SUNY at Buffalo Center of Excellence blazes new trails in scientific research with high performance computing clusters and extreme networks.

The State University of New York (SUNY) at Buffalo Center of Excellence in Bioinformatics (BCEB) merges high-end technology, including super-computing and visualization, with expertise in genomics, proteomics, and bioimaging, to enable major contributions in science and healthcare. The Center of Excellence in Bioinformatics is poised to be a central player in this emerging revolution in biology. The theme of the center is systems biology--what the proteins do, how they interact and how drugs to treat diseases can be designed. Among major universities, the BCEB is unique in the scope and breadth of areas addressed.

High Performance Clustering 24x7

The BCEB planned to rely on a high performance computing cluster to accelerate cancer research, including human genome research, bioinformatics, protein structure prediction and large-scale computer simulations. In order to study the structure and orientation of human proteins--a crucial step in finding cures for many diseases--the cluster had to be able to support the demands of researchers and the massive simulations and calculations needed for their work. In this exacting 24x7 environment, system degradation and downtime could significantly impact cluster performance.

"Network performance, reliability and redundancy were vital to ensure the consistent performance of the cluster," explains Bartosz Ilkowski, head of cluster computing. "Extreme delivered a low cost infrastructure solution for our cluster that not only meets our performance needs, but improves our ability to ensure continuous operation."

Extreme Networks[R] high availability Layer 3 Ethernet switching solution delivers a non-blocking switch fabric and non-blocking architecture to support redundant, dual-homed connections for scalable and fault tolerant meshed topologies--all at wire-speed. And, because the cluster will empower scientists who use it to potentially make vitally important discoveries, the solution's wire-speed Ethernet capabilities ensure that mission critical clustering applications perform as expected, regardless of the complexity or network and server load, with the lowest possible incremental latency.

Reducing Costs and Complexity of the Cluster

Costs were a consideration when designing the cluster, but with Extreme Networks the BCEB was able to reap the benefits of a high-performance infrastructure without increasing costs or administrative complexities. With a core network already consisting of Extreme Networks BlackDiamond[R] switches, the BCEB decided to deploy additional BlackDiamonds in the cluster network to ensure consistency and simplify network management, area network (SAN) and also acts as an entry point for building aggregation.

"Even though the cluster required unparalleled performance and availability, we were able to deploy a single platform," notes Ilkowski. "Extreme enabled us to reduce management, training and on-going costs with a single platform capable of meeting all our needs."

By sharing a consistent architecture between all Extreme Networks switches, the operating system and the management platform, Extreme's solution can deliver high-performance clustering while reducing network ownership costs and management complexity. With the ExtremeWare[R] software suite, similar tasks can be handled for all switches at onceaesaving time and resources and making it easy to manage across the network. In addition, the high port density available in Extreme Networks switches ensures support for a large number of server clusters without increasing the cost of supporting those clusters.

The BCEB worked closely with partners Dell, IBM and Extreme Networks to design the final solution, and today the Center's multi-vendor computing infrastructure consists of 1,900 two-processor Pentium III servers and 100 two-processor Xeon servers from Dell, 266 two-processor Xeon blade servers from IBM, eight storage servers providing access to 14-terabyte storage area network from Dell and EMC and 5-terabyte storage area network (SAN) from IBM and six servers for managing workloads. The cluster runs on Linux and utilizes Platform Computing's LSF 5 work management software and relies on Extreme Networks BlackDiamond switches for Gigabit I/O connectivity between the nodes. In all, the cluster fills 60 computer racks.

"The initial Buffalo cluster cost only about $4 million," said Dr. Jeffrey Skolnick, who will use the system to conduct protein research in the university's Center of Excellence. "it is the poor man's supercomputer. We're basically stringing together a large number of quasi-independent computers."

The resulting solution has been so successful; SUNY Buffalo was the first university to receive Dell's research excellence award.

Continuing Excellence

The BCEB looks forward to continuing the growth of its network in support of the university's research efforts, including the development of a complete infrastructure for a grid of computing clusters.

"So far the cluster has performed up to expectations," says Ilkowski. "With the unparalleled performance and power of the network, our researchers have the tools to truly achieve the highest level of research excellence."