SYSTEMS
HP announced enhancements to its HPTC cluster systems
- Written by: Writer
- Category: SYSTEMS
The new offerings broaden the price and performance options for HP's powerful HPTC clusters, providing customers with more choices as they move to standards-based cluster computing. The new benchmark results underscore HP's continued leadership in providing the industry's leading HPTC technology and products. Announcements include: • Full and newly expanded line of HP Integrity rx2600 and rx1600 server options available in scalable Linux-based HP XC6000 clusters, doubling the number of XC6000 server options • InfiniBand 4x for Integrity HP-UX clusters, which triples the HP-UX Integrity clusters interconnect bandwidth while substantially improving interconnect latency • High Performance Cluster LC Series is expanding with the LC1000 Series, based on the HP ProLiant DL140 Intel® Xeon? DP-based performance server designed for scale out deployment and superior price/GigaFlop • HP's leadership on the Lustre-based scalable, shareable, open file system for HPTC Linux clusters • Record benchmark results for HP Integrity servers on several important high-performance computing benchmarks, including: SPEC HPC2002, SPEC OMP and the new Purdue TAP benchmark ratings Doubling server options for HP XC6000 and LC clusters Continuing to raise the bar in high-performance computing, the HP XC and LC clusters for HPTC offer simple and affordable solutions to support the needs of technical and commercial numeric-intensive computing environments. The scalable, Linux-based HP XC6000 clusters for HPTC now support the full and newly expanded line of HP Integrity rx2600 servers, which are powered by Intel Itanium® 2 processors. This doubles the number of server-node options available for the XC6000 clusters and provides customers with more affordable entry points into HP's line of powerful Linux-based clusters. Four rx2600 options are available: 1.0 GHz with 1.5 MB cache, 1.3 GHz with 3 MB cache, 1.4 GHz with 1.5 MB cache, and 1.5 GHz with 6 MB cache. The XC6000 also now supports the rx1600 servers (1.0 GHz with 1.5 MB cache), available in spring 2004. The High Performance Cluster LC2000 Series based on the industry-leading HP ProLiant DL360 (3.2 GHz with 1MB caches) providing world-class performance, robustness and manageability is now joined by the LC1000 Series based on the HP ProLiant DL140 Xeon DP-based performance server designed for scale-out deployment and outstanding price/GigaFlop. The DL140 is available at 2.4, 3.0, 3.2 GHz (with 1/2 MB caches at 2.4 to 3 GHz and 1 MB caches at 3.2 GHz). With the addition of these products, the HP XC6000 and LC cluster solutions now support a full range of server options, providing customers with an easier way to increase performance and lower costs using industry-standard servers. Asia Pacific's largest Linux cluster running on HP Integrity servers Among those customers seeking to increase performance and lower costs with HP Linux clusters is Taiwan's National Center for High performance Computing (NCHC), Taiwan's first national institute dedicated to HPTC and network applications research. HP Integrity servers are powering one of Asia Pacific's largest Linux clusters at NCHC. The center has invested in a 100-node Linux cluster, consisting of HP Integrity rx2600 servers with more than 200 Intel Itanium 2 processors. "We are pleased to partner with HP on this important project. With our firm commitment to research and innovation, we needed a computing environment to meet our scalability, sustainability and performance requirements," said Dr. Joe Juang, director, NCHC. "HP was able to deliver accurately what we needed to stay on course with our high-performance computing strategy. HP's industry-standard architecture, competitive pricing and customized on-site engineering service made HP an excellent choice to help us stay ahead in today's fast-changing IT industry." Tripling interconnect bandwidth for Integrity HP-UX clusters with InfiniBand 4x HP-UX Integrity clusters based on hptc/ClusterPack, HP's industry-leading cluster management suite, now support InfiniBand 4x, a message passing interconnect that provides three to10 times faster I/O performance over the previous HyperFabric2 and Ethernet options. HP is the first major vendor to release InfiniBand solutions for UNIX® platforms. InfiniBand's low latency and very high bandwidth are crucial for the performance of compute-intensive and message-passing applications for technical computing customers. InfiniBand 4x delivers over 750 MB/s of bandwidth and under seven microseconds of latency as measured by the MPI Ping-pong benchmark. This is over three times higher bandwidth and three times better latency compared to HyperFabric2 and about 10 times better performance (bandwidth and latency) than gigabit Ethernet. HP-UX Integrity clusters with InfiniBand significantly improve the scalability and performance of many distributed computing applications by lowering the overhead and allowing applications to efficiently scale to larger numbers of processors. New scalable, shareable Lustre-based file systems for Linux Increasing performance and lowering costs will be even further enhanced for Linux cluster customers by Lustre-based global file servers. Currently in development, Lustre demonstrates a new paradigm in designing scalable file system solutions. A highly scalable, shareable, open, parallel file system for Linux, Lustre will significantly increase I/O performance and decrease storage costs. HP is the prime contractor for a multi-year Tri-Labs (Sandia, Los Alamos, and Lawrence Livermore National Laboratories) PathForward project to develop Lustre in conjunction with Cluster File Systems Incorporated and Intel. The Tri-Labs project specifies a set of milestones where defined functionality and capability will be achieved. HP is building on this to produce Lustre-based products for HPTC markets. HPTC-oriented benchmarks HP Integrity servers, based on the Itanium 2 processor, have again achieved record results on several important high-performance technical computing benchmarks, including: SPEC HPC2002, SPEC OMP and the new Purdue Top Application Performance benchmark ratings. These new results, along with previously reported record results, place HP Integrity systems at the top of many of the prestigious HPTC-oriented benchmarks lists, including Linpack, SPECfp, Purdue's TAP list, (1) SPEC OMP (OpenMP) and the SPEC HPC benchmarks. • The HP Integrity Superdome server running 64 1.5-GHz Itanium 2 processors on the HP-UX operating system is the first system of any size to break the 300,000 barrier for SPEComp large performance with an official score of 303,161 on the SPECompLpeak2001 benchmark.(2) This same system also set the official leading score for any size server of 289,967 for the SPECompLbase2001 benchmark.(2) These SPEComp Large results are more than 53 percent higher than SGI's fastest result (an Origin 3900 system), faster than a Fujitsu HPC2500 system that used twice as many processors, and faster than results reported by all other vendors. • With these new record-breaking SPEC HPC2002 results on SPECchemM2002 of 56.66 and SPECenvM2002 of 309.95,(2) HP now holds the No. 1 position on the primary TAP list maintained by Purdue University.(1) HP also leads the TAP OMP list based on SPEC OMPM2001. TAP Lists are new high-performance computer rank lists that use realistic application benchmarks to measure performance. The primary TAP list (also referred to as the TAP HPC list) ranks high-performance computers based on the SPEC HPC2002 benchmarks, which are the largest, most realistic computational application benchmarks that are available to the public today. • In addition, HP Integrity leads the industry on 64-processor shared-memory Linpack N*N performance, sustaining 341.685 Gflop/s. HP Integrity systems also have maintained the No. 1 position on the Linpack report,(3) with no challengers since the 1.5-GHz Itanium 2-based HP Integrity systems were released in June 2003. These servers secured 1,635 Mflop/s for the 100x100 result and 5,303 Gflop/s for the 1000x1000 single-processor results. Since its introduction in June 2003, the HP Integrity rx2600 server has held the record for the fastest server SPECfp 2000 result in the industry: 2,106.(2) This result is yet to be challenged by any other vendor. The HP Integrity systems also maintained their leadership on the SPEC ENV2002 benchmark, with no challengers in this last round of updates. (1) Competitive TAP benchmark results stated above reflect results published on http://peak.ecn.purdue.edu/ParaMount/TAP/ as of Dec. 10, 2003. (2) Competitive SPEC benchmark results stated above reflect results published on www.spec.org as of Feb. 4, 2004. (3] More information on the Linpack benchmark is available at http://netlib.org/benchmark/performance.ps.