20 June 2002 - News Item - Myrinet Clusters in the June 2002 Top500 List

News Item
20 June 2002

Myrinet Clusters in the
June 2002 TOP500 List

Myrinet clusters made an excellent showing in the June 2002 TOP500 list. The list includes 21 Myrinet clusters, led by the University of Heidelberg HELICS cluster, ranking #35. An additional 145 HP Superdome/Hyperplex systems, classified by the TOP500 authors as "MPPs," use Myrinet for communication between the Superdome SMP hosts. Thus, one-third of the supercomputers in this June 2002 TOP500 list use Myrinet technology.

The TOP500 list, published twice a year at the ISC conference in June and at the SC conference in November, ranks supercomputers worldwide according to their performance on the LINPACK benchmark. This 19th TOP500 list was published on 20 June at the International Supercomputer Conference (ISC2002) in Heidelberg, Germany.

Two of the scientists who publish the TOP500 list, Prof.Dr. Hans Meuer (University of Mannheim) and Prof. Jack Dongarra (University of Tennesee, Knoxville), gave invited presentations at the 2002 Myrinet Users Group conference. Their slide presentations can be downloaded from this MUG-2002 page. Prof.Dr. Meuer's presentation was specifically about the TOP500 project.

The TOP500 list nearly defines what is meant by a "supercomputer," and reveals interesting trends throughout the last decade. Clusters started to appear in the TOP500 list in 1997. The Berkeley NOW (Network of Workstations), a Myrinet cluster of 100 UltraSPARC-I computers, ranked #479 in the June 1997 TOP500 list with a LINPACK performance of 10.140 Gflops. According to the classification used by the TOP500 authors, there are now 80 clusters (16%) in the June 2002 list, up from 43 clusters (8.6%) in the November 2001 list, and up from from 33 clusters (6.6%) in the June 2001 list. The "cluster" classification is somewhat arbitrary. For example, it would not be obvious to a computer architect why an IBM SP would be classified as a (distributed-memory) MPP rather than as a (distributed-memory) cluster.

In any case, clusters are clearly the fastest growing category of supercomputers in the TOP500 list.

Here are some of the notable Myrinet clusters in the June 2002 TOP500 list:

#35
825 Gflops The HELICS cluster at the University of Heidelberg, Germany, Interdisciplinary Center for Scientific Computing (IWR) is the top Myrinet cluster in the June 2002 TOP500 list. This cluster of 256 dual 1.4GHz Athlon hosts (512 processors) was installed in March 2002, and is thus making its first appearance in the TOP500 list. This Linux supercluster was strikingly less expensive -- reported to have cost ~$1.15M -- than any of the higher ranked supercomputers. For more information, the web site for the HELICS cluster, http://helics.iwr.uni-heidelberg.de/, is interesting reading, and Dr. Wolfgang Hafemann, the leader of this project, gave an excellent talk about HELICS at the MUG-2002 conference.

#47
716.1 Gflops The Presto III cluster at TITECH (Tokyo Institute of Technology) Global Scientific Information and Computing Center is a self-made Myrinet/Linux cluster with 240 dual 1.2 GHz Athlon hosts (480 processors). This page by the "cluster team" is available in Japanese or English.

#50
706.7 Gflops
The CPlant/Ross cluster at Sandia National Laboratory. This self-made Myrinet cluster contains 1,369 single-processor Alpha hosts, and runs Linux. The Ross cluster was installed in 2001, and ranked #30 in November 2001 list. Note that 20 new TOP500 systems faster than the Ross cluster entered the TOP500 list in the past 7 months!

CPlant has grown during the 1998-2001 period. CPlant ranked #97 in the November 1998 TOP500 list at 54.24 Gflops achieved by 150 Alphas, and ranked #44 in the November 1999 TOP500 list at 232.6 Gflops with 580 Alphas.

#53
677.9 Gflops The Titan cluster at NCSA is a Myrinet/Linux cluster of 160 IBM dual 800MHz Itanium-1 hosts (320 processors). Titan, integrated by IBM, was installed in 2001, and first appeared in the November 2001 TOP500 list ranking #34.

#56
654 Gflops Magi cluster at the Tsukuba, Japan, Advanced Computing Center is a Myrinet/Linux cluster of 520 NEC 933MHz Pentium-III hosts (1040 processors), integrated by NEC, installed in 2001.

#57
618.3 Gflops
SCore IIIe cluster at the Real World Computing Partnership, Tsukuba, Japan, is a Myrinet/Linux cluster of 512 NEC 933MHz Pentium-III hosts (1024 processors), self-made, installed in 2001.

The two clusters above both run the SCore software, developed at RWCP.

#58
594 Gflops
The Platinum Netfinity cluster at NCSA is a Myrinet/Linux cluster of 512 IBM dual 1GHz Pentium-III hosts (1024 processors), installed by IBM in 2001.

To see the benefits of Myrinet versus Ethernet for the relatively tightly coupled LINPACK computation, compare Platinum with the otherwise identical ethernet clusters ranking #169 - #172. When the interprocess communication is limited by ethernet, these otherwide identical ethernet clusters achieved only 289 Gflops, less than half the performance of Platinum, a Myrinet cluster.

#64
564 Gflops The Joint Supercomputer Center in Moscow built this Myrinet/Linux cluster of 384 dual 667MHz Alpha EV-67s (768 processors), the most powerful computer in Russia.

#79
526 Gflops A visualization cluster at Sandia National Laboratory, but with plenty of compute power as well! Dell supplied this cluster of 128 dual 2.0 GHz Pentium-4 Xeon Precision 530 hosts (256 processors).

#96
470.9 Gflops
A SuperDome 750MHz / HyperFabric cluster at HP in Richardson, Texas.

HP HyperFabric is Myrinet, developed and produced for HP by Myricom, but with software developed by HP. This TOP500 list includes 145 SuperDome/HyperFabric or SuperDome/Hyperplex systems. See this news story for additional background.

#106
442.7 Gflops The ACL-580 cluster at the Forecast Systems Laboratory, NOAA, Boulder, Colorado. HPTi was the integrator for this Myrinet/Linux cluster of 580 Alpha processors. This article is a good description of the applications of the original FSL cluster, which is currently being expanded.

#107
442.5 Gflops The Prairiefire cluster at the University of Nebraska, Lincoln. Atipa was the integrator for this Myrinet/Linux cluster of 128 dual 1.4GHz Athlon hosts.

#176
285 Gflops The Huinalu cluster at the Maui High-Performance Computing Center. IBM was the integrator for this Myrinet/Linux cluster of 260 dual 933MHz Pentium-III hosts. "Huinalu" is Hawaiian for "a group of waves."

#229
237 Gflops The Los Lobos cluster at the University of New Mexico. IBM was the integrator for this Myrinet/Linux cluster of 256 dual Pentium-III hosts. Los Lobos was installed in 2000, and uses second-generation (1.28+1.28 Gb/s) Myrinet components. Los Lobos ranked #80 in its first appearence in the TOP500 list in November 2000.

#238
232 Gflops The SARA (Stichting Academisch Rekencentrum Amsterdam) Netfinity cluster, a cluster of 200 IBM dual 1GHz Pentium-III hosts.

#243
226 Gflops The Ohio Supercomputer Center Itanium Cluster is a Myrinet/Linux cluster of 73 (64 as benchmarked) SGI dual 733MHz Itanium hosts (146 processors), produced for OSC by SGI.

The TOP500 authors also maintain a special site and list for clusters, "Clusters@TOP500," http://clusters.top500.org/.

20 June 2002

#35 825 Gflops	The HELICS cluster at the University of Heidelberg, Germany, Interdisciplinary Center for Scientific Computing (IWR) is the top Myrinet cluster in the June 2002 TOP500 list. This cluster of 256 dual 1.4GHz Athlon hosts (512 processors) was installed in March 2002, and is thus making its first appearance in the TOP500 list. This Linux supercluster was strikingly less expensive -- reported to have cost ~$1.15M -- than any of the higher ranked supercomputers. For more information, the web site for the HELICS cluster, http://helics.iwr.uni-heidelberg.de/, is interesting reading, and Dr. Wolfgang Hafemann, the leader of this project, gave an excellent talk about HELICS at the MUG-2002 conference.
#47 716.1 Gflops	The Presto III cluster at TITECH (Tokyo Institute of Technology) Global Scientific Information and Computing Center is a self-made Myrinet/Linux cluster with 240 dual 1.2 GHz Athlon hosts (480 processors). This page by the "cluster team" is available in Japanese or English.
#50 706.7 Gflops	The CPlant/Ross cluster at Sandia National Laboratory. This self-made Myrinet cluster contains 1,369 single-processor Alpha hosts, and runs Linux. The Ross cluster was installed in 2001, and ranked #30 in November 2001 list. Note that 20 new TOP500 systems faster than the Ross cluster entered the TOP500 list in the past 7 months! CPlant has grown during the 1998-2001 period. CPlant ranked #97 in the November 1998 TOP500 list at 54.24 Gflops achieved by 150 Alphas, and ranked #44 in the November 1999 TOP500 list at 232.6 Gflops with 580 Alphas.
#53 677.9 Gflops	The Titan cluster at NCSA is a Myrinet/Linux cluster of 160 IBM dual 800MHz Itanium-1 hosts (320 processors). Titan, integrated by IBM, was installed in 2001, and first appeared in the November 2001 TOP500 list ranking #34.
#56 654 Gflops	Magi cluster at the Tsukuba, Japan, Advanced Computing Center is a Myrinet/Linux cluster of 520 NEC 933MHz Pentium-III hosts (1040 processors), integrated by NEC, installed in 2001.
#57 618.3 Gflops	SCore IIIe cluster at the Real World Computing Partnership, Tsukuba, Japan, is a Myrinet/Linux cluster of 512 NEC 933MHz Pentium-III hosts (1024 processors), self-made, installed in 2001. The two clusters above both run the SCore software, developed at RWCP.
#58 594 Gflops	The Platinum Netfinity cluster at NCSA is a Myrinet/Linux cluster of 512 IBM dual 1GHz Pentium-III hosts (1024 processors), installed by IBM in 2001. To see the benefits of Myrinet versus Ethernet for the relatively tightly coupled LINPACK computation, compare Platinum with the otherwise identical ethernet clusters ranking #169 - #172. When the interprocess communication is limited by ethernet, these otherwide identical ethernet clusters achieved only 289 Gflops, less than half the performance of Platinum, a Myrinet cluster.
#64 564 Gflops	The Joint Supercomputer Center in Moscow built this Myrinet/Linux cluster of 384 dual 667MHz Alpha EV-67s (768 processors), the most powerful computer in Russia.
#79 526 Gflops	A visualization cluster at Sandia National Laboratory, but with plenty of compute power as well! Dell supplied this cluster of 128 dual 2.0 GHz Pentium-4 Xeon Precision 530 hosts (256 processors).
#96 470.9 Gflops	A SuperDome 750MHz / HyperFabric cluster at HP in Richardson, Texas. HP HyperFabric is Myrinet, developed and produced for HP by Myricom, but with software developed by HP. This TOP500 list includes 145 SuperDome/HyperFabric or SuperDome/Hyperplex systems. See this news story for additional background.
#106 442.7 Gflops	The ACL-580 cluster at the Forecast Systems Laboratory, NOAA, Boulder, Colorado. HPTi was the integrator for this Myrinet/Linux cluster of 580 Alpha processors. This article is a good description of the applications of the original FSL cluster, which is currently being expanded.
#107 442.5 Gflops	The Prairiefire cluster at the University of Nebraska, Lincoln. Atipa was the integrator for this Myrinet/Linux cluster of 128 dual 1.4GHz Athlon hosts.
#176 285 Gflops	The Huinalu cluster at the Maui High-Performance Computing Center. IBM was the integrator for this Myrinet/Linux cluster of 260 dual 933MHz Pentium-III hosts. "Huinalu" is Hawaiian for "a group of waves."
#229 237 Gflops	The Los Lobos cluster at the University of New Mexico. IBM was the integrator for this Myrinet/Linux cluster of 256 dual Pentium-III hosts. Los Lobos was installed in 2000, and uses second-generation (1.28+1.28 Gb/s) Myrinet components. Los Lobos ranked #80 in its first appearence in the TOP500 list in November 2000.
#238 232 Gflops	The SARA (Stichting Academisch Rekencentrum Amsterdam) Netfinity cluster, a cluster of 200 IBM dual 1GHz Pentium-III hosts.
#243 226 Gflops	The Ohio Supercomputer Center Itanium Cluster is a Myrinet/Linux cluster of 73 (64 as benchmarked) SGI dual 733MHz Itanium hosts (146 processors), produced for OSC by SGI.