in-memory nov 13 2013 ronan bourlier ibm flash solutions on power systems

34
© 2013 IBM Corporation IBM Client Center Ronan Bourlier [email protected] IBM Flash Systems And Oracle Improving your IO throughput and response time with IBM FlashSystems

Upload: dario-simbana

Post on 25-Jan-2016

6 views

Category:

Documents


0 download

DESCRIPTION

Computo in memory

TRANSCRIPT

Page 1: In-Memory Nov 13 2013 Ronan Bourlier IBM Flash Solutions on Power Systems

© 2013 IBM Corporation

IBM Client Center

Ronan Bourlier [email protected]

IBM Flash Systems And Oracle

Improving your IO throughput and response time with IBM FlashSystems

Page 2: In-Memory Nov 13 2013 Ronan Bourlier IBM Flash Solutions on Power Systems

© 2013 IBM Corporation

IBM Client Center

2

Agenda

IBM Oracle Center (IOC)

IBM Flash Systems TechnologyWhy Flash ? Why now ?

IBM Flash Systems Portfolio

IBM Flash Systems with Oracle DatabaseSome Benchmarks at IOC

Oracle configuration exemple

Page 3: In-Memory Nov 13 2013 Ronan Bourlier IBM Flash Solutions on Power Systems

© 2013 IBM Corporation3 18-Nov-13

The IBM Oracle Center (aka IOC)

Mixed IBM / Oracle Architectures

COVERED PRODUCTS

• IBM Platforms (System z,Power, System x, Total Storage)

• Oracle Technologies (Oracle DB, RAC, ASM, Dataguard)

• Oracle Applications (EBS, Siebel & OBI & OWI)

• Entry point to other on Industry Solutions (BRM, iFlex, RETEK, Weblogic…)

OUR MISSION

Help IBM customers to deliver integrated solutions with Oracle Software

Products on IBM Infrastructures

OUR STRENGH

Cross platform team with strong knowledge on Oracle products and a wide

network within IBM and Oracle ecosystem

OUR ACTIVITIES

• Convince : Briefings & Conferences

• Build : Architecture, Design, Sizing

• Demonstrate : Proof-of-Concept, Benchmarks

• Deliver : Publications & Workshops

Page 4: In-Memory Nov 13 2013 Ronan Bourlier IBM Flash Solutions on Power Systems

© 2013 IBM Corporation4 18-Nov-13

Agenda

IBM Oracle Center (IOC)

IBM Flash Systems TechnologyWhy Flash ? Why now ?

IBM Flash Systems Portfolio

IBM Flash Systems with Oracle DatabaseSome Benchmarks at IOC

Oracle configuration exemple

Page 5: In-Memory Nov 13 2013 Ronan Bourlier IBM Flash Solutions on Power Systems

© 2013 IBM Corporation

Evolution of Education in Storage Performance

In the

beginning,

there were iron rings and

copper coils

Data growth occurs and storage is measured by what it can hold

Storage measured by the quantity of data transferred in a given interval

Capacity Bandwidth IOPS

I/O throughput

now a function

of requests and less focus

on bytes

1970s 1980s 1990s 2000s

You

Are Here

2010s

Persistence

Page 6: In-Memory Nov 13 2013 Ronan Bourlier IBM Flash Solutions on Power Systems

© 2013 IBM Corporation

Applications see time waiting, not IOPS

I/O TimeI/O TimeNetwork

Time

Network Time

CPU TimeCPU Time

I/O Time

I/O Time

Network Time

Network Time

CPU Time

CPU Time

I/O TimeI/O TimeNetwork

Time

Network

TimeCPU

Time

CPU

Time

All client requests take time

Newer CPUs help

Newer CPUs + FlashSystem magnify the gains

Time Recovered

Query Time

Page 7: In-Memory Nov 13 2013 Ronan Bourlier IBM Flash Solutions on Power Systems

© 2013 IBM Corporation

Why Flash……..Why Now?

In the last 10 years…

� CPU Speed: Performance increase roughly 8-10x

� DRAM Speed: Performance increase roughly 7-9x

� Network Speed: Performance increase of 100x

� Bus Speed: Performance increased roughly 20x

� Disk speed: Performance increased 2X

Page 8: In-Memory Nov 13 2013 Ronan Bourlier IBM Flash Solutions on Power Systems

© 2013 IBM Corporation

Why Flash……..Why Now?

In the last 10 years… with Flash

� CPU Speed: Performance increase roughly 8-10x

� DRAM Speed: Performance increase roughly 7-9x

� Network Speed: Performance increase of 100x

� Bus Speed: Performance increased roughly 20x

� Disk speed: Performance increased 30x

Page 9: In-Memory Nov 13 2013 Ronan Bourlier IBM Flash Solutions on Power Systems

© 2013 IBM Corporation

The world of storage

Lowest Latency

Very Expensive

Consistent Performance

Low Latency

Fair Latency

Expensive HDD

Unbalanced Performance

Lowest Performance

Lower Cost

Server FlashServer Flash

SAN Shared Flash Only ArraySAN Shared Flash Only Array

Hybrid Array or SAN SSDHybrid Array or SAN SSD

Array HDDArray HDD

Server SSDServer SSD

Tier or CacheTier or Cache

ηs

ms

sec

us

Page 10: In-Memory Nov 13 2013 Ronan Bourlier IBM Flash Solutions on Power Systems

© 2013 IBM Corporation10 18-Nov-13

Agenda

IBM Oracle Center (IOC)

IBM Flash Systems TechnologyWhy Flash ? Why now ?

IBM Flash Systems Portfolio

IBM Flash Systems with Oracle DatabaseSome Benchmarks at IOC

Oracle configuration exemple

Page 11: In-Memory Nov 13 2013 Ronan Bourlier IBM Flash Solutions on Power Systems

© 2013 IBM Corporation

IBM Flash Systems

Flash Choice SLC eMLC

Models FlashSystem-710 FlashSystem-720 FlashSystem-810 FlashSystem-820

Capacity 1 -5 TB 5 or 10 TB 2 - 10 TB 10 or 20 TB

Latency R/W (microseconds)

90/25 90/25 110/25 110/25

IOPS 450K 500K 400K 450K

Bandwidth 4GB/s 5GB/s 4GB/s 4GB/s

Interface 4-Ports8Gb FC or 40Gb IB 8Gb FC or 40Gb IB

Data Protection VSR VSR+2D VSR VSR+2D

All units 1U form factor, less than 500 Watts

Page 12: In-Memory Nov 13 2013 Ronan Bourlier IBM Flash Solutions on Power Systems

© 2013 IBM Corporation

IBM FlashSystem 820/720: Logical Components

CPUs (16)

Interface Controller

Management Module

RAID Controller

Flash Modules (12)

• Hardware-only data path with extreme LOW latency– Custom FPGA-based data movement decreases latency vs. software

• Distributed out-of-data-path CPU processing• High Performance (IOPS), High Bandwidth

Page 13: In-Memory Nov 13 2013 Ronan Bourlier IBM Flash Solutions on Power Systems

© 2013 IBM Corporation

Variable Stripe RAID™ (VSR)

� Patented VSR allows RAID stripe sizes to vary.

� If one die fails in a ten-chip stripe, only the failed die is bypassed, and then data is

restriped across the remaining nine chips – Reusing the remaining capacity in both the chip & the stripe

� Higher availability; Fewer maintenance events due to flash failures

16 Planes

10 Chips

FAIL

Page 14: In-Memory Nov 13 2013 Ronan Bourlier IBM Flash Solutions on Power Systems

© 2013 IBM Corporation

Variable Stripe RAID™ (VSR)

Form Factor SSD• Flash failure = Disk failure• Requires top-level RAID• Relatively frequent hot-swaps

Enterprise Flash Drive or Memory Module• Flash failure = Degraded state

within module• Performance impact on RAID set• Hot-swap to resolve

FlashSystem with Variable Stripe RAID• Preserves Flash life• Preserves performance• Re-parity data in microseconds

Parity

Parity

No Parity

Less maintenance touches

while still preserving the life, protection, and

performance of the Day-1

experience

Page 15: In-Memory Nov 13 2013 Ronan Bourlier IBM Flash Solutions on Power Systems

© 2013 IBM Corporation

2D Flash RAID™ (FlashSystem 720/820)

RAID 5 across Flash Modules(10 data + 1 parity + 1 hot spare)

ExternalInterfaces

(FC, IB)

RAID Controllers

Variable Stripe RAID within

Flash Modules(9 data + 1 parity)

TMS2D Flash RAID™

Self-healing Flash Mgmt:•Preserves life•Preserves performance

• Flash Ctrlr Protection• Brings hot-swap to

Flash cards

Page 16: In-Memory Nov 13 2013 Ronan Bourlier IBM Flash Solutions on Power Systems

© 2013 IBM Corporation

time

CPU Utilization Waiting for I/OWaiting for I/O

Work

I/O Serviced by Disk

Issue/Process I/O request ~ 100+100 µs

Wait for I/O to be serviced ~ 5,000 µs

Waiting for I/OWaiting for I/O

Work

Waiting for I/OWaiting for I/O

Work

CPU Utilization = ~4% Less Work, More WaitLess Work, More Wait

time

CPU Utilization WorkWork

Wait

I/O Serviced by FlashSystem

Issue/Process I/O request ~ 100+100 µs

Wait for I/O to be serviced ~ 200 µs

WorkWork

Wait

WorkWork

Wait

WorkWork

Wait

WorkWork

Wait

WorkWork

Wait

How Latency translates to Application Efficiency

Page 17: In-Memory Nov 13 2013 Ronan Bourlier IBM Flash Solutions on Power Systems

© 2013 IBM Corporation17 18-Nov-13

Agenda

IBM Oracle Center (IOC)

IBM Flash Systems TechnologyWhy Flash ? Why now ?

IBM Flash Systems Portfolio

IBM Flash Systems with Oracle DatabaseSome Benchmarks at IOC

Oracle configuration exemple

Page 18: In-Memory Nov 13 2013 Ronan Bourlier IBM Flash Solutions on Power Systems

© 2013 IBM Corporation

Benchmark Setup – Stressing Test Reference on traditional hard diskbased storage

SwingBench

LPAR0 : Running SwingBench

SwingBench Partition :

Power7+

12 Go RAM

AIX 7.1 TLXXX

Hard disk based storage

Rootvg20 Go

ora_bin20 Go

ora_data2 To

LPAR1 : Running Oracle Database Enterprise Engine

VIOS2VIOS1

Oracle 11gR2

Oracle Database Partition : 11.2.0.3 Single Instance

Power7+

64 Go RAM

AIX 7.1 TLXXX

3 Volume Groups on storage (rootvg,oracle binaries,oracledatafiles)

Filesystem : JFS2

Workload

Page 19: In-Memory Nov 13 2013 Ronan Bourlier IBM Flash Solutions on Power Systems

© 2013 IBM Corporation

Benchmark Setup – Stressing Test Mix IBM Flash & Traditional hard diskbased storage

SwingBench

LPAR0 : Running SwingBench

SwingBench Partition :

Power7+

12 Go RAM

AIX 7.1 TLXXX

Hard disk based Storage

LPAR1 : Running Oracle Database Enterprise Engine

VIOS2VIOS1

Oracle 11gR2Workload

FlashSystems

rootvg20 Go

ora_bin20 Go

ora_data2 To

Oracle Database Partition : 11.2.0.3 Single Instance

Power7+

64 Go RAM

AIX 7.1 TLXXX

2 Volume Groups on storage (rootvg,oracle binaries)

1 Volume Group on IBM Flash System (oracle datafiles)

Filesystem : JFS2

Page 20: In-Memory Nov 13 2013 Ronan Bourlier IBM Flash Solutions on Power Systems

© 2013 IBM Corporation

IO rate seen from OS

� Around 5000 IOPS on the standard storage system disk/LUN

� Around 250 000 IOPS on the FlashSystems disk/LUN

� Factor 50 in performance increase

� How will this improve Oracle performances ?

Page 21: In-Memory Nov 13 2013 Ronan Bourlier IBM Flash Solutions on Power Systems

© 2013 IBM Corporation

Orion

� Orion simulates Oracle queries to storage systems

� FlashSystems improves by a factor 20 the average latency seen by Oracle Database

Orion Normal Test

0

50000

100000

150000

200000

250000

300000

350000

1 4 8 16 32 64

128

256

512

10242048409681921638432768655361310722621445242881048576

Latency (us)

IO (

#)

0

1000

2000

3000

4000

5000

6000

7000

8000

IO (

#) FlashSystems

Legacy

Page 22: In-Memory Nov 13 2013 Ronan Bourlier IBM Flash Solutions on Power Systems

© 2013 IBM Corporation

AWR Comparaisons with OLTP

Hard disk based storage

IBM FlashSystem

the “db file sequential

read” with is 15 times

better.

Give more time for

CPU.

Page 23: In-Memory Nov 13 2013 Ronan Bourlier IBM Flash Solutions on Power Systems

© 2013 IBM Corporation

Give it a try

� Estimation of performance improvement–Send us a AWR report on one hour of your most busy workload–We will send back an estimation of performance improvement thanks

to FlashSystem

� Try and Buy–Want to try it in your infrastructure ?–Contact your local IBM commercial for a try–Keep it if you like it !

Page 24: In-Memory Nov 13 2013 Ronan Bourlier IBM Flash Solutions on Power Systems

© 2013 IBM Corporation24 18-Nov-13

Agenda

IBM Oracle Center (IOC)

IBM Flash Systems TechnologyWhy Flash ? Why now ?

IBM Flash Systems Portfolio

IBM Flash Systems with Oracle DatabaseSome Benchmarks at IOC

Oracle configuration exemple

Page 25: In-Memory Nov 13 2013 Ronan Bourlier IBM Flash Solutions on Power Systems

© 2013 IBM Corporation

Solutions in Order of Ease

Page 26: In-Memory Nov 13 2013 Ronan Bourlier IBM Flash Solutions on Power Systems

© 2013 IBM Corporation

WRITES

READS

ASM FG2

ACTIVE DATA

TRANSITIONAL DATA

20 TB

5 TB

Oracle – Architecture

26

ARCHIVE DATA

100 TB

ASM FG1

ACTIVE DATA

TRANSITIONAL DATA

20 TB

5 TB

ASM

LVM

ASM

LVMBoost Performance

Boost Redundancy

- Without Disruption

- Without Risk

- Without Feature Loss

SANSAN SAN

SAN

DB Servers

MirrorMirror

IBM Flash System

Page 27: In-Memory Nov 13 2013 Ronan Bourlier IBM Flash Solutions on Power Systems

© 2013 IBM Corporation

Preferred Read Mirror

� Reads are serviced from the preferred mirror, in this case the Flash Storage

� Writes are synchronous to both mirrors

� Eliminates read poisoning of the disk array– Read poisoning: Writes are delayed by multiple head repositioning due to

reads

� Can be used with any existing SAN as long as it has preferred read capability.

� In initial testing Oracle ASM was used, but SVC will work as well

Page 28: In-Memory Nov 13 2013 Ronan Bourlier IBM Flash Solutions on Power Systems

© 2013 IBM Corporation

8,000 Reads / Sec now at extremely low latency

Preferred Read - Example

Storage does 10,000 Writes & IBM FlashStorage does 10,000 Writes & 40,000 Reads

Storage performance @10,000 IOPS for a given appRead/Write Ratio @ 80% Reads / 20% Writes

Reads: 8,000 / SecWrites: 2,000 / Sec

Introduce IBM FlashStorage as Primary Copy of new mirror

Storage was 10,000 IOPS� Now 10,000+ Writes / Sec

R/W ratio does not change; No change in the app

= 5x Previous storage performance

= 5x Previous storage performance

Texas Memory Systems, Inc., an IBM Company - The World's Fastest Storage®

Page 29: In-Memory Nov 13 2013 Ronan Bourlier IBM Flash Solutions on Power Systems

© 2013 IBM Corporation

� 1.3 Million IOPS

� 43K+ Transactions p/sec

� 13K Updates per second

A way to look at TCO and Storage Economics

Normalized $ / IOPS

Energy Space

IBMFlashSystem

2,500Spindles+ 128 SSDs

5,000 Spindles

11x Less

80x Less26x Less

IBMFlashSystem

2,500Spindles+ 128 SSDs

5,000 Spindles

IBMFlashSystem

2,500Spindles+ 128 SSDs

5,000 Spindles

ibm.com\storage\flash

Page 30: In-Memory Nov 13 2013 Ronan Bourlier IBM Flash Solutions on Power Systems

© 2013 IBM Corporation

IBM FlashSystem 820 Midrange SSD-based Array

Source: SPC-1/E Result #AE00006 (IBM FlashSystem 820), 8/16/2013, and SPC-1 Result #A00134 (HP 3PAR StoreServ 7400 Storage System (with SSDs)), 5/23/2013.

Data indicated by * is derived from SPC data, not reported directly. More footnotes coming…

Price per GBLatency

(minimum)Power Space

Capacity density (GB/U)

IOPS density (IOPS/U)

75%better

45%better

30%better

75%better

97%better

67%better

Beyond Latency: Data EconomicsSPC-1 Comparison – FlashSystem vs. SSDs

Under Embargo Until Nov. 19, 2013

Page 31: In-Memory Nov 13 2013 Ronan Bourlier IBM Flash Solutions on Power Systems

© 2013 IBM Corporation

IBM Flash Storage Sweet SpotsDo More, Do it Faster…

Cloud-scale

Infrastructures

On-demand computing, content distribution, web, caching, metadata, GPFS, active file management…

OLTP

Databases

Financial, gaming, real-time billing, trading, real-time monitoring, query

acceleration (DB2/Oracle)…

Virtual

Infrastructures

VDI, Consolidated virtual infrastructures, user profiles…

HPC,

ComputationalApplications

Simulation, modeling, rendering, FS metadata, scratch space, video on

demand, thread efficiency…

Analytical

Applications (OLAP)

Business intelligence, batch processing, ERP systems, reporting, massive data feeds…

Page 32: In-Memory Nov 13 2013 Ronan Bourlier IBM Flash Solutions on Power Systems

© 2013 IBM Corporation

85% ReductionIn batch processing

times

90% ReductionIn OLTP times

150-200 µs Latency

80% ReductionEnergy Usage

IBM Flash Impact

75% Footprint Reduction

Store one petabyte in

a single floor tile. Add

compression and add

up to 100% more

Enterprise ReliabilityHigh Availability,

2D Flash RAIDTM

and Variable

Stripe RAIDTM

Boost performance without re-architecting applications!

Page 33: In-Memory Nov 13 2013 Ronan Bourlier IBM Flash Solutions on Power Systems

© 2013 IBM Corporation33 18-Nov-13

[email protected]

How to contact us

Page 34: In-Memory Nov 13 2013 Ronan Bourlier IBM Flash Solutions on Power Systems

© 2013 IBM Corporation

DISCLAIMERNo part of this document may be reproduced or transmitted in any form without written permission from IBM Corporation.

Product data has been reviewed for accuracy as of the date of initial publication. Product data is subject to change without

notice.

This information could include technical inaccuracies or typographical errors. IBM may make improvements and/or changes in

the product(s) and/or program(s) at any time without notice. Any statements regarding IBM's future direction and intent are

subject to change or withdrawal without notice, and represent goals and objectives only.

The performance data contained herein was obtained in a controlled, isolated environment. Actual results that may be obtained

in other operating environments may vary significantly. While IBM has reviewed each item for accuracy in a specific situation,

there is no guarantee that the same or similar results will be obtained elsewhere. Customer experiences described herein are

based upon information and opinions provided by the customer. The same results may not be obtained by every user.

Reference in this document to IBM products, programs, or services does not imply that IBM intends to make such products,

programs or services available in all countries in which IBM operates or does business. Any reference to an IBM Program

Product in this document is not intended to state or imply that only that program product may be used. Any functionally

equivalent program, that does not infringe IBM's intellectual property rights, may be used instead. It is the user's responsibility

to evaluate and verify the operation on any non-IBM product, program or service.

THE INFORMATION PROVIDED IN THIS DOCUMENT IS DISTRIBUTED "AS IS" WITHOUT ANY WARRANTY, EITHER

EXPRESS OR IMPLIED. IBM EXPRESSLY DISCLAIMS ANY WARRANTIES OF MERCHANTABILITY, FITNESS FOR A

PARTICULAR PURPOSE OR INFRINGEMENT. IBM shall have no responsibility to update this information. IBM products are

warranted according to the terms and conditions of the agreements (e.g. IBM Customer Agreement, Statement of Limited

Warranty, International Program License Agreement, etc.) under which they are provided. IBM is not responsible for the

performance or interoperability of any non-IBM products discussed herein.

Information concerning non-IBM products was obtained from the suppliers of those products, their published announcements

or other publicly available sources. IBM has not tested those products in connection with this publication and cannot confirm

the accuracy of performance, compatibility or any other claims related to non-IBM products. Questions on the capabilities of

non-IBM products should be addressed to the suppliers of those products.

IBM Corporation 2013