, comprehensive, turn-key solution offering multi-tenant and secure data centers, delivering bare-metal performance at the cost efficiencies and simplicity of hyperscale data centers.

It comprises pre-configured racks of disaggregated compute and storage servers powered by the Fungible DPU™, managed by the Fungible Data Center Composer (DCC) software. The FDC ships with the following components:

• Two or more Fungible FS1600 nodes. The FS1600 is a high-performance, secure, scale-out NVMe-oF storage platform. Two Fungible F1 DPUs power a single FS1600. The Fungible DPU enables the FS1600 to deliver unrivaled performance in the most demanding scenarios, featuring a random read rate of 15M IOPS (4KB), a throughput of 60 GB/s, and read latencies of +10μs vs. direct-attached storage (DAS) systems. The Fungible DPU hardware accelerators in FS1600 (including compression, erasure coding, encryption) operate at a line rate of 800 Gb/s. FS1600 addresses the constraints that have traditionally limited true storage disaggregation by delivering DAS-like performance, linearly scaled in large clusters.

• A set of compute servers with industry-standard processors, with no local disks, and one Fungible Data Services Platform (DSP) FC50/FC100/FC200 card. A Fungible S1 DPU powers each DSP card. The DSP card is a purpose built to fully offload the data center network, storage, and security processing from the CPUs, enabling the CPUs to focus on application processing. The DSP card implements a separation of control and data plane, with a fully programmable data path optimized for highly efficient data-centric computations.

• A set of three dedicated management nodes that hosts the Fungible Data Center Composer, a comprehensive software solution that supports bare-metal data center composition, provisioning, management, and orchestration.

• A set of pre-qualified, industry-standard Top-of-Rack (ToR) switches for data and management networks and routers. In each rack, there is one ToR switch for the data network and one for BMC management network. Each of the ToRs comes with DHCP relay capability, connecting to the data center spine network.

At the core of the FDC solution components lies the Fungible DPU - unifying the trifecta of compute, storage, and networking and making data centers that are truly fungible a reality! A unique Fungible innovation, the Fungible DPU represents a new class of microprocessors designed from the ground up to solve fundamental bottlenecks in the traditional CPU-centric data center architecture. Specifically, the Fungible DPU executes data-centric computations an order of magnitude more efficiently than general-purpose CPUs and facilitates highly efficient data interchange among server and storage nodes through its TrueFabricTM technology. The Fungible DPU is the foundational technology that enables the hyperdisaggregation and pooling of all data center resources to be realized at a massive scale. Refer to the Fungible DPU product brief for more information.

THE FUNGIBLE DATA CENTER COMPOSER SOFTWAREThe Fungible Data Center Composer (DCC) is the “mission control” of the Fungible Data Center. The DCC is a comprehensive software solution that supports bare-metal data center composition, provisioning, management, and orchestration of infrastructure at all scales.

Data Center Administrator (DCA) - a role that manages the overall FDC. The DCA

partitions the compute, storage, and network resources and allocates them to different Logical Partitions (LPs). An LP is an abstraction representing one tenant. The DCA is thus responsible for supporting multiple tenants and ensures there is complete isolation amongst the tenants.

Logical Partition Administrator (LPA) - a role that supports the tenant to which the

partition is allocated to. The LPA uses the resources allocated by the DCA to compose bare-metal servers based on the requirements of application workloads. The LPA is also responsible for deploying these application workloads within the LP.

BARE-METAL COMPOSITION FOR RIGHT-SIZED, ON-DEMAND DATA CENTERSBy harnessing the full capabilities of the Fungible DPU, the Fungible Data Center Composer (DCC) allows the data center infrastructure to dynamically adapt to changing workload conditions by composing compute, storage and network resources from a disaggregated pool of resources. The ability to hyperdisaggregate resources, then compose them efficiently enables a Fungible Data Center to maintain a minimal set of server SKUs, allowing for economies of scale and management simplicity.

The DCC’s composition is bare-metal, removing the overhead of software virtualization and hypervisors, thus delivering performance equivalent to that of custom-designed bare-metal data centers. The FDC uses standard drivers for Linux distributions, simplifying the setup experience, and applications can run directly on the bare-metal composed servers with no changes to applications.

The DCC also delivers real-time load balancing* of resource allocations to address hot spots in data centers. For instance, some environments need additional server resources, while others have an excess of said resources. The logical partition administrator (LPA) can temporarily remove the physical server attached to an environment while preserving its entire state (disks, NIC IP address, MAC address, etc.) and move that server to another environment that needs additional compute capacity. The LPAs have the flexibility to adjust allocated resources in each environment dynamically, enabling optimum resource utilization across all data center environments. This novel and unique capability is also known as “thin provisioning” for servers.

* represents road map item.

https://www.fungible.com/wp-content/uploads/2020/08/PB0028.01.02020820-Fungible-F1-Data-Processing-Unit.pdf

https://www.fungible.com/wp-content/uploads/2020/08/PB0028.01.02020820-Fungible-F1-Data-Processing-Unit.pdf


3

EFFORTLESS WORKLOAD PROVISIONING FOR SEAMLESS AND RAPID DEPLOYMENTThe DCC supports one-click deployment of complex, scale-out workloads via pre-configured templates from an application marketplace, enabling workload deployment in mere minutes! TheFDC offers two types of template stores: one for the data center andthe other for tenant-specific logical partitions. Each logical partitioncan have templates specific to their application workloads. The DCAcan grant permissions to templates for use in one or more logicalpartitions. Logical partition administrators (LPAs) can use the datacenter templates they have been granted access or create and usetheir own templates. This feature enables a repeatable and seamlessdeployment of applications (without a single line of code change tothe application itself!). As with any good modern architecture, thesystem is built on a foundation of infrastructure-as-code forrepeatability of deployments.

SIMPLIFIED & SECURE IT: MANAGE BARE-METAL INFRASTRUCTURE LIKE VIRTUAL MACHINESThe Fungible DCC is a fully web-based, single-pane-of-glassmanagement software that manages the entire Fungible DataCenter environment. It is designed with simplicity in mind to handlediscovery, initialization, registration, and decommissioning ofhardware resources, identity management, monitoring oftelemetry data, management of failures, firmware updates, andmany other functions.

Multi-Tenancy, Customer Isolation and Identity Management

The DCC is capable of creating completely isolated partitions to support a multi-tenant environment. Security and access control are achieved through role-based access. The role-based access system supports two roles, one for data center administrators and one for each tenant (i.e., logical partition administrators). Each role has its own independent management and authentication planes.

The Fungible Data Center leverages KeyCloak to provide robust identity management to support the role-based access control and multi-tenancy. The FDC plugs into existing customers’ identity management systems, such as LDAP or Active Directory, to authenticate both the DCA and LPA roles. Each logical partition can also be configured to work with that tenant organization’s own identity provider. This makes it possible for CSP or MSP customers (i.e., the DCA role) to seamlessly onboard their tenants (LPAs) without requiring tenants to use yet another authentication system. LPAs are self-sufficient via the tenant’s independent management portal and APIs.

Hardware Lifecycle Management and Monitoring

The FDC offers complete hardware lifecycle management* and monitoring capabilities. A DCA can update software versions

for the DCC and firmware versions for components, such as the Fungible FS1600s storage nodes, the Fungible Data Services Platform FC50/100/200 PCIe cards, and compute servers. Health status, telemetry data, failure data, and other exception issues in the hardware assets can also be aggregated and monitored by the FDC. Critical events are pushed to users in the form of notifications, and administrators can customize the volume and the type of notifications they would like to receive. Notifications are published to standard industry formats, pushed through the management portal, and sent via email. The scope of monitoring and notification coverage is limited to the permission level of each role. The DCC also includes a dashboard* that displays aggregate hardware status and critical items requiring immediate attention. A detailed log of all system events is maintained and published using industry-standard log management technologies.

Environment Deployment and Integration with Industry-Standard Orchestration Tools

The DCC is built on a cloud-native design, leveraging an API-first approach and supports APIs for the entire infrastructure, allowing deep application integration and automation opportunities for infrastructure management.

LPAs can create environments from scratch or instantiate environments from existing templates. In either case, LPAs can deploy environments using industry-leading orchestration automation technologies, such as Ansible and Terraform. The First-boot configuration is managed via Cloud-init.

Flexible Networking Control

The DCC also enables easy integration with existing networksvia a built-in network controller or a customer’s existing networkcontroller. The FDC provides a complete set of networkingcapabilities to manage a data center via a built-in network controllerthat operates in conjunction with data center Virtual Routing &Forwarding (VRFs). DCAs have the flexibility to create virtualnetworks and permit them to specific logical partitions or todelegate the capability to create virtual networks from assignedVRF/CIDR blocks. Customers can bypass* the FDC’s built-innetwork controller and use their existing network controller withoutlosing other FDC functionality. In this case, the underlay networksof deployment are managed by the FDC, and customers’ existingnetwork controller manages everything else. The passthrough*network is physical and not managed by the FDC. It offers hooks tohigher-layer network management systems such as the Open VirtualNetwork (OVN) hierarchy where the configuration and managementare handled. The FDC maintains NICs’ names and membership inpassthrough mode, which is then used by Open vSwitch (OVS) or asimilar system to set up one or more virtual overlay networks.

+ =

BARE-METAL COMPOSITION AND RE-COMPOSITION OF SERVERSwith the Desired Disk, GPU, and NIC Geometry

General purposediskless server

Disk, NIC, GPUspecification

Composed server Detach server while preserving complete state

Re-hydrate server state of a different workload to eliminate hotspots


4

THE FUNGIBLE DATA CENTERHARDWARE SPECIFICATIONS

Juniper QFX5200-32C

Juniper EX2300-48P

Fungible Data Center Composer Nodes

Compute Nodes powered by Fungible Data Services Platform PCIe Card

Fungible FS1600 Storage Nodes

FUNGIBLE FS1600 STORAGE NODES

• High-performance, secure, scale-out NVMe-oF storage platform

• Support for up to 24x U.2 NVMe SSD bays, with raw capacity options of 46, 92, and 184 TB per FS1600 node

• Rack unit: 2 RU

• Dimensions: H x W x D: 3.5” (88 mm) x 17.4” (440 mm) x 30.6” (776 mm)

• Weight: 43.4 lb (20Kg) without SSDs

• Power input/supply: (Redundant) 2 x 1500 W AC to DC power-factor-corrected (PFC), Universal Input Voltage Range: 90–246 VAC

• Power usage: 750 W/node (typical), 850 W/node (peak)

• Data Network: 12 x 100 Gb/s Ethernet ports

• Management Network: 1 x 1/10 Gb/s Ethernet port, 1 x 1Gb/s Ethernet IPMI/BMC port

For complete details, please refer to the Fungible Storage Cluster and Fungible FS1600 product briefs.

COMPUTE NODES POWERED BY FUNGIBLE DATA SERVICES PLATFORM PCIE CARDSupermicro HigH-end enterpriSe Server

• Family of high-end enterprise servers with industry-standard processors, with no local disks used for composing bare-metal servers for hosting customer applications

• Rack unit: 1 RU


• Weight: Net: 26 lb (11.8 kg) / gross: 41 lb (18.6 kg)

NOTE: Illustrative purposes only. The placement of actual servers and storage may be distributed across multiple rack.

• Power input/supply: (redundant) 2 x 750W AC to 100–127 VAC, 50–60 Hz; 200–240 VAC, 50–60 Hz

• CPU: Dual Socket P (LGA 3647) support 2nd Gen Intel® Xeon® Scalable processors (Cascade Lake/Skylake)

• Memory Capacity: 24 DIMMs; up to 6TB 3DS ECC DDR4-2933MHz RDIMM/LRDIMM

• Power Usage: 750W Redundant Power Supplies Platinum Level

• PCI-Express: 2 PCI-E 3.0 x16 slots (FH, 10.5” L), 2 PCI-E 3.0 x8 slots (1 LP, 1 internal LP)

For complete details, please refer to the SuperServer 1029U-TR4 brief.

Fungible data ServiceS platForm

• Family of high-performance, dual-port PCIe cards powered by the Fungible S1 DPU™, accelerating data center network, storage, and security while optimizing application efficiency of compute nodes

• Dimensions: Full height, half length (FHHL)

• Power: < 75W

• Features: ■ Storage Initiator

■ Supports remote boot of host and guest (VM) OSes (NVMeoTCP)

■ High-performance bare-metal virtualized NIC with multiple accelerators:

■ NVMe/TCP Storage Initiator Offload

■ IP/TCP/UDP inner and outer Checksum offload

■ LSO/TSO Offload

■ Receive Side Steering (RSS)

■ VXLAN Encapsulation/Decapsulation

■ VLAN Insertion/Stripping

■ Security Accelerators

■ Lookaside Crypto (PF)

■ Jumbo frame support (up to 16KB)

FUNGIBLE DATA CENTER COMPOSER NODES

• A set of three dedicated management nodes to host the FDC Composer software

• Rack unit: 1 RU


• Weight: Net: 26 lb (11.8 kg) / gross: 41 lb (18.6 kg)

• Power input/supply: (redundant) 2 x 750W AC to 100–127 VAC, 50–60 Hz; 200–240 VAC, 50–60 Hz

• Power usage: 700 W (peak)

• Network: 4 x 1 G (4 x RJ-45 GbE ports), 2 x 10/25G NIC

TOP OF RACK SWITCHESdata plane SwitcH: Juniper QFX5200-32c• Line-rate, low-latency platform for building large spine and leaf,

IP-fabric data center networks

• Rack units: 1 RU

• Dimensions: H x W x D: 1.72” (43.7 mm) x 17.36” (440.9 mm) x 20.48” (520.2 mm)

• Weight: 23.8 lb (10.8 Kg)

• Power: Redundant (1+1) hot-pluggable 850 W AC/DC power supplies, 110-240 V single-phase AC power, -36 to -72 V DC power

https://www.fungible.com/wp-content/uploads/2020/12/PB0020.04.02021214-Fungible-Storage-Cluster-FSC-Disaggregated-Storage-Platform.pdf

https://www.fungible.com/wp-content/uploads/2020/11/PB0018.04.02021118-Fungible-FS1600-Disaggregated-Storage-Platform.pdf

https://www.supermicro.com/en/products/system/1U/1029/SYS-1029U-TR4.cfm


5

ABOUT FUNGIBLESilicon Valley-based Fungible is reimagining the agility, performance, economics, reliability and security of today’s data centers.

CONTACT US [email protected]

Fungible, Inc.3201 Scott Blvd., Santa Clara, CA 95054, USA669-292-5522

www.fungible.com

PB0051.00.12020330

© 2020 Fungible, Inc. All rights reserved. Fungible, the Fungible logo and all product and service names mentioned herein are registered trademarks or trademarks of Fungible, Inc. in the United States and other countries. All other brand names mentioned herein are for identification purposes only and may be the trademarks of their respective holder(s).

• CPU: Intel Quad Core Ivy Bridge 1.8 GHz CPU, 16 GB SDRAM, 64 GB SSD

• Operating system: Junos OS

• System throughput: Up to 6.4 Tbps (bi-directional)

• Forwarding capacity: Up to 2.4 Bpps

• Maximum 100GbE port density: 32

For complete details, please refer to the Juniper QFX5200 Switch datasheet.

management and bmc SwitcH: Juniper eX2300-48p• Compact, high-performance switches for deploying

management network in the FDC

• Rack units: 1 RU

• Dimensions: H x W x D: 1.75” (44.5 mm) x 17.5” (441.0 mm) x 12.2” (309. 8mm)

• Weight: 11.07 lb (5.02 Kg)

• Power: Redundant (1+1) hot-pluggable 850 W AC/DC power supplies, 110-240 V single-phase AC power, -36 to -72 V DC power

• CPU: 1.25GHz ARM CPU

• Operating system: Junos OS

• System throughput: Up to 176 Gbps

• Backplane speed: 80Gbps

• Maximum 10GbE port density: 48

For complete details, please refer to the Juniper EX2300 Switch datasheet.

Compute.Tiny Compute.Medium Compute.Large Compute.XLarge

Storage.Tiny

• (2) FS1600, 92TB

• (4) Compute Nodes, 2 *8x3.1Ghz, 256GB RAM

• Fungible DCC nodes

• (2) FS1600, 92TB

• (8) Compute Nodes, 2*8x3.1Ghz, 512GB RAM


• (2) FS1600, 92TB



• (2) FS1600, 92TB

• (32) Compute Nodes, 2*20x2.1Ghz, 1.5TB RAM


Storage.Small

• (2) FS1600, 92TB



• (2) FS1600, 92TB



• (2) FS1600, 92TB



• (2) FS1600, 92TB



Storage.Medium

• (3) FS1600, 138TB



• (3) FS1600, 138TB



• (3) FS1600, 138TB



• (3) FS1600, 138TB



Storage.Large.Perf

• (7) FS1600, 323TB



• (7) FS1600, 323TB



• (7) FS1600, 323TB



• (7) FS1600, 323TB



Storage.Large.Density

• (7) FS1600, 1290TB



• (7) FS1600, 1290TB



• (7) FS1600, 1290TB



• (7) FS1600, 1290TB



The FDC comes with various sizing options. Below are the available options:

Note: The capacity listed is RAW capacity

https://www.juniper.net/us/en/products-services/switching/qfx-series/datasheets/1000560.page

https://www.juniper.net/us/en/products-services/switching/ex-series/datasheets/1000579.page

Documents

Hyperscale Infrastructure for All