WorldWide-English

» Contact HP

The Real Story about IBM system p5 servers

Large Enterprise Business


	»	Products
	»	Business & IT Services
	»	Solutions
	»	Technologies
	»	Partners
	»	Support & Drivers
	»	Business Technology
	»	Media & Library



	»	Support & Troubleshooting
	»	Software & Drivers

the facts you may not know about IBM p5 January 2007. When comparing enterprise systems it is important to understand the entire picture of features and functionality. Customers may not know that IBM’s p5 server design does not contain some key features, or that highly touted technologies come with limitations.

Before choosing an IBM system p5 server solution, please consider the following:

The Real Story home page

The Real Story about the IBM Mainframe

The Real Story about Integrity Servers

Fact 1: IBM system p5 servers do not support hard partitions with full electrical isolation.[1]

HP Hard partitions, nPartitions, offer a higher level of fault isolation than does a soft partition approach like IBM, LPARs.²

One cell-based HP Integrity server can be partitioned into separate fault isolated partitions. Lack of electrical isolation between IBM LPARs may cause some users to cluster mission-critical applications in partitions on a separate backup server.

Hard partitions, like HP nPartitions, offer electrical isolation and cell board granularity, allowing you to service one partition while others are online.³

Fact 2: IBM does not have a single-system (non-clustered) TPC-H result in the 10 terabyte database size.[4]

System

# servers

Processors

cores/threads

Type of Processor

$/QphH@ 10000GB

$QphH@ 10000GB

Total System Cost

Availability

HP Integrity Superdome

128/128

1.6 GHz Itanium2 18M

171,380

$38.98

$6,681,114

4/1/2007

IBM System p5 575

128

128/256

2.2 GHz POWER5+

180,108

$47.00

$8,467,124

8/30/2006

Best 10-TBTPC-H results from each vendor. Results as of 1/24/07. See: http://www.tpc.org/.

A single HP system delivered 95% of the performance of a cluster of 16 IBM systems at a total system cost that was $1.8 million less expensive.

HP delivered similar performance is a much more energy efficient configuration:

System	Servers	Maximum Rated Wattage
HP Integrity Superdome	1	24.4 kW⁵
IBM System p5 575	16⁶	55.5 kW⁷

In this benchmark HP delivered 95% of the performance with a system that is rated at less than ½ the power. This would indicate that HP delivered 2X the performance per watt (rated).

This benchmark raises the question of why did IBM need a cluster of single core POWER5+ systems instead of the dual core p5 595? Is the p5 595 not suitable for the most demanding TPC-H database size?⁸

Fact 3: The minimum charge for IBM’s on/off capacity on demand is 24 hrs (one processor day) regardless of whether the processor activation was for a much shorter period of time.[9]

When you pre-pay for 30 processor days with HP’s Temporary Instant Capacity (TiCAP), your active processor time is measured in much more granular 30 minute slices.¹⁰

Why pay for 24 hours when you need just one?

Fact 4: In order for the IBM’s eWLM to manage a system based on transaction response times the server application must be instrumented with ARM. [11]

ARM is an Open Group standard composed of a set of APIs that can be used to collect response time information for enterprise applications. Is your application ARM enabled?

IBM admits that if you run a lot of non-ARM instrumented applications the partition management benefits of EWLM are limited.¹²

Among other notable benefits, HP’s WLM can improve the response time for critical users and applications through effective use of workloads and Service Level Objectives (SLOs) without the need to ARM enable.

To learn more see: http://h20338.www2.hp.com/hpux11i/cache/328430-0-0-0-121.html.

Fact 5: IBM states that, “Dynamic LPAR memory operations are possible but ... NOT recommended frequently due to paging out performance hit".[13]

How useful is a feature that can’t be used frequently?

Fact 6: A single IBM Virtual I/O Server is a single point of failure and requires a dedicated LPAR increasing the virtualization overhead.[14]

The IBM Virtual I/O Server can be made redundant by running a second instance of it in another partition however, processor resources dedicated to the virtual I/O server add to the virtualization overhead. IBM recommends a dedicated processor allocated for high VIOS workload.¹⁵ Dedicated resources must be allocated to the virtual I/O server and cannot be made available for customer workloads.

Fact 7: IBM’s POWER virtualization technology is made up of individual pieces with restrictions that limit the amount of integration between them.

For example:

IBM Partition Load Manager (PLM) restrictions:

IBM’s PLM cannot manage the IBM Virtual I/O server (VIOS).¹⁶
The IBM PLM supports just a single Hardware Management Console (HMC). Failover to a back up HMC must be done manually.¹⁷
The IBM PLM requires the HMC and does not support systems managed by Integrated Virtualization Manager (IVM).¹⁸

IBM Integrated Virtualization Manager (IVM) restrictions:

The IVM and HMC are incompatible.¹⁹
IBM admits that, “Migration from an HMC-managed environment to an IVM environment is a difficult scenario.”²⁰
IBM also says. “There is no officially announced procedure to migrate between an HMC and an IVM environment.”²¹
The IVM is not supported on the p5 570, 575, 590 or 595.²²

Fact 8: System p5 limits operating system choices by not offering Windows®.

IBM p5 servers do not support Windows.

When you see the fact you can see that HP Integrity servers offer much more than a glitzy benchmark result delivered at the cost of many important customer features (detailed above).

To learn more about HP Integrity servers see:

http://www.hp.com/go/integrity

[1] IBM Whitepaper, “IBM p5: A Highly Available Design for Business-Critical Applications, “http://www-03.ibm.com/systems/p/hardware/whitepapers/power5_ras.pdf, page 29
[2] Illuminata report, http://www-1.ibm.com/servers/eserver/pseries/library/consult/illuminata_p5.pdf, pages 3 -4
[3] http://h71028.www7.hp.com/enterprise/cache/4282-0-0-0-121.aspx
[4] See http://www.tpc.org/, as of 24 Jan 2007
[5] HP Quick Specs http://h18000.www1.hp.com/products/quickspecs/11717_div/11717_div.HTML#Technical%20Specifications
[6] IBM details that a single p5 575 cluster node is 12 servers. 16 would require two racks. ftp://ftp.software.ibm.com/common/ssi/rep_sp/n/PSD00762USEN/PSD00762USEN.PDF
[7] full rack of 12 p575’s is rated to consume 41.6kW of power maximum. 16 would have a max power consumption of 55.5 kw. See: ftp://ftp.software.ibm.com/common/ssi/rep_sp/n/PSB01628USEN/PSB01628USEN.PDF
[8] As of 24 Jan 2007 IBM does have a single system p5 595 TPC-H in the smaller 3 TB TPC-H size. The TPC believes that comparisons of TPC-H results measured against different database sizes are misleading and discourages such comparisons. See; http://www.tpc.org/tpch/results/tpch_perf_results.asp
[9] IBM SystemI and System P Capacity on Demand
http://publib.boulder.ibm.com/infocenter/eserver/v1r3s/topic/ipha2/ipha2.pdf
[10] http://docs.hp.com/en/B9073-90081/B9073-90081.pdf
[11] IBM Redbook Enterprise Workload Manager V2.1, http://www.redbooks.ibm.com/redbooks/pdfs/sg246785.pdf, section 1.2.5
[12] IBM Redbook Enterprise Workload Manager V2.1, http://www.redbooks.ibm.com/redbooks/pdfs/sg246785.pdf, Chapter 1 page 19, “If you run a lot of non-ARM-instrumented applications in a 1-tier pSeries environment or if you run a very simple 3-tier architecture where all transactions run through the same three servers, the partition management benefits of EWLM are limited and PLM and/or AIX WLM might be more suitable.””
[13] IBM Presentation Sizing virtual micropartitions IBM System p, AIX 5L and Linux
Technical University ,course description at http://www-304.ibm.com/jct03001c/services/learning/ites.wss?pageType=page&c=a0010655
[14] IBM Redbook, “Advanced POWER Virtualization on IBM System p5” , January 2007, http://www.redbooks.ibm.com/redpieces/pdfs/sg247940.pdf
[15] IBM Redbook, “Advanced POWER Virtualization on IBM System p5” , January 2007, page 126 http://www.redbooks.ibm.com/redpieces/pdfs/sg247940.pdf
[16] IBM Redbook, “Advanced POWER Virtualization on IBM System p5” , January 2007, Section 6.5 http://www.redbooks.ibm.com/redpieces/pdfs/sg247940.pdf
[17] IBM Redbook, “Advanced POWER Virtualization on IBM System p5” , January 2007, Section 6.5 http://www.redbooks.ibm.com/redpieces/pdfs/sg247940.pdf
[18] IBM Redbook, “Advanced POWER Virtualization on IBM System p5” , January 2007, Section 6.5 http://www.redbooks.ibm.com/redpieces/pdfs/sg247940.pdf
[19] IBM Redbook “Integrated Virtualization Manager on IBM System p5”, Section 1.1.,“The internal design of the IVM requires that no HM should be connected to a working IVM system” http://www.redbooks.ibm.com/redpapers/pdfs/redp4061.pdf
[20] IBM Redbook “Integrated Virtualization Manager on IBM System p5”, Section 5.2.3
[21] IBM Redbook “Integrated Virtualization Manager on IBM System p5”, Section 5.2
[22] IBM Redbook “Integrated Virtualization Manager on IBM System p5”, Section 1.1. 1 http://www.redbooks.ibm.com/redpapers/pdfs/redp4061.pdf

Intel and Itanium are trademarks or registered trademarks of Intel Corporation or its subsidiaries in the United States and other countries. Microsoft and Windows are U.S. registered trademarks of Microsoft Corporation. UNIX is a registered trademark of The Open Group. Linux is a U.S. registered trademark of Linus Torvalds. TPC-H is a trademark of the Transaction Performance Processing Council.

Printable version


Privacy statement	Using this site means you accept its terms

© 2009 Hewlett-Packard Development Company, L.P.