Korean Institute of Science and Technology Information

Story Details

KISTI is a public research institute tasked with supporting Korea's research community. KISTI established its Supercomputing Center in 1967. Over the decades, the center has grown to provide high-end technologies in supercomputing and networking, along with active support and services, to various research communities in academia and industry, as well as research institutes and governmental and scientific organizations. Most of the center's users are scientists and engineers who need to process large calculations with high precision. The present KISTI Supercomputing Center was established in 2001 and is the largest provider of supercomputing resources and high performance networks in Korea, supplying 40 percent of supercomputing resources in the country.

In 2003 KISTI built its third supercomputer based on an IBM p690-Series SMP cluster comprised of 21, 32-processor compute nodes, with a peak performance of approximately 4.3 Teraflops (TFLOPS). Increasing demand from users working on complex projects requiring greater computing power was putting a strain on resources and delaying research progress. With a policy of renewing its supercomputer every five years, KISTI wanted to develop its fourth supercomputer built on a High Performance Computer (HPC) cluster that would meet its existing and future performance needs, and provide the best price-performance ratio.

IBM, HP, DELL, Cray, Samsung and Sun Microsystems presented proposals for the KISTI HPC project, with IBM, HP and Sun being shortlisted. Sun was awarded the KISTI HPC project for its 300 TFLOPS Sun Constellation System HPC cluster. KISTI was impressed by the proposed switch architecture and Sun’s performance. “The performance of the Sun system was the best among the three vendors, and Sun also offered an excellent collaboration plan,” explains Sik Lee, Ph.D, Team Leader, Supercomputing Applications Team, KISTI.

The Sun Constellation System is the world's first open HPC architecture capable of scaling from departmental clusters all the way up to the largest supercomputer in the world. It provides an open Petascale computing environment combining ultra-dense high performance computing, networking, storage and software into an integrated system. The KISTI HPC project started in December 2007. The fourth KISTI supercomputer, named “Tachyon” was designed as general-purpose supercomputer to be used predominately for academic research. In the first phase of the KISTI HPC project, 188 compute nodes (3,008 cores), four login nodes consisting of 32 CPUs, four debugging nodes consisting of 64 CPUs, 203 TB of disk storage, and 410 TB of tape storage, were installed. In the second phase of the project 300 TFLOPS of compute nodes (21,504 cores), 1 PB of disk storage, and 2 PB of tape storage will be deployed.

The KISTI supercomputer comprises innovative HPC cluster components including Sun Blade 6048 chassis with Sun Blade servers acting as compute nodes. Unique to the Sun Blade 6048 modular system is the Sun Blade 6048 InfiniBand Switched Network Express Module, the industry's only Chassis Integrated Leaf Switch. It provides the lowest cost per port with Sun Datacenter Switch 3456, which will be deployed in phase two of the project to consolidate intercommunication among computing nodes running parallel applications. Phase Two of this project is expected to deploy in 2009.

Sun Fire X4600 servers and Sun Fire X4200 M2 servers act as infrastructure nodes. KISTI also deployed Sun Fire X4500 servers and Sun Fire X4100 M2 servers, with all servers running the Solaris 10 Operating System and CentOS, an open-source Linux distribution. Sun xVM Ops Center provides KISTI with the ability to manage both its Linux and Solaris Operating System environments within a single unified solution. KISTI uses Sun xVM Ops Center to quickly and easily update, monitor and manage all servers, saving the organization time and money. All compute jobs are delivered to the Sun servers using Sun Grid Engine 6.2 software, which provides policy-based workload management and dynamic provisioning of application workloads.

KISTI is using the Sun StorageTek 6140 array as disk storage, configured with five SAN arrays, 83 TB for data backup and archiving, 224 FC-SATA drives, and RAID 5+1. The Sun StorageTek SL8500 Modular Library System has been deployed as tape storage for data archiving and data backup to meet KISTI's needs for scalability, performance, and reliability. Currently KISTI has 16 Sun StorageTek LTO3 tape drives and 1050 media slots. “Sun StorageTek arrays provide us with efficient price-performance,” confirms Lee. KISTI is using Sun StorageTek QFS (SAM-QFS) Shared File System software for data archiving. SAM-QFS migrates and recalls huge user data to and from tape storage. To accelerate the transfer of stored data on the cluster, KISTI uses the Lustre file system.

Tachyon was made available to users in late August 2008. The primary “grand challenge” applications running on the HPC cluster currently include a large scale molecular dynamics study on membrane protein, collision simulation on black-holes, and multi-physics problems on solid surfaces.

Compared to the previous cluster based on a IBM p690-Series SMP cluster, performance has improved by approximately eight times. KISTI aims to achieve 24 TFLOPS in performance by the end of phase two, and 300 TFLOPS peak performance by 2010. The KISTI supercomputer achieved a rank of 130 in the recently published Top500 project list for supercomputers, enabling KISTI to maintain its position as one of the leading supercomputing centers in Asia.

“In the near future, we expect tremendous scalability from the compute nodes and the Sun Datacenter Switch 3456. We are also optimistic about the future of Sun x64 servers because Sun has a versatile road map for both blade and rack servers for HPC. Furthermore, Sun is the only company that has a complete HPC portfolio, from hardware to software,” says Lee.