Upload
mongodb
View
856
Download
2
Embed Size (px)
Citation preview
High Performance MongoDB Applications on IBM Power8
Keshav RanganathanAnalytics Solutions Offering Manager – Power Systems, IBM
Buzz MoschettiSenior Solutions Architect, MongoDB
May 12, 2016
+
A few reminders…
Please ask any questions in the chat box on the left side of your screen.
Buzz and Keshav will answer all questions at the end of the session.
This webinar is being recorded. Slides and recording will be sent to you tomorrow.
MongoDB
The leading post-relational database
Document Data Model
Open-Source
Fully Featured
High Performance
Scalable
{ name: “John Smith”, pfxs: [“Dr.”,”Mr.”], address: “10 3rd St.”, phone: {
home: 1234567890, mobile: 1234568138 }}
MongoDB Company Overview
~550 employees 2200+ customers
Over $311 million in fundingOffices in NY & Palo Alto and
across EMEA, and APAC
The best way to run MongoDB in your data
center.
Automated.
Supported.
Secured.
What’s included?
Enterprise-Grade Support
Commercial License
Ops Manager or Cloud Manager Premium
Encrypted & In-Memory Storage Engines
Compass (GUI for exploration)
BI Connector
Advanced Security
Platform Certification
On-Demand Training
MongoDB Enterprise Edition
Leading Organizations Use MongoDB
`
Gartner Says MongoDB is a Leader
{ _id: “123”, title: "MongoDB: The Definitive Guide", authors: [ { _id: "kchodorow", name: "Kristina Chodorow“ }, { _id: "mdirold", name: “Mike Dirolf“ } ], published_date: ISODate(”2010-09-24”), pages: 216, language: "English", thumbnail: BinData(0,"AREhMQ=="), publisher: { name: "O’Reilly Media", founded: 1980, locations: ["CA”, ”NY” ] }}
What is a Document?
Nexus Architecture
Scalability& Performance
Always On,Global Deployments
FlexibilityExpressive Query Language
& Secondary Indexes
Strong Consistency
Enterprise Management& Integrations
MongoDB 3.0 Set The Stage…
7x-10x Performance, 50%-80% Less Storage
How: WiredTiger Storage Engine•Same data model, query language, & ops•100% backwards compatible API•Non-disruptive upgrade•Storage savings driven by native compression•Write performance gains driven by
• Document-level concurrency control• More efficient use of HW threads
•Much better ability to scale vertically
MongoDB 3.0MongoDB 2.6
Performance
MongoDB 3.4 + POWER: Efficient, Performant MongoDB
• Much better ability to scale vertically
+ • Document Validation Rules• Encryption at rest• BI Connector (SQL bridge)• Compass• New Relic & AppDynamics integration• Backup snapshots on filesystem• $lookup (“left outer JOIN”)
4X
Threads per core*
4X Mem. Bandwidth*
4XMore cache* @ Lower Latency
SMT=Simultaneous Multi-Threading OLTP = On-Line Transaction Processing
These design decisions result in best performance for data centric workloads like:
Database, NoSQL, Big Data Analytics, OLTP
POWER8: Designed for data to deliver breakthrough performance
POWER8SMT8
x86Hyperthread
Parallel Processing
POWER8pipe
Data flow
x86 pipe POWER8
x86 POWER8 + OpenPOWER
x86
Power Scale-out servers are affordable, easy-to-deploy and energy efficient with superior cloud economics and security.
Scale-out Systems• 1 or 2 sockets
• Up to 24 cores
• Up to 192 threads
• Up to 2 TB memory
• Running AIX, IBM i, Linux
Enterprise Systems*• Up to 16 sockets
• Up to 192 cores
• Up to 1536 threads
• Up to 16 TB memory
• Running AIX, IBM i, Linux
Power Enterprise servers are designed for data and offer businesses the ultimate in resiliency, availability, security, and performance.
IBM Power Systems with POWER8, designed to take on the most complex data challenges
Supported by Canonical
Community / 3rd Party Support
Converged
The IBM Power Systems Linux Portfolio
•Design and TCA cost optimized for cognitive workloads on scale out deployments (cloud and cluster)
•Solutions for cognitive, HPC and Big Data IBM Support
running
The LC Line
The L Line
PurePower
Enterprise& IFLs
IaaS
Scale-Out, Linux-Only
ConvergedInfrastructure
Scale-Up
•Enterprise level RAS for single system deployments
•Solutions for Big Data & Analytics
•Converged infrastructure offering
•Rapid time to value and simplicity of management
•Enterprise level robustness and IFL capability
•Solution editions for in memory databases
•(HANA, DB2 BLU)
•Hosted cloud and hybrid cloud solutions
•Rapid deployments and POCs
System CostServer + RHEL OS + MongoDB Subscription
$34,161($20,872 + $1,299 + $11,990)
$38,976($25,687 + $1,299 + $11,990)
MongoDB result - YCSBTotal operations per second (OPS)
297.5 k 169.5 k
OPS per dollar 8.7 ops/$ 4.35 ops/$
100% Better
IBM Power S822LC
(16-core, 256GB)
$20,872
HP DL380 Gen9(24-core, 256GB)
$25,687
MongoDBon
Linux on POWER8vs.
Linux on Intel Haswell
Significantly reduce operating costs running MongoDB on IBM Power SystemsLower TCA + higher performance = more efficient use of IT dollars
• Based on IBM internal testing of single system and OS image running Yahoo Cloud Services Benchmark (YCSB) 0.6.0, workload at 50/50 read/write factor. Conducted under laboratory condition, individual result can vary based on workload size, use of storage subsystems & other conditions.
• IBM Power System S822LC; 16 cores (2 x 8c chips) / 128 threads, POWER8; 3.3 GHz, 256 GB memory, MongoDB 3.3.5 RHEL 7.2. Competitive stack: HP Proliant DL380 Gen9; 24 cores (2 x 12c chips) / 48 threads; Intel E5-2690 v3; 2.6 GHz; 256 GB memory, MongoDB 3.3,5 RHEL 7.2 . Both server priced with 2 x 1TB SATA 7.2K rpm HDD, 1 Gb 2-port, 2 x 16gbps FCA. Configurations represent the highest processor frequency for that specific processor running the MongoDB server on 1 socket & the YCSB application workload on the 2nd socket. Both systems used fiber attached file system on IBM Flash 900.
Pricing is based on web pricing for S822LC http://www-03.ibm.com/systems/power/hardware/s822lc-commercial/buy.html and HP DL380 Gen9 https://h22174.www2.hp.com/SimplifiedConfig/Index MongoDB https://www.mongodb.com/compare/mongodb-oracle Page: 6
2.6Xbetter per
core
So Why are We Seeing an Advantage on P8??
Power 8 Threads per coreSMT 8
4.2 GHz
X86 2 Threads per coreSMT 2
2.5 GHz
Client 2: On-premises infrastructure – Consumer creditWorkload: Internal customer proprietary application workload
Internal business process management softwareObjective: Show over 2 x higher system level performance and delivers price performance advantage.Comparison: Intel 2 socket x86 IvyBridgeIBM solution: S824L servers – 4 core VM
Integrate with existing customer storage environmentPOC result: POC result: 3.7x per-core over x86, Initial 50 servers sold.
Experiences on Key Workload – in Customer POCs
Client 1: Database as a Service – Major TelComWorkload: Yahoo Cloud Services Benchmark (YCSB) – Java-
based workload with non-industry specific data with selectable DB size & read/write mix
Objective: Show better per-VM performance and higher VM density to deliver price performance advantage.
Comparison: Intel 12-c / 2 socket x86 Haswell (HP)IBM solution: S822LC 20-c server – 4 core VMs
Integrate with existing FCoE 40GB Cloud Storage Offer additional storage growth on IBM ESS
POC result: 5-20% better performance per VM with 4 x VM density
Both highly virtualized open source cloud environments & on-premises system of record
deployments are showing major gains in performance and price efficacy on Linux
on Power
Rapid Build Program
“The combination of Linux on IBM POWER8 with MongoDB Enterprise Server makes an ideal enterprise Database-as-a-Service platform. Leveraging POWER8’s performance benefits and the RapidBuild program, enterprises can build MongoDB-as-a-Service and private cloud environments that reduce installation time, enhance security, reduce TCO, and ensure high availability.”
Alan Chhabra, Vice President, WW Partners at MongoDB
19
MongoDB-As-A-Service
The Main Goal:•Efficient provisioning of MongoDB
But Also:•Better control / transparency of database use (data too!)•Economy of scale in monitoring and management
• Backup / restore• Version upgrades
•Opportunity to employ several cost allocation models•Opportunity for self-service
20
Macroarchitecture
Developer / DBA
PortalOpsManager
OtherProvisioningTechnology
(puppet, chef, …)
Application
ManagedPool Of
Servers &Storage
ServiceBookkeeping
Service Support
App Servers &
DBsCompassBI Connector
ApplicationApplication
Application
DevOps Infrastructure Engineering & Operations
21
Dedicated Replica Set Per Application
4 cores32G RAM
4 cores32G RAM
4 cores32G RAM
App 1
8 cores64G RAM
8 cores64G RAM
8 cores64G RAM
App 2
4 cores32G RAM
4 cores32G RAM
4 cores32G RAM
App n
…
Pros:Greatest de-coupling of infrastructure risk and cross application riskAbility to have very different, specific machine configs per appEasy to model allocations / chargeback
Cons:XNo on-demand capability and generally lengthy rack and wire-up timeXPower and data center space hogXFear of underprovisioning bare metal leads to chronic overprovisioning
GRADE: B-
= locally attached storage
22
Dedicated Replica Set + Centralized Storage
4 cores32G RAM
4 cores32G RAM
4 cores32G RAM
App 1
4 cores32G RAM
4 cores32G RAM
4 cores32G RAM
App 2
4 cores32G RAM
4 cores32G RAM
4 cores32G RAM
App n
…
Pros:Potentially easier to manageLarger volume sizes possible
Cons:XNo on-demand capability and generally lengthy rack and wire-up timeXPower and data center space hogXRisk of storage fabric failure
GRADE: C+
23
“Striping”: Efficient MaaS
Pros:Far fewer machinesOn-demand creation of clusterEasy to vary size of “stripe” Can use locally attached storage or arrays
Cons:XMachine failure OR maintenance triggers many voting events
=
Ap
p 1
Ap
p 2
Ap
p 3
Ap
p n
GRADE: A
Q&A
+
Thank you