Skip to main content
 
developerworks > Community >  Dashboard > HPC Central Wiki > HPC Central > InfiniBand
developerWorks
Log In   View a printable version of the current page.
Overview New to Forums Wikis
InfiniBand
Added by gcorneau, last edited by robinhan on Jun 06, 2011  (view change)
Labels: 

Latest News

June 3, 2011

HPC software updates IBM POWER6 and IBM POWER7 support with InfiniBand

IBM* High Performance Computing (HPC) cluster software extends InfiniBand support to include updated
levels of the IBM AIX 6.1 TL6 SP5

o IBM Power 755 (8236-E8C) and IBM Power 750 (8233-E8B) interconnected with the GX Dual-port DDR Host Channel Adapter and
supported QLogic DDR InfiniBand switches
o IBM Power Systems* 575 (9125-F2A) interconnected with the Dual 2 Port 4x Host Channel Adapter and supported QLogic DDR InfiniBand switches
o Power Systems 520+ (8203-E4A) and Power Systems 550+ (8204-E8A) interconnected with the GX Dual-port DDR Host Channel Adapter and supported QLogic DDR InfiniBand switches
o Power Systems 520 (8203-E4A) and Power Systems 550 (8204-E8A) interconnected with the GX Dual-port DDR Host Channel Adapter or the GX Dual-port SDR Host Channel Adapter and supported
QLogic DDR InfiniBand switches
o IBM BladeCenter* JS22 (7998-61X) and JS23 (7778-63X) interconnected with 4X InfiniBand DDR Expansion Card for BladeCenter and supported QLogic DDR InfiniBand switches
o QLogic DDR InfiniBand switches supported: 7874-024, 7874-040, 7874-120, 7874-240

The following cluster software levels are supported running AIX 6.1 TL6 SP5:
o IBM Tivoli* Workload Scheduler LoadLeveler* for AIX, V4.1.1.4
o Parallel Environment for AIX, V5.2.2.3
o General Parallel File System* (GPFS*) for AIX, V3.4.0.6
o Parallel Engineering and Scientific Subroutine Library (Parallel ESSL) for AIX, V3.3.0.4
o Engineering and Scientific Subroutine Library (ESSL) for AIX, V5.1.0.2
o Extreme Cloud Administration Toolkit (xCAT) V2.6.0.0

Previously documented limitations still apply.

Please refer to the README for IBM clusters with the InfiniBand switch at the following URL for specific details:
http://www14.software.ibm.com/webapp/set2/sas/f/networkmanager/home.html

GPFS provides more extensive support for InfiniBand running AIX than what is supported in the HPC solution stack. Please refer to the GPFS FAQs for additional information:
http://publib.boulder.ibm.com/infocenter/clresctr/vxrx/topic/com.ibm.cluster.gpfs.doc/gpfs_faqs/gpfs_faqs.html

Please refer to the xCAT Release Notes for additional information:
http://sourceforge.net/apps/mediawiki/xcat/index.php?title=XCAT_2.6_Release_Notes

Service can be obtained from the IBM Electronic Fix Distribution service Web site:
http://www-933.ibm.com/support/fixcentral/?productGroup0=ibm/fcpower&productGroup1=ibm/ClusterSoftware&productGroup2=ibm/power/IBM

* Trademark or registered trademark of International Business Machines Corporation.
Linux is a trademark of Linus Torvalds in the United States, other countries, or both.
Other company, product, or service names may be trademarks or service marks of others.

December 10, 2010

HPC software updates IBM POWER7 support with InfiniBand running Linux

IBM* High Performance Computing (HPC) cluster software extends InfiniBand support to include Red Hat Enterprise Linux (RHEL) 6 on the IBM Power Systems* 755.  This update includes Huge Page support as well as updated HPC software levels.

The following cluster software levels are supported running RHEL 6:
o  IBM Tivoli* Workload Scheduler LoadLeveler* for Linux, V4.1.1.1
o  Parallel Environment for Linux, V5.2.2-1
o  GPFS* for Linux, V3.3.0-9 and V3.4.0-2
o  Parallel ESSL for Linux on POWER, V3.3.3-0
o  ESSL for Linux on POWER, V5.1.0-0
o  xCAT 2.5.1

Note:  Select cluster licensed programs listed above require a modification level update and fix level.

Large (Huge) Page support:
o Huge Page support is available with the release of RHEL 6.
o With kernel level 2.6.32.12-0.7.1.1609.0, SUSE Linux Enterprise Server 11 SP1 continues to support Huge Page.

The following RHEL 6 server limitations apply:
o  IBM System x InfiniBand User Space is supported at 8 node scaling
o  IBM Power Systems 755 is supported on RHEL6, no other POWER models are supported with InfiniBand
o  Power 755 scaling is limited to 8 nodes

Please refer to the README for IBM clusters with the InfiniBand switch at the following URL for specific details:
http://www14.software.ibm.com/webapp/set2/sas/f/networkmanager/home.html

Please refer to the GPFS FAQs for additional information:
http://publib.boulder.ibm.com/infocenter/clresctr/vxrx/topic/com.ibm.cluster.gpfs.doc/gpfs_faqs/gpfs_faqs.html

Please refer to the xCAT Release Notes for additional information:
https://sourceforge.net/apps/mediawiki/xcat/index.php?title=XCAT_2.5.1_Release_Notes

Service can be obtained from the IBM Electronic Fix Distribution service Web site:  
http://www-933.ibm.com/support/fixcentral/?productGroup0=ibm/fcpower&productGroup1=ibm/ClusterSoftware&productGroup2=ibm/power/IBM

* Trademark or registered trademark of International Business Machines Corporation.
Linux is a trademark of Linus Torvalds in the United States, other countries, or both.
Other company, product, or service names may be trademarks or service marks of others.

August 9, 2010

 HPC software updates IBM POWER6 & IBM POWER7 support with InfiniBand running AIX & Linux

IBM* High Performance Computing (HPC) cluster software extends InfiniBand support to include updated levels of the IBM AIX and Linux operating systems as well as updated HPC software levels.  Additionally, support on the IBM BladeCenter JS23 is now provided.   

IBM AIX* 5.3 Technology Level (TL) 12 Service Pack (SP) 1:
o  IBM Power Systems* 575 (9125-F2A) interconnected with the Dual 2 Port 4x Host Channel Adapter and supported QLogic DDR InfiniBand switches
o  Power Systems 520+ (8203-E4A) and Power Systems 550+ (8204-E8A) interconnected with the GX Dual-port DDR Host Channel Adapter and supported QLogic DDR InfiniBand switches
o  Power Systems 520 (8203-E4A) and Power Systems 550 (8204-E8A) interconnected with the GX Dual-port DDR Host Channel Adapter or the GX Dual-port SDR Host Channel Adapter and supported
QLogic DDR InfiniBand switches
o  IBM BladeCenter* JS22 (7998-61X) and JS23 (7778-63X) interconnected with 4X InfiniBand DDR Expansion Card for BladeCenter and supported QLogic DDR InfiniBand switches
o  QLogic DDR InfiniBand switches supported: 7874-024, 7874-040, 7874-120, 7874-240

IBM AIX 6.1 TL5 SP1:
o  IBM Power 755 (8236-E8C) interconnected with the GX Dual-port DDR Host Channel Adapter and supported QLogic DDR InfiniBand switches
o  IBM Power Systems* 575 (9125-F2A) interconnected with the Dual 2 Port 4x Host Channel Adapter and supported QLogic DDR InfiniBand switches
o  Power Systems 520+ (8203-E4A) and Power Systems 550+ (8204-E8A) interconnected with the GX Dual-port DDR Host Channel Adapter and supported QLogic DDR InfiniBand switches
o  Power Systems 520 (8203-E4A) and Power Systems 550 (8204-E8A) interconnected with the GX Dual-port DDR Host Channel Adapter or the GX Dual-port SDR Host Channel Adapter and supported QLogic DDR InfiniBand switches
o  IBM BladeCenter* JS22 (7998-61X) and JS23 (7778-63X) interconnected with 4X InfiniBand DDR Expansion Card for BladeCenter and supported QLogic DDR InfiniBand switches
o  QLogic DDR InfiniBand switches supported: 7874-024, 7874-040, 7874-120, 7874-240

SUSE Linux Enterprise Server (SLES) 11 SP1:
o  IBM Power 755 (8236-E8C) interconnected with the GX Dual-port DDR Host Channel Adapter and supported QLogic DDR InfiniBand switches
o  IBM Power Systems* 575 (9125-F2A) interconnected with the Dual 2 Port 4x Host Channel Adapter and supported QLogic DDR InfiniBand switches
o  Power Systems 520+ (8203-E4A) and Power Systems 550+ (8204-E8A) interconnected with the GX Dual-port DDR Host Channel Adapter and supported QLogic DDR InfiniBand switches
o  Power Systems 520 (8203-E4A) and Power Systems 550 (8204-E8A) interconnected with the GX Dual-port DDR Host Channel Adapter or the GX Dual-port SDR Host Channel Adapter and supported QLogic DDR InfiniBand switches
o  IBM BladeCenter* JS22 (7998-61X) and JS23 (7778-63X) interconnected with 4X InfiniBand DDR Expansion Card for BladeCenter and supported QLogic DDR InfiniBand switches
o  QLogic DDR InfiniBand switches supported: 7874-024, 7874-040, 7874-120, 7874-240

The following cluster software levels are supported running AIX 5.3 TL12 SP1 or AIX 6.1 TL5 SP1:
o  IBM Tivoli* Workload Scheduler LoadLeveler* for AIX, V4.1.0.7  
o  Parallel Environment for AIX, V5.2.1.5
o  General Parallel File System* (GPFS*) for AIX, V3.3.0.7 and V3.4.0.1
o  Parallel Engineering and Scientific Subroutine Library (Parallel ESSL) for AIX, V3.3.0.2
o  Engineering and Scientific Subroutine Library (ESSL) for AIX, V5.1.0.0
o  Extreme Cloud Administration Toolkit (xCAT) 2.4.2

The following cluster software levels are supported running SLES 11 SP1:
o  IBM Tivoli Workload Scheduler LoadLeveler for Linux, V4.1.0.7
o  Parallel Environment for Linux, V5.2.1.5
o  GPFS for Linux, V3.3.0.7 and V3.4.0.1
o  Parallel ESSL for Linux, V3.3.2.0
o  ESSL for Linux, V4.4.1.1
o  xCAT 2.4.2

The following server limitations apply:
o  IBM Power 755 scaling is limited to 64 nodes
o  Power Systems 575, 520, 520+, 550, 550+ and BladeCenter JS22 and JS23 scaling is limited to 64 nodes
o  Power Systems 520, 520+ and 550, 550+ are supported on IPoIB  Protocol only
o  GPFS NSD server support dependent upon device drivers being qualified/supported on SLES11 SP1.
   Until your device driver is supported on SLES 11 SP1, NSD servers should remain at SLES 11.

Service fixes are required.  Specific APAR numbers are listed in the README for IBM clusters with the InfiniBand switch.  Please refer to the README at the following URL for specific details: http://www14.software.ibm.com/webapp/set2/sas/f/networkmanager/home.html

Please refer to the xCAT Release Notes for additional information:
https://sourceforge.net/apps/mediawiki/xcat/index.php?title=XCAT_2.4.2_Release_Notes

Service can be obtained from the IBM Electronic Fix Distribution service Web site:  
http://www-933.ibm.com/support/fixcentral/?productGroup0=ibm/fcpower&productGroup1=ibm/ClusterSoftware&productGroup2=ibm/power/IBM

* Trademark or registered trademark of International Business Machines Corporation.
Linux is a trademark of Linus Torvalds in the United States, other countries, or both.
Other company, product, or service names may be trademarks or service marks of others.

April 6, 2009

HPC extends clustering support for InfiniBand on IBM Power Systems with RHEL 5.3 and SLES10 SP2

IBM High Performance Computing (HPC) cluster software extends InfiniBand support to include:

  • IBM Power Systems 575 (9125-F2A) interconnected with the Dual 2 Port 4x Host Channel Adapter and supported QLogic DDR InfiniBand switches
  • Power Systems 520 (8203-E4A) and Power Systems 550 (8204-E8A) interconnected with the GX Dual-port DDR Host Channel Adapter or the GX Dual-port SDR Host Channel Adapter and supported QLogic DDR InfiniBand switches
  • IBM BladeCenter JS22 (7998-61X) interconnected with 4X InfiniBand DDR Expansion Card for BladeCenter and supported QLogic DDR InfiniBand switches
  • QLogic DDR InfiniBand switches, supported Machine Type Models: 7874-024, 7874-040, 7874-120, 7874-240

This support is provided running Red Hat Enterprise Linux (RHEL) 5.3 or SUSE Linux Enterprise Server (SLES) 10, Service Pack (SP) 2 with the OFED 1.3 update.

The following cluster software levels are supported:

  • Cluster Systems Management for Linux, V1.7.0-18
  • IBM Tivoli Workload Scheduler LoadLeveler for Linux, V3.5.0-4
  • Parallel Environment for Linux, V5.1.0-4
  • General Parallel File System (GPFS) for Linux, V3.2.1-9
  • Parallel Engineering and Scientific Subroutine Library for Linux, V3.3.1-0

The following server limitations apply:

  • On Power Systems 575, 550, 520 and BladeCenter JS22, scaling is limited to 64 nodes
  • Power Systems 520 and 550 are supported on IPoIB Protocol only
  • IP mode is restricted to 8 tasks per node if running mixed MPI and LAPI 32 bit applications

Notes: Only first in, first out (FIFO) mode is supported on Red Hat 5.3, Remote Direct Memory Access (RDMA) is not supported.

Please refer to the README at the following URL for specific details:
http://www14.software.ibm.com/webapp/set2/sas/f/networkmanager/home.html

Service can be obtained from the cluster support Web site:
http://www14.software.ibm.com/webapp/set2/sas/f/cluster/home.html

January 19, 2009

HPC extends clustering support for InfiniBand on IBM Power Systems with AIX

IBM High Performance Computing (HPC) cluster software extends InfiniBand support to include:

  • IBM Power Systems 575 (Machine Type Model 9125-F2A) interconnected with the Dual 2 Port 4x Host Channel Adapter and supported QLogic DDR InfiniBand switches
  • Power Systems 520 (Machine Type Model 8203-E4A) and Power Systems 550 (Machine Type Model 8204-E8A) interconnected with the GX Dual-port DDR Host Channel Adapter or the GX Dual-port SDR Host Channel Adapter and supported QLogic DDR InfiniBand switches
  • IBM BladeCenter JS22 (Machine Type Model 7998-61X) interconnected with 4X InfiniBand DDR Expansion Card for IBM BladeCenter and QLogic DDR InfiniBand switches
  • QLogic DDR InfiniBand switches, supported Machine Type Models: 7874-024, 7874-040, 7874-120, 7874-240

This support is provided running IBM AIX 5.3, Technology Level (TL) 9, Service Pack 1 and IBM AIX 6.1, TL 2, Service Pack 1

The following cluster software levels are supported:

  • Cluster Systems Management for AIX, V1.7.0.16
  • IBM Tivoli Workload Scheduler LoadLeveler for AIX, V3.5.0.1
  • Parallel Environment for AIX, V5.1.0.1
  • General Parallel File System (GPFS) for AIX, V3.2.1.7
  • Parallel Engineering and Scientific Subroutine Library for AIX, V3.3.0.2

The following server limitations apply:

  • On Power Systems 575, 550, 520 and BladeCenter JS22, scaling is limited to 64 nodes
  • Power Systems 520 and 550 are supported on IPoIB Protocol only
  • JS21 and JS22 are not supported with Checkpoint Restart on the AIX OS

Service fixes are required. Specific APAR numbers are listed in the README.

Please refer to the README at the following URL for specific details:
http://www14.software.ibm.com/webapp/set2/sas/f/networkmanager/home.html

Service can be obtained from the IBM Electronic Fix Distribution service Web site at:
http://www-933.ibm.com/eserver/support/fixes/fixcentral/main/pseries/aix

November 11, 2008

IBM InfiniBand Clusters with QLogic InfiniBand switches help address HPC and commercial clustering requirements

Link to the announcement letter here: 108-843

August 2, 2008

HPC extends clustering support for InfiniBand on IBM Power Systems with SLES10

IBM* High Performance Computing (HPC) cluster software extends InfiniBand support to include:

  • IBM Power Systems* 575 (Model 9125-F2A) interconnected with the Dual 2 Port 4x Host Channel Adapter and supported QLogic DDR InfiniBand switches
  • Power System 520 (Model 8203-E4A) and Power System 550 (Model 8204-E8A) interconnected with the GX Dual-port SDR Host Channel Adapter and supported QLogic DDR InfiniBand switches
  • Power System* 575 (Model 9118-575) and Power System* 550 (Model 9133-55A) supported on Cisco InfiniBand SDR switches
  • QLogic DDR InfiniBand switch Model 9024, 9040, 9080, 9120, 9240 supported

This support is provided running SUSE Linux Enterprise Server (SLES) 10, Service Pack 2 with the OFED 1.3 update.

The following cluster software levels are supported:

  • Cluster Systems Management for Linux, V1.7.0-13
  • IBM Tivoli* Workload Scheduler LoadLeveler* for Linux, V3.4.3-3
  • Parallel Environment for Linux, V4.3.2-3
  • General Parallel File System* (GPFS*) for Linux, V3.2.1-4
  • Parallel Engineering and Scientific Subroutine Library for Linux, V3.3.0-1

The following server limitations apply:

  • Power System 575, 550, and 520: scaling limited to 64 nodes
  • RDMA is not supported
  • IPoIB Connected Mode is not supported

Please refer to the README at the following URL for specific details:
http://www14.software.ibm.com/webapp/set2/sas/f/networkmanager/home.html

Service can be obtained from the cluster support Web site:
http://www14.software.ibm.com/webapp/set2/sas/f/cluster/home.html

July 24, 2008

HPC extends clustering support for InfiniBand to JS22 running SLES 10

IBM* High Performance Computing (HPC) cluster software extends InfiniBand support to include the IBM BladeCenter* JS22 (Model 7998-61X) interconnected with the PCI-e Dual-port DDR Host Channel Adapter and QLogic DDR InfiniBand switch Models 9024, 9040, 9080, 9120, 9240.

This support is provided running SUSE Linux Enterprise Server (SLES) 10, Service Pack 2, with the OFED 1.3 update and includes failover and recovery.

The following cluster software levels are supported:

  • Cluster Systems Management for Linux, V1.7.0-13
  • IBM Tivoli* Workload Scheduler LoadLeveler* for Linux, V3.4.3-3
  • Parallel Environment for Linux, V4.3.2-3
  • General Parallel File System* (GPFS*) for Linux, V3.2.1-4
  • Parallel Engineering and Scientific Subroutine Library for Linux, V3.3.0-1

The following blade server limitations apply:

  • scaling limited to 64 blades
  • I/O servers for clustering are limited to POWER6* servers.

Please refer to the README at the following URL for specific details:
http://www14.software.ibm.com/webapp/set2/sas/f/networkmanager/home.html

Service can be obtained from the Cluster Support Web site at:
http://www14.software.ibm.com/webapp/set2/sas/f/cluster/home.html

June 23, 2008

HPC extends clustering support for InfiniBand to JS22

IBM* High Performance Computing (HPC) cluster software extends InfiniBand support to include the IBM BladeCenter* JS22 (Model 7998-61X) interconnected with the PCI-e Dual-port DDR Host Channel Adapter and QLogic DDR InfiniBand switch Models 9024, 9040, 9080, 9120, 9240.

This support is provided running IBM AIX* Version 5.3, Technology Level (TL) 8 and includes user space pre-emption, RDMA, failover and recovery for multiple adapters and adapter affinity.

The following cluster software levels are supported:

  • AIX V5.3 TL 5300-08 with Service Pack 2 and APARs IZ24424, IZ24423, IZ24420
  • Cluster Systems Management for AIX, V1.7 with APAR IZ21571
  • IBM Tivoli* Workload Scheduler LoadLeveler* for AIX, V3.4.3 with APAR IZ21565
  • Parallel Environment for AIX, V4.3.2 with APAR IZ21566
  • General Parallel File System* (GPFS*) for AIX, V3.2.1.3 with APAR IZ22023
  • Parallel Engineering and Scientific Subroutine Library for AIX, V3.3.0.2

The following blade server limitations apply:

  • scaling limited to 64 nodes

The following capabilities are restricted for TWS LoadLeveler:

  • Checkpoint Restart of jobs

Please refer to the README at the following URL for specific details: http://www14.software.ibm.com/webapp/set2/sas/f/networkmanager/home.html

Service can be obtained from the IBM Electronic Fix Distribution service Web site at: http://www-03.ibm.com/servers/eserver/support/unixservers/aixfixes.html

  • Trademark or registered trademark of International Business Machines Corporation.
    Other company, product, and service names may be trademarks or service marks of others.]
Docs InfiniBand POWER5 Cross Reference (HPC Central Wiki)
Docs InfiniBand POWER6 Cross Reference (HPC Central Wiki)
Docs InfiniBand POWER7 Cross Reference (HPC Central Wiki)
Docs InfiniBand Switches (HPC Central Wiki)