© 2014 IBM Corporation The Economics and Value of Power Systems versus Linux on Intel

© 2014 IBM Corporation

The Economics and Value of Power Systems

versus Linux on Intel

What adds value when evaluating a platform for Critical Commercial Workloads?

Investment Protection – minimal platform disruptions with timely technology advancements (both HW and ISV SW)

Reliability, Availability and Serviceability Performance and Workload Priority Scalability for intended and unexpected growth Security – No one wants to be (a) Target And cost savings would be icing on the cake.

2

Investment ProtectionFor both HW and SW

3

Oracle Customer Base (Estimates)

6,000+ IBM Migration Factory competitive migrations since 2009

~2,300+ from Sun/Oracle

< Oracle Claims over 10,000 “Exa” Systems installed since 2008 (1/4 Rack UoM) ~3,000 customers

Exadata OLTP workloads (custom + all ISV)

Exalogic is a fraction of these

390,000 Oracle customers

>56% UNIX footprint on Power(19% on SPARC)

40,000 Oracle ISV apps on IBM

5

SPARC SuperCluster

.

Oracle Customer Base for Engineered Systems

Reliability, Availability and Serviceability

6

Area POWER8 Processor Based X86 Processor Based (2014)

Processor Core

Advanced Soft Error Resilient Technology

Yes: Hardened Stacked Latches and eDRAM on SOI

Limited: Improved Over Previous Generations

Hardware Directed Core/Cache Soft Error Handling

Yes:Processor Instruction RetryCache Line Purge and Delete

No: Software driven Machine Check Handling Only

Transparent Repair/Self-Healing In Processor for solid faults

Yes: Cache line/set delete/purgeL2/L3 Cache dynamic column repairAlternate Processor Recovery

No:Predictive Processor Deallocation and machine check handling only

Processor and I/O Infrastructure

Dedicated Service Processor Yes:OS Independent Automatic Error Analysis/Repair

For All failure modes

With full access to processor fault internals

No:Optional Service Processor

Not capable of full Analysis/Repair

Lacks full access to processor fault internals

Redundant Infrastructure Support Yes:Redundant Clock Interfaces SupportedRedundant Service Processor Interfaces

No:Single clock input for each processor

Power/Thermal Handling Yes:On Chip Controller

Yes:Built in to processor

Complete I/O Enhanced Error Handling

Yes:Integrated PCI Controller FreezePCIe Link retrain/retry

Limited:PCIe Link retrain/retry

Features may not be supported in all systems or software configurations

RAS Compare: POWER8 to E5 v2 and E7 v2 Processors

Area POWER8 Processor Based X86 Processor Based (2014)

Memory

Processor to Memory Bus Protection

Yes: Retry for soft errors in

memory controller

Memory bufferBus Retrain with Data Lane Sparing

Limited: Improved Over Previous Generations

Advanced DRAM Module Error Correction

Yes:Chipkill plus protection all DIMMs

Can handle 1 bad DRAM module for every 18 DRAMs

Limited: Chipkill plus protection on x4 DIMMS only

Can handle 1 bad DRAM module for every 18 DRAMS (performance mode) or

Can handle 2 bad DRAM module for every 36 DRAMS (RAS mode)

DIMM/DRAM Module Sparing Yes: Additional Spare DRAM for every 9 DRAMs

Limited: No DRAM module Sparing Can dedicate an entire Rank for each DIMM as spare for predictive errors (reduced memory capacity)

Solid Core Fault Handling Yes:Alternate Processor RecoveryCore Contained Checkstops

Limited: Software driven Machine Check Handling Only

Memory Mirroring Yes: Able to mirror of logical memory blocksSupports PowerVM (hypervisor) mirroring

Yes: Can mirror entire system memorySupport for hypervisor mirroring of OS and partitions depending on software used

Features may not be supported in all systems or software configurations

RAS Compare: POWER8 to E5 v2 and E7 v2 Processors

Power RAS is built into the platform (HW, OS, PVM) so clients do not have to dedicate scarce resources to prepare for downtime

• With built-in RAS, the platform comes close

to maintaining itself• 67% of corporations now require a minimum

of 99.99% uptime or better for mission critical

hardware, operating systems and main line of

business (LOB) applications• AIX on Power consistently has the least

amount of downtime in ITIC studies for

several years• Industry leading availability for all workloads,

including SAP

Source: ITIC 2013 Global Server Hardware, Server OS Reliability Survey, ITIC, (All rights reserved); January 2013.

Power exhibits only6.6 minutes of

planned downtime per year

9

Performance and Workload Priority

for confidence in consolidation

10

Key concept: workload management and virtualization

• Virtualized hosting platforms must be able to differentiate between high priority and low priority workloads when sharing resources

– The ability to maintain priority enables maximum utilization of the hosting platform– An inability to maintain priority means that separate platforms for high and low priority

workloads will be required, increasing the cost and complexity

• Power Systems workload management is nearly perfect when mixing workloads– Low priority workloads “donate” resources to high priority workloads when required, soak up

unused resources when available

• Hypervisors like VMware on Intel and Oracle VM on SPARC permit “leakage” where low priority workloads steal resources from high priority workloads

11

PowerVM workload management is nearly perfect when mixing workloads

High Priority Workload

Workload MetricsTotal Throughput: 14.42M

Run High And Low Priority Workloads Together

High Priority Workload MetricsTotal Throughput: 12.95M

10.2% throughput reduction

12

VMware workload management loses 30% throughput when mixing workloads

High Priority Workload

Workload MetricsTotal Throughput: 6.48M

Run High And Low Priority Workloads Together

High Priority Workload MetricsTotal Throughput: 4.48M

30.7% throughput reduction

13

Power8 E870 systems provide the best virtualization platform in the industry

• 2.60x VM’s per core73% more VM’s2 supported on only 40 POWER8 cores versus 60 E7-4890 v2 cores (codename: Ivy Bridge-EX)

• 2.70x Throughput per core 80% more system throughput3 on only 40 POWER8 cores versus 60 E7-4890 v2 cores (codename: Ivy Bridge-EX)

POWER8 Enterprise Servers – World’s Leading Virtualization PlatformEach POWER8 core supports over 2.6x the VMs1 of the best Intel Xeon x86 E7 class server core

Do not leave behind with client; presentation use only

1 Performance based on IBM Internal test of TPoX benchmark TPoX is an open-source benchmark developed by IBM in collaboration with Intel and others. It is available at: http://tpox.sourceforge.net/tpoxresults.htm. This study used TPoX version 2.1, accessing an IBM DB2 version 10.1 backend database.

2 Over 70% more VM’s on a POWER8 Enterprise E870 (40c/4.19 GHz POWER8) versus an HP DL580p (60c/2.7GHz E7-4890 v2) at equal throughput per VM with VMware vSphere 5.5 Upd1 at equal throughput per VM3 Up to 80% more throughput on a POWER8 Enterprise E870 (40c/4.19 GHz POWER8) versus an HP DL580p (60c/2.7GHz E7-4890 v2) at equal number of VMs with VMware vSphere 5.5 Upd1 at equal throughput per VM

• Published Benchmarks – • ALL data is PUBLISHED

x86“Ivy Bridge”

IBMPOWER S824

POWER8 vs. x86 Core Performance Ratio

Intel Xeon E5-2697 v2(except where noted)

POWER8 @ 3.5 GHz

# Cores 24 24

SAP 2-Tier 10253 21212 2.1

SPECint_rate2006 1020 1750 1.7

SPECfp_rate2006 734 1370 1.9

SPECjbb2013 (max-jOPS) 63079 361293 5.7

SPECjEnterprise2010 11260 22543 2.0

Oracle eBS 12.1.3 Payroll 10176391090909

(12-core)2.1

Siebel CRM Release 8.1.1.4 10000(16-core E5-2690)

50000(6-core)

13.3

Performance Comparison – POWER8 vs. x86 E5IBM POWER8 core and system performance is up to 5.7x the x86 Xeon E5-2697 v2 core performance

1) IBM Power System S824 on the two-tier SAP SD standard application benchmark running SAP enhancement package 5 for the SAP ERP 6.0 application; 4 processors / 24 cores / 96 threads, POWER8; 3.52GHz, 512 GB memory, 21,212 SD benchmark users, running AIX® 7.1 and DB2® 10.5, Certification # 2014016. Source: http://www.sap.com/benchmark All results valid as of October 3, 2014

2) Dell PowerEdge R730, on the two-tier SAP SD standard application benchmark running SAP enhancement package 5 for the SAP ERP 6.0 application; 2 processors/36 cores/72 threads, Intel Xeon Processor 2699v3; 2.30 GHz, 256 GB memory; 16,500 SD benchmark users, running RHEL 7 and SAP ASE 16; Certification # 2014033. Source: http://www.sap.com/benchmark.

3) SPECcpu2006 results are submitted as of 9/8/2014. For more information go to http://www.specbench.org/cpu2006/results/ 4) SPECjbb2013 results are submitted as of 10/15//2014. For more information go to http://www.specbench.org/jbb2013/results 5) SPECjEnterprise2010 results are valid as of 9/8/2014. For more information go to http://www.specbench.org/jEnterprise2010/results/ 6) Oracle eBS 12.1.3 Payroll Batch Extra Large Kit and are current as of 3/24/2014. For more information go to http://www.oracle.com/us/solutions/benchmark/apps-benchmark/results-166922.html7) Siebel 8.1.1.4 PSPP Kit and are current as of 3/24/2014. For more information go to http://www.oracle.com/us/solutions/benchmark/white-papers/siebel-167484.html

http://www.specbench.org/cpu2006/results/

http://www.specbench.org/jbb2013/results

http://www.specbench.org/jEnterprise2010/results/

http://www.oracle.com/us/solutions/benchmark/apps-benchmark/results-166922.html

http://www.oracle.com/us/solutions/benchmark/white-papers/siebel-167484.html

• Industry Standard Benchmarks – • All Ivy Bridge performance numbers are IBM internal projections and publishes where available• IBM S824 data is published/projected

x86“Ivy

Bridge”

IBMPower S824


Intel Xeon E5-2697 v2

Power 8 @ 3.5 GHz

P8 Util: 100%x86 Util: 100%

P8 Util: 65%x86 Util: 40%


# Cores 24 24Benchmark Utilization

Utilization with virtualized x86

Utilization without virtualized x86

OLTP 2100 3585 1.7 2.8 5.5

ERP SAP 2-Tier 10253 21212 2.1 3.4 6.7

SPECjbb2013 (max-jOPS) 63079 361293 5.7 9.3 18.6

SPECint_rate 1020 1750 1.7 2.8 5.6

SPECfp_rate 734 1370 1.9 3.0 6.0

SPECjEnterprise2010 11260 22543 2.0 3.3 6.5

Published Projected

LEGEND:

Do not leave behind with client, presentation use only

Core Performance Comparison – POWER8 vs. x86IBM POWER8 core performance is up to 18.6x the x86 Xeon E5-2697 v2 core performance (typical customer utilization)

• Published Benchmarks – ALL data is PUBLISHED

x86“Haswell”

IBM

POWER S824



(except where noted)

POWER8

@ 3.5 GHz

# Cores 36 24

SAP 2-Tier 16500 21212 1.9

SPECint_rate2006 1400 1750 1.8

SPECfp_rate2006 942 1370 2.1

SPECjbb2013 (max-jOPS) 195119 361293 2.7

SPECjEnterprise2010 11260(24-core E5-2697 v2)

22543 2.0

Oracle eBS 12.1.3 Payroll 1017639(24-core E5-2697 v2)

1090909(12-core)

2.1

Siebel CRM Release 8.1.1.4 10000(16-core E5-2690)

50000(6-core)

13.3

Performance comparison – POWER8 vs. x86 E5IBM POWER8 core and system performance is leadership versus the x86 Xeon E5-2699 v3

Updated with Haswell



3) SPECcpu2006 results are submitted as of 9/8/2014. For more information go to http://www.specbench.org/cpu2006/results/ 4) SPECjbb2013 results are submitted as of 10/15//2014. For more information go to http://www.specbench.org/jbb2013/results 5) SPECjEnterprise2010 results are valid as of 9/8/2014. For more information go to http://www.specbench.org/jEnterprise2010/results/ 6) Oracle eBS 12.1.3 Payroll Batch Extra Large Kit and are current as of 3/24/2014. For more information go to http://www.oracle.com/us/solutions/benchmark/apps-benchmark/results-166922.html7) Siebel 8.1.1.4 PSPP Kit and are current as of 3/24/2014. For more information go to http://www.oracle.com/us/solutions/benchmark/white-papers/siebel-167484.html






• Industry Standard Benchmarks – • All Intel performance numbers are IBM internal projections and publishes where available• IBM S824 data is published/projected

x86E5

IBMPower S824



Power 8 @ 3.5 GHz

P8 Util: 100%x86 Util: 100%



# Cores 36 24Benchmark Utilization


Utilization without virtualized x86

OLTP 2400 3585 2.2 3.6 7.2

ERP SAP 2-Tier 16500 21212 1.9 3.2 6.3

SPECjbb2013 (max-jOPS) 195119 361293 2.7 4.5 9.0

SPECint_rate 1430 1750 1.8 2.9 5.9

SPECfp_rate 965 1370 2.1 3.4 6.8

SPECjEnterprise2010 16500 22543 2.0 3.3 6.5

Published Projected

LEGEND:

Do not leave behind with client, presentation use only

Core Performance Comparison – POWER8 vs. x86IBM POWER8 core performance is up to 9.0x the best x86 Xeon E5 performance (typical customer utilization)




3) SPECcpu2006 results are submitted as of 9/8/2014. For more information go to http://www.specbench.org/cpu2006/results/ 4) SPECjbb2013 results are submitted as of 10/15/2014. For more information go to http://www.specbench.org/jbb2013/results 5) SPECjEnterprise2010 results are valid as of 9/8/2014. For more information go to http://www.specbench.org/jEnterprise2010/results/ 6) Oracle eBS 12.1.3 Payroll Batch Extra Large Kit and are current as of 3/24/2014. For more information go to http://www.oracle.com/us/solutions/benchmark/apps-benchmark/results-166922.html7) Siebel 8.1.1.4 PSPP Kit and are current as of 3/24/2014. For more information go to http://www.oracle.com/us/solutions/benchmark/white-papers/siebel-167484.html






• Infrastructure Software Price-performance has been REDUCED on Intel servers– Assumes flat system pricing

Software Licensing has increased by 1.5x– 12 cores versus 8 cores OR 18 versus 12

Performance has not increased proportionally to the chip core count resulting in higher software costs

– x86 publishes on 2-socket systems

POWER8 moves forwards while Xeon moves backwards IBM POWER systems continue to deliver improved system performance and more value per SW $ spent


x86“Sandy Bridge”

x86“Ivy

Bridge”

x86“Haswell”

SystemPerformance

Ratio

POWER7+ POWER8 SystemPerformance

Ratio

2-socket E5-2690

2-socket E5-2697v2

2-socket E5-2699v3

SNB to

IVB

SNB to

HAS

IVB to

HAS

2-socket POWER7+

2-socket POWER8

POWER7+ to POWER8

# Cores 16 24 36 1.50 2.25 1.50 16 24 1.50

ERP SAP 2-Tier 7960 10253 16500 1.29 2.07 1.61 10000 21212 2.12

SPECint_rate 693 1020 1400 1.47 2.02 1.37 884 1750 1.98

SPECfp_rate 510 734 942 1.44 1.85 1.28 602 1370 2.28

SPECjbb2013 N/A 63079 195119 - - 3.09 NA 361293 -

SPECjEnterprise2010

8310 11260 N/A 1.35 - - 13161 22543 1.71

3.09 performance gain came from new version of Java and increased

memory (4x more)



3) SPECcpu2006 results are submitted as of 9/8/2014. For more information go to http://www.specbench.org/cpu2006/results/ 4) SPECjbb2013 results are submitted as of 9/8/2014. For more information go to http://www.specbench.org/jbb2013/results 5) SPECjEnterprise2010 results are valid as of 9/8/2014. For more information go to http://www.specbench.org/jEnterprise2010/results/ 6) Oracle eBS 12.1.3 Payroll Batch Extra Large Kit and are current as of 3/24/2014. For more information go to http://www.oracle.com/us/solutions/benchmark/apps-benchmark/results-166922.html7) Siebel 8.1.1.4 PSPP Kit and are current as of 3/24/2014. For more information go to http://www.oracle.com/us/solutions/benchmark/white-papers/siebel-167484.html

POWER8 is 89% better at the system level and 2.7x the core

performance






• Published Benchmarks – • ALL data is PUBLISHED or SUBMITTED (SPECint_rate2006 and SPECfp_rate2006 on the

POWERS824)

Performance Comparison – POWER8 vs. x86 E7IBM POWER8 core performance is up to 7x the x86 Xeon E7-4890 v2 core performance

x86“Ivy Bridge”

IBMPOWER S824


Intel Xeon E7-x890 v2

POWER8 @ 3.5 GHz

P8 Util: 100%x86 Util: 100%



# Cores 60 24 Benchmark Utilization


Utilization with nonvirtualized

x86

SAP 2-Tier 25000 21212 2.1 3.4 6.9

SPECint_rate2006 2400 1750 1.8 3.0 5.9

SPECfp_rate2006 1770 1370 1.9 3.1 6.3

1) SAP results are based on the two-tier SAP SD standard application benchmark running SAP enhancement package 5 for the SAP ERP 6.0 application. Results valid as of April 28, 2014. Source: http://www.sap.com/benchmark 2) SPECcpu2006 results are submitted as of 4/22/2014. For more information go to http://www.specbench.org/cpu2006/results/

http://www.sap.com/benchmark


POWER8 core performance2 improvements surpass lagging x86 core performance

POWER8 core provides 35% better performance per SW license $ than POWER7+

– Intel Ivy Bridge Core degrades performance per SW license $ by 10% versus Sandy Bridge

1 Based on generational comparisons of SW that utilizes per core pricing and 50% more cores in per system (Power: 8c POWER7 to 12c POWER8; x86:8c E5-2690 to 12c E5-2697 v2)

2 Performance is based on published x86 data and published/projected POWER7+ & POWER8. Workloads are ERP, Integer, Floating Point, Java

POWER8 moves forwards while x86 moves backwards IBM POWER processors continue to deliver improved core performance – up to +35% versus POWER7+ while Intel went backwards (-10%) with Ivy Bridge versus Sandy Bridge.

21

2.17x ERP users per core with Power E870 with DB2 versus the competition on SAP Sales and Distribution 2-Tier BenchmarkNearly 1,000 Users per Core with POWER8 based E870

SD Benchmark Users per

Core

2.17X more users

3.18X more users

(1) IBM Power Enterprise System E870 on the two-tier SAP SD standard application benchmark running SAP enhancement package 5 for the SAP ERP 6.0 application; 8 processors / 80 cores / 640 threads, POWER8; 4.19GHz, 2048 GB memory, 79,750 SD benchmark users running AIX® 7.1 and DB2® 10.5, Certification #: 2014034 Result valid as of October 3, 2014. Source: http://www.sap.com/benchmark.(2) IBM Power System S824 on the two-tier SAP SD standard application benchmark running SAP enhancement package 5 for the SAP ERP 6.0 application; 4 processors / 24 cores / 96 threads, POWER8; 3.52GHz, 512 GB memory, 21,212 SD benchmark users, running AIX® 7.1 and DB2® 10.5, Certification # 2014016. Source: http://www.sap.com/benchmark All results valid as of October 3, 2014(3) Dell PowerEdge R730, on the two-tier SAP SD standard application benchmark running SAP enhancement package 5 for the SAP ERP 6.0 application; 2 processors/36 cores/72 threads, Intel Xeon Processor 2699v3; 2.30 GHz, 256 GB memory; 16,500 SD benchmark users, running RHEL 7 and SAP ASE 16; Certification # 2014033. Source: http://www.sap.com/benchmark. (4) Oracle SPARC Server M5-32 on the two-tier SAP SD standard application benchmark running SAP enhancement package 5 for the SAP ERP 6.0 application; 32 processors/192 cores/1536 threads, SPARC M5; 3.60 GHz, 4,096 GB memory; 85,050 SD benchmark users, running Solaris® 11 and Oracle 11g; Certification # 20013009. Source: http://www.sap.com/benchmark. (5) Dell PowerEdge R920 on the two-tier SAP SD standard application benchmark running SAP enhancement package 5 for the SAP ERP 6.0 application; 4 processors/60 cores/120 threads, Intel Xeon Processor 4890 v2; 2.80 GHz, 1024 GB memory; 25,451 SD benchmark users, running SUSE Linux Enterprise Server 11 and SAP ASE 16; Certification # 2014011. Source: http://www.sap.com/benchmark. (6) Oracle SPARC Server M6-32 on the two-tier SAP SD standard application benchmark running SAP enhancement package 5 for the SAP ERP 6.0 application; 32 processors/384 cores/3072 threads, SPARC M6; 3.60 GHz, 16 TB memory; 140,000 SD benchmark users, running Solaris® 11 and Oracle 11g; Certification # 20014008. Source: http://www.sap.com/benchmark. (7) Oracle SPARC Server T5-8 on the two-tier SAP SD standard application benchmark running SAP enhancement package 5 for the SAP ERP 6.0 application; 8 processors/128 cores/1024 threads, SPARC T5; 3.60 GHz, 2,048 GB memory; 40,000 SD benchmark users, running Solaris® 11 and Oracle 11g; Certification # 2013008. Source: http://www.sap.com/benchmark.

SAP and all SAP logos are trademarks or registered trademarks of SAP AG in Germany and in several other countries. All other product and service names mentioned are the trademarks of their respective companies.

SAP SD Standard Application Benchmark Results, 2-Tier: SD Benchmark Users per CoreSAP enhancement package 5 for SAP ERP 6.0

Source: http//www.sap.com//benchmark

IBM E870

POWER8 AIX / DB2

8p/80c/640t

DellPowerEdge R920

E7-4890 v2SUSE / SAP ASE

4p/60c/120t

OracleT5-8T5

Sol / Oracle EE8p/128c/1024t

IBM S824

POWER8 AIX / DB2

4p/24c/192t

DellPowerEdge R730

E5-2699 v3RHEL / SAP ASE

2p/36c/72t

OracleM5-32

M5Sol / Oracle EE32p/192c/1536t

OracleM6-32

M6Sol / Oracle EE 32p/384c/3072t








POWER8 processing cores are the fastest in the industry for Java code

• 2.84x more performance per core than Intel E5-2699 v3 offerings (codename: Haswell-EP)

• 3.92x more performance per core than Intel E7-4890 v2 offerings (codename: Ivy Bridge-EX)

• 4.53x more performance per core than Oracle T5 offerings

POWER8 demonstrates the fastest Java code1 performance in the industry3.92x per core performance of the best Intel Xeon x86 E7- server core

OracleT5-2T5

32c/256t

CiscoUCS C460 M4

E7-4890 v260c/120t

HPDL 380p

E5-2699 v336c/72t

IBM E870

POWER8 80c/640t

1 Performance comparisons based on published SPECjbb2013 results as of October 1, 2014 http://www.spec.org/jbb2013/results/ . All IBM benchmark results will be submitted to spec.org on October 15, 2014.

http://www.spec.org/jbb2013/results/

2x to 2.4x core performance advantage with E870versus 8-socket x86 Ivy Bridge-EX across key workloads

Java – SPECjbb2013 (Max-jOPS)4.1x Performance

ERP – SAP 2-Tier (Users) 2.4x Performance

SPECint_rate20062.0x Performance

SPECfp_rate20062.2x Performance

• Results are based on best published per core results on Xeon E7-8890 processor.• SAP results are based on the two-tier SAP SD standard application benchmark running SAP enhancement package 5 for the SAP ERP 6.0 application. Results valid as of October 3, 2014. IBM Power Enterprise System E870

on the two-tier SAP SD standard application benchmark running SAP enhancement package 5 for the SAP ERP 6.0 application; 8 processors / 80 cores / 640 threads, POWER8; 4.19GHz, 2048 GB memory, 79,750 SD benchmark users running AIX® 7.1 and DB2® 10.5, Certification #: 2014034 Result valid as of October 3, 2014. Source: http://www.sap.com/benchmark. IBM System x3950 X6 on the two-tier SAP SD standard application benchmark running SAP enhancement package 5 for the SAP ERP 6.0 application; 8 processors/ 120 cores/ 240 threads, Intel Xeon Processor 8890 v2; 2.80 GHz, 1024 GB memory; 49,000 SD benchmark users, running Windows Server 2012 Standard Edition and DB2 10; Certification # 2014024. Source: http://www.sap.com/benchmark .

• SPECjbb2013 results are valid as of 10/2/2014. For more information go to http://www.specbench.org/jbb2013/results/ All IBM benchmark results will be submitted to spec.org on October 15, 2014.• SPECcpu2006 results are submitted as of 10/2/2014. For more information go to http://www.specbench.org/cpu2006/results/ All IBM benchmark results will be submitted to spec.org on October 6, 2014.

Intel Xeon E7-8890 v2IBM x3950 X6

Win 2012 / DB28s/120c/240t

POWER8IBM E870

8s/80c/640t

Intel Xeon E7-8890 v2HP Converged System

8s/120c/240t

POWER8IBM E870AIX / DB2

8s/80c/640t

POWER8IBM E870

8s/80c/640t

Intel Xeon E7-8890 v2Fujitsu PRIMEQUEST2800E

8s/120c/240t

POWER8IBM E870

8s/80c/640t

Intel Xeon E7-8890 v2Huawei RH8100 V3

8s/120c/240t



http://www.specbench.org/jbb2013/results/


POWER8 S822L will deliver over 2.5x the performance of the best published x86 system

… and continues to offer far superior RAS

POWER8 exploits additional cores, more threads, larger caches, memory bandwidth

Terasort is a popular benchmark to measure the performance of a Hadoop solution

Sorts a large dataset (10 TB) in parallel Exercises the Map-reduced framework and

Hadoop Distributed File System (HDFS)

IBM Analytics Stack: IBM Power System S822L; 24 cores / 192 threads, POWER8; 3.0GHz, 512 GB memory, RHEL 6.5, InfoSphere BigInsights 3.0

http://www.cisco.com/en/US/solutions/collateral/ns340/ns517/ns224/ns944/le_tera.pdf

POWER8 delivers 2.5x performance on Big Data / HadoopTerasort benchmark on a POWER8 doubles the system capacity of the best x86 published result

25

2.5x2.5x

http://www.cisco.com/en/US/solutions/collateral/ns340/ns517/ns224/ns944/le_tera.pdf

(1) Performance data based on best published peak SPECint_rate2006 results. All results can be found at www.spec.org(2) Intel chip prices based on Intel published list prices. All chips and chip prices can be found at http://www.intc.com/pricelist.cfm

26

Intel Ivy Bridge-EPSPECint2006_rate

(2-socket)

Proc. # Cores Freq.List

Price (kU)

ResultS824 vs x86 SKU

E5-2697 v2 12 2700 2614 1020 1.72

E5-2695 v2 12 2400 2336 921 1.90

E5-2690 v2 10 3000 2057 900 1.94

E5-2680 v2 10 2800 1723 860 2.03

E5-2670 v2 10 2500 1552 816 2.14

E5-2660 v2 10 2200 1329 748 2.34

E5-2650 v2 8 2600 1166 689 2.54

E5-2640 v2 8 2000 885 542 3.23

E5-2630 v2 6 2300 612 506 3.46

E5-2620 v2 6 2100 406 430 4.07

E5-2609 v2 4 2500 294 250 7.00

E5-2603 v2 4 1800 202 186 9.41

Ivy Bridge (E5 family of chips) is the most current commodity 2-socket x86 offering

Varies from 4 to 12 cores and $200 to $2600 per chip

x86 performance marketing messages are focused on 12-core E5-2697 v2 (top of the line)

POWER8 performance is 2x versus E5-2697 v2 (top of the line) across multiple benchmarks (i.e. SPECjEnterprise2010, SAP 2-tier, etc.)

2.5x to 4x the performance of the x86 sweet spot (based on SPECint2006_rate)

Up to 9x compared to the bottom of the line Intel chip (based on SPECint2006_rate)

POWER8 delivers up to 4x the performance of x86 “sweet spot”Don’t allow x86 vendors to mix top of the line performance claims with low-performance chip & system pricing

http://www.spec.org/

http://www.intc.com/pricelist.cfm

(1) Performance data based on best published peak SPECint_rate2006 results. All results can be found at www.spec.org(2) Intel chip prices based on Intel published list prices. All chips and chip prices can be found at http://www.intc.com/pricelist.cfm

POWER8 delivers up to 3.3x the performance of x86 “sweet spot”Don’t allow x86 vendors to mix top of the line performance claims with low-performance chip & system pricing

Intel Haswell-EPSPECint2006_rate

(2-socket)

Proc. # Cores Freq.List Price (kUnits)

ResultS824 vs. x86 SKU

E5-2699 v3 18 2300 4025 1400 1.25

E5-2698 v3 16 2300 3167 1270 1.38

E5-2697 v3 14 2600 2702 1230 1.42

E5-2695 v3 14 2300 2424 1120 1.56

E5-2690 v3 12 2600 2090 1110 1.58

E5-2680 v3 12 2500 1745 1070 1.64

E5-2670 v3 12 2300 1589 1000 1.75

E5-2660 v3 10 2600 1445 912 1.92

E5-2650 v3 10 2300 1166 850 2.06

E5-2640 v3 8 2600 939 728 2.40

E5-2630 v3 8 2400 667 691 2.53

E5-2620 v3 6 2400 417 526 3.33

E5-2609 v3 6 1900 309 317 5.52

E5-2603 v3 6 1600 213 276 6.34

Haswell (E5 family of chips) is the most current commodity 2-socket x86 offering Varies from 6 to 18 cores and $200 to

$4,000 per chip x86 performance marketing messages

are focused on 18-core E5-2697 v2 (top of the line)

POWER8 performance is 2x versus E5-2699 v3 (top of the line) across multiple benchmarks (i.e. SPECjEnterprise2010, SAP 2-tier, SPECjbb2013, etc.)

2.4x to 3.3x the performance of the x86 sweet spot (based on SPECint2006_rate)

Over 6x compared to the bottom of the line Intel chip (based on SPECint2006_rate)


http://www.spec.org/

http://www.intc.com/pricelist.cfm

ScalabilityControl and Contain Your Spend

Buy what you need, when you need it

28

POWER Systems Flexibility Advantage withOracle Database

Size individual database LPARs to match specific CPU, I/O and memory needs

Scale from very small to very large LPARs and Oracle instances

Create independent security domains Deploy varying versions of Oracle

Isolate critical databases in different LPARs Isolate database by department or other Mix test and production on the same frame Mix application and database on the same

machine

AIXWPARs

DB

DB

App

DB

OS

DB

OS

App

OS

DB

OS OS

DBDB

OS

RAC

OS

RAC

OS

RAC

OS

RAC

PowerVM Hypervisor PowerVM Hypervisor

= IBM Advantages

Implement and deploy an appropriate mix of RAC and non-RAC Oracle database instances as well as application instances

29

Market Positioning of PowerVM and VMware

PowerVM

Virtualization and partitioning since 2001

Microcode/eeprom hardware implementation

Enterprise mission critical platform that hosts multi-tier consolidation

I/O subsystem is managed in separate hardware and software partitions

Scalability: from 1 thread to 1024 simultaneous treads ( 32 chips / 256 cores)

Integrated in hardware and additional enhancements priced as features

AIX, i/OS, RedHat, SuSe, but no Windows support

VMware

PC virtualization since 1999, server virtualization since 2001

Proven 20 to 1 compression ration of real to virtual servers (of the right type)

Standard for x86 virtualization - Infrastructure and small applications

Pure software virtualization with 15% to 30% overhead in production workloads

Scalability: v4 1 to 8 threads ( .4 chips / 4 cores) v5 1 to 32 threads ( 1.6 chips /16 cores)

No partitioning technology

I/O is handled within virtual machines

No Oracle technical support or sub-capacity pricing support

Expensive and now priced on virtual resources

Oracle Certification For VMware and KVM• Running Oracle in a VMware ESX cluster you must license ALL of the cores in the cluster• Oracle DOES NOT recognise VMware as "hard partitioning"• http://blogs.gartner.com/chris-wolf/2010/11/10/oracle-broadens-x86-virtualisation

-support-but-work-remains/• Running Oracle in a VMware ESX cluster is not certified. If support is required for unknown problems then

you must recreate the problem without VMware installed view Oracle Metalink document 249212.1

• Red Hat Enterprise Linux 5 integrates Kernel-based Virtual Machine (KVM) and ships Xen as the default hypervisor, so they are supported by Oracle under the Oracle Linux support program. However, Oracle does not support Oracle products on RHEL's KVM/Xen.

• http://www.oracle.com/us/technologies/027617.pdf

http://blogs.gartner.com/chris-wolf/2010/11/10/oracle-broadens-x86-virtualization-support-but-work-remains/

http://blogs.gartner.com/chris-wolf/2010/11/10/oracle-broadens-x86-virtualization-support-but-work-remains/

http://www.oracle.com/us/technologies/027617.pdf

Security No one wants to be (a) Target

32

33

IT security is key to any your business as security breeches can cause irreparable damage

• The cost to your business can be in many forms

– Loss of customers– Customer dissatisfaction– Bad press– Ruined reputation– Government or regulatory investigations– Large fines or penalties– Lower productivity– Lower stock price

• Consider the monetary cost of a data breach1

– 204 USD per compromised customer record in 2009 for companies in the United States

– 6.75M USD per-incident total cost in 2009 for companies in the United States

Source: 1. Ponemon Study Shows the Cost of a Data Breach Continues to Increase; Ponemon Institute; November 2011. 2. Chronology of Data Breaches, Security Breaches 2005 – Present; Privacy Rights Clearinghouse, http://www.privacyrights.org; November 2011.

Target apologizes for data breach, retailers embrace security upgrade

Target disclosed on December 19 that it was victim to one of the biggest credit card breaches on record. It said it ran for 19 days in the busy holiday shopping season through December 15.

Stores and card processing companies have reported a steady stream of security breaches for years without a major backlash from consumers, such as those disclosed by TJX Cos in 2007 and by Heartland Payment Systems Inc in 2009.But the latest thefts could mark a watershed moment for security standards as calls grow for changes in the protection of consumer information.

By Ross Kerber, Phil Wahba and Jim Finkle January 13, 2014 12:26 PM http://news.yahoo.com/target-apologizes-data-breach-retailers-embrace-security-upgrade-172658839--sector.html

34

Home Depot confirms security breach following Target data theft

“Home Depot Inc confirmed on Monday its payment security systems have been breached, a data theft analysts warn could rival Target Corp's massive breach last year.Home Depot said the data theft could impact its customers in stores across the United States and Canada…”http://www.reuters.com/article/2014/09/09/us-usa-home-depot-databreach-idUSKBN0H327E20140909

http://www.privacyrights.org/

Security of critical workload (SAP) deployments on Power is beyond reproach

• SAP on Power versus competitive SAP deployments study with over 54,150 clients analyzed

• The security for ERP systems, including SAP, can be very challenging – by nature, the mixture of application modules, user profiles, plug-in components and so on, provide many avenues for security breaches

Source: Business Impacts on SAP Deployments; Solitaire Interglobal Ltd (All rights reserved); January 2013.

0 reported security breeches with SAP and IBM DB2 or

Oracle DB on Power

35

Server virtualization security is critical for DB workloads since many are run in virtual environments

• The PowerVM hypervisor has never had

a reported security vulnerability and

provides the bullet-proof security that

customers demand for mission-critical

workloads• Dare to compare – search any

security tracking DB and compare

Power against x86

0 reported security breeches

on the PowerVM hypervisor

36

Cost Savings on Power

37

Source: Capacity based on IBM Sizing of typical SPECint_rate landscape and 3rd party analysis of system utilization.

Current Configuration 100 2-socket x86 servers Xeon X5690 processor 12 cores per server, 2

threads per core 3.46 GHz 1200 total cores VMWare

Ivy Bridge Configuration 69 2-socket x86 servers Xeon E5-2697 v2 processor 24 cores per server, 2 threads

per core 2.8 GHz 1656 total cores VMWare

1.5X

increased

throughput

Per-core SW Costs INCREASE 34%!!!

Current x86

Platform

(circa 2011)

POWER8 DELIVERS EQUAL CAPACITY with

1/3 of the Servers (LOWER MANAGEMENT COSTS)

1/3 of the Cores (LOWER SW COSTS)

1/3 of the Space (LOWER INFRASTRUCTURE COSTS)

ALL for <50% of the x86 HW TCA

POWER8

Platform

(2014)

New x86

Platform

(2014)

Per-core SW Costs DECREASE up to 54%!!!

Power S822L Configuration 23 2-socket POWER8 servers POWER8 processor 24 cores per server, 8 threads

per core 3.0 GHz 552 total cores

Intel: Reduce HW infrastructure in a virtualized environment but increase per-core SW costs…POWER8: Reduce HW infrastructure EVEN MORE and REDUCE per-core SW costs up to 54%

http://www.sil-usa.com/pub_papers/QR2012A672.pdf

Power S822L

TCA/TCO is for 34 servers, 816 cores

3x better virtualized throughput vs. an HP 2 socket Ivy Bridge

2S, 24 cores each POWER8, 3.0GHz PowerVM

HP DL 380 G8


100 HP servers needed for ~ equal virtualized throughput of 34 Power S822L

2S, 24 cores each Ivy Bridge, 2.7GHz VMware vSphere Ent

58%lower total HW TCA vs.

Ivy Bridge w/ VMware

POWER8 delivers lower cloud infrastructure costs Thirty-four 2-socket Power S822L servers do the job of 100 2-socket x86 (HP DL380) servers running equal virtualized capacity

39

$1,307,776

$2,951,257

$-

$1,000,000

$2,000,000

$3,000,000

$4,000,000

S822L/24c HP DL380p IvB(2s)

Total HW TCA

HW TCA

Source: Capacity based on IBM Sizing of typical SPECint_rate landscape and 3rd party analysis of system utilization. Pricing from www.hp.com.

66% Less Systems & Cores Lower SW License Fees

Reduced Management CostsReduced Floor Space

Power S824

TCA/TCO is for one servers, 24 cores

4x better virtualized throughput vs. an HP 2 socket Ivy Bridge

2S, 24 cores each POWER8, 3.5GHz PowerVM Oracle EE Database

HP DL 380 G8

TCA/TCO is for four servers, 96 cores

4 HP servers needed for ~ equal ERP throughput of 1 Power S824

2S, 24 cores each Ivy Bridge, 2.7GHz VMware vSphere Ent. Oracle EE Database

47% lower total TCA

vs. HP Ivy Bridge w/

VMware

$531,0993-yr. TCO savings

vs. HP Ivy Bridge w/ VMware

Source: Capacity based on IBM Sizing of typical ERP landscape and IBM estimates of system utilization. Pricing from www.hp.com

$85,431

$342,000

$118,050

$684,000

$-

$200,000

$400,000

$600,000

$800,000

$1,000,000

S824/24c HP DL380p IvB(2s)

Total TCA

SW TCA

HW TCA

$586,911

$1,118,010

$-$200,000

$400,000$600,000

$800,000$1,000,000$1,200,000

S824/24c HP DL380p IvB(2s)

3-yr TCO

3-yr TCO

Save $500K in your Oracle DB ERP environment with POWER8 vs x86 One 2-socket Power S824 server does the job of four 2-socket HP DL380p Gen8 servers running equal virtualized ERP capacity with a Oracle EE Database

40

41

Power S822



HP DL 380 G8

TCA/TCO is for ten servers, 240 cores

2.5x HP servers needed for ~ equal OLTP throughput of Power S822

2S, 24 cores each Ivy Bridge, 2.7GHz VMware vSphere Ent. Oracle EE Database

41

Save over $1.3M in your Oracle DB OLTP environment with POWER8 vs x86 Four 2-socket Power S822 servers do the job of Ten 4-socket HP DL380 servers running equal virtualized OLTP capacity with Oracle EE Database

60% Lower HW TCA

51%Lower Total TCA

$1,329,9543-yr. TCO savings

vs. HP Ivy Bridge with

VMWare

POWER8 delivers equal ERP performance at 51% lower total TCA than Intel 8-socket IvBTwo 8-socket Power E870 servers with Oracle DB ERP compared against Three 8-socket x86 E7-8890 v2 servers shows lower TCA and TCO while delivering equal performance

Source: Capacity based on IBM Sizing of typical ERP landscape and IBM estimates of system utilization. This is an IBM sizing designed to replicate a typical IBM customer workload used in the marketplace. The results are calculated and not an actual customer environment. IBM's internal workload studies are not benchmark applications, nor are they based on any benchmark standard. As such, customer applications, differences in the stack deployed, and other systems variations or conditions may produce different results and may vary based on actual configuration, applications, specific queries and other variables in a production environment. Prices, where applicable, are based on published US list prices for both IBM and competitor. Pricing from www.oracle.com

51% Lower Total TCA

$2.3M3-yr. TCO savings

vs.. 8-socket x86 Servers

8-socket E7 Servers


8S/120 cores per server E7-8890 v2, 2.8GHz VMWare Oracle EE

$601,694

$912,000

$517,676

$2,565,000

$-

$1,000,000

$2,000,000

$3,000,000

$4,000,000

E870 / 64c / 4.02 HP DL980/IvB (8s)

Total TCA

SW TCA

HW TCA

$1,923,974

$4,256,276

$-

$2,000,000

$4,000,000

$6,000,000

E870 / 64c / 4.02 HP DL980/IvB (8s)

3-yr TCO

3-yr TCO

Power E870

TCA/TCO is for two servers, 128 cores

8S/64 cores per server POWER8, 4.02 GHz PowerVM Oracle EE

Power S822

TCA/TCO is for one servers, 20 cores


Intel 2s Haswell


2S, 24 cores each Haswell 2.3GHz VMware vSphere Ent. Oracle EE Database

68% lower total TCA

vs. Intel Haswell 2s

w/ VMware

$483,0033-yr. TCO savings

vs. Intel Haswell 2s

w/ VMware

Source: Capacity based on IBM Sizing of typical ERP landscape and IBM estimates of system utilization. Pricing from www.hp.com

Save $480K in your Oracle DB ERP environment with POWER8 vs x86 Haswell One 2-socket Power S822 server does the job of two 2-socket Intel Haswell servers running equal virtualized ERP capacity with a Oracle EE Database

43

S822/20c/3.42 HP DL380p/HAS (2s) $-

$200,000

$400,000

$600,000

$800,000

$1,000,000

$65,428 $59,355

$423,000

$761,400

Total TCA

SW TCA

S822/20c/3.42 HP DL380p/HAS (2s) $-

$200,000

$400,000

$600,000

$800,000

$1,000,000

$1,200,000

$683,548

$1,166,551

3-yr TCO

3-yr ...

IBM and Business Partner Use ONLY

3 x 2-socket Haswell108 Total CoresVMWare40% utilizationWebsphere

1 x POWER8 S822L24 Total CoresPowerVM65% utilization (guaranteed)Websphere

Return to x86 Tactics

$38,464$115,920

$98,655

$521,640

$-

$200,000

$400,000

$600,000

$800,000

S822L/24c/3.02 HP DL380p/HAS (2s)

Total TCA

SW TCA

HW TCA

$209,752

$845,121

$-

$500,000

$1,000,000

S822L/24c/3.02 HP DL380p/HAS (2s)

3-yr TCO

Now let’s look at Intel’s latest 2-socket Haswell-EP boxes running a java workload (SPECjbb2013)

1x2-socket Power S822L outperforms 3x2-socket x86 with

LOWER HW/SW TCA and TOTAL TCO

IBM S822L has:Better performance

61% Lower HW TCA75% Lower Total TCA

Source: Capacity based on IBM Sizing of typical SPECjbb2013 and 3rd party analysis of system utilization. . This is an IBM sizing designed to replicate a typical IBM customer workload used in the marketplace. The results are calculated and not an actual customer environment. IBM's internal workload studies are not benchmark applications, nor are they based on any benchmark standard. As such, customer applications, differences in the stack deployed, and other systems variations or conditions may produce different results and may vary based on actual configuration, applications, specific queries and other variables in a production environment. Prices, where applicable, are based on published US list prices for both IBM and competitorPrices, where applicable, are based on published US list prices for both IBM and competitor.

Production US DollarSelectable

Power System Competitive SystemOverwritable

Workload ERP ResultsDB Licensing OracleEE OracleEEFraction of DB Cores 100% 100%Virtualization PowerVM NoneClustering None NoneOracle RAC RAC RAC

DB SW Contract Discount 70% 70%

Based on DB Licensing - overwritable

Architecture Power Intel/x86

System S822/20c HP DL380p/IvB (2s)# Systems 2 10Frequency (MHz) 3400 2700Mem Option Mem per SystemGB per System 320 384

HW Discount 0% 35%

Power system includes discounted HW, SW (OS/PVM), Maint.

Sustained Utilization 65% 20%

System Cores 40 240 500%Database License Cores 40 240Memory 640 3840

Performance 19875 18432

HW TCA $ 92456 $ 140719 52%advantage

SW TCA $ 846000 $ 2538000 200%advantage

TOTAL TCA $ 938456 $ 2678719 185%advantage

SW TCO $ 1218240 $ 3654720 200%advantage

TOTAL 3-yr TCO $ 1328696 $ 3832939 188%advantage

3-Year Savings $ 2504243

Power requires fewer cores (and DB cores), as well as overall memory due to PowerVM utilization, compared to this example of running an ERP workload on a physical x86 server

Power/AIX vs Linux on Intel TCA/TCO 2s Ivy Bridge Servers versus S822 for ERP Workloads

45

Power System Competitive SystemOverwritable

Workload ERP Results

DB Licensing OracleEE OracleEE

Fraction of DB Cores 100% 100%

Virtualization PowerVM VMWare

Clustering None None

Oracle RAC RAC RAC

DB SW Contract Discount 70% 70%

Based on DB Licensing - overwritable

Architecture Power Intel/x86

System S822/20c HP DL380p/IvB (2s)

# Systems 3 10

Frequency (MHz) 3400 2700

Mem Option Mem per System

GB per System 320 384

HW Discount 0% 35%

Power system includes discounted HW, SW (OS/PVM), Maint.

Sustained Utilization 65% 40%

System Cores 60 240 300%Database License Cores 60 240

Memory 960 3840

Performance 29812 31334

HW TCA $ 138684 $ 200719 45%advantage

SW TCA $ 1269000 $ 2538000 100%advantage

TOTAL TCA $ 1407684 $ 2738719 95%advantage

SW TCO $ 1827360 $ 3654720 100%advantage

TOTAL 3-yr TCO $ 1993044 $ 3892939 95%advantage

3-Year Savings $ 1899895

Lintel - What happens to TCA/TCOWhen VMWare is used for the x86 systems?

While adding VMWare will assist in driving up the utilization level on x86 systems, it does not drive them to the level that PVM can, and adds cost to the x86 TCA

And look at the SW TCA benefit we bring over x86!

46

IBM POWER7/8® Systems Linux on Intel

HW/OS RAS Features

IBM Power with AIX offers the least amount of downtime per year

Fastest Patch Time – 11 minutes

Less than 30 minutes of downtime/server/year

Linux on Intel requires up to 4.5x more downtime

27 minutes to patch each Lintel server

Nearly 1.5 hours of downtime/server/year

PerformanceIndustry-leading performance and benchmarks from 2-16 sockets per server

Delays in roadmap execution have resulted in lagging performance

Planned DowntimeLPM can be used to ensure the workloads and users are not disrupted during planned maintenance

Expensive software investments are required to support planned downtime, including RAC and server cluster software

Virtualization

Allows optimization of Oracle licenses to reduce number required by driving sustained utilization levels higher through dynamic resource reallocation

All features are certified and supported for use with Oracle

Fragmented virtualization options that rely on OS, and are limited in functionality due to OS restrictions – no dynamic reallocation of resources

VMWare/KVM/Xen not certified or supported for use with Oracle

SecurityZERO Critical or High Security Threats – Most secure platform/OS in industry

Climbing number of security threats reported – Linux is the OS where 31% of all security threats reported

Roadmap

History of a clear roadmap with timely execution

Intel has slipped in their execution of roadmap, leaving gaps in addressing increasing workload demands. No 8 socket EX for SB

Top Reasons IBM POWER Systems are Better for Critical Workloads Compared with Linux on Intel

All of these reasons add up to significant TCA/TCO savings for YOU when you move your Oracle DB workloads to IBM Power Systems

Avg 150/mo Reported Security Vulnerabilities in RHEL 2013

Conclusion: The Other Systems are NOT as Good

• POWER Systems Deliver a Lower TCA and TCO Compared againstLinux on Intel

• Power Delivers– Better Investment Protection for both HW and SW– Better RAS features at a lower Cost– Better Performance at Higher System Utilization/Performance Levels

Which we Guarantee – POWER8 65% Sustained Utilization Guarantee with No Degradation in Performance

– Better Scalability for intended and unexpected growth – buy what you need, when you need it

– Better SecurityDon’t be (a) Target

Don’t waste resources (time and money) patching endless systems– And cost savings with POWER IS the icing on the cake

48

Power Systems have a Lower TCA Why Compromise?

Thank YouThank You

Merci

Grazie

Gracias

Obrigado

Danke

Japanese

Hebrew

English

French

Russian

German

Italian

Spanish

Brazilian Portuguese

Arabic

Traditional Chinese

Simplified Chinese

Hindi

Tamil Korean

Thai

TesekkurlerTurkish

Documents

© 2014 IBM Corporation The Economics and Value of Power Systems versus Linux on Intel