Windows® Compute Cluster Server 2003 is helping researchers support their complex analysis with the appropriate computing power while reducing high support costs and the need for specialized skills.
Traditionally, open source platforms including Linux® or UNIX®, have been the engine for large server clusters. While providing massive computational power, research teams are often tasked with system management duties, and the associated costs are high. Those issues along with the need to extend high-performance computing capabilities are pushing research-focused organizations to seek alternative options.
Cluster meets budget, compute power needs
Three years ago, when Professor David Holmes joined the Life Science Foundation in Santiago, Chile, the not-for-profit had no HPC environment. He set out to investigate the various computing approaches to support the foundation’s wide ranging research work.
“I knew, given the series of research programs in progress, we couldn’t do without computing power,” says Holmes, Director of the Center for Bioinformatics and Genome Biology. Holmes considered various cluster options—from Macintosh configurations to a Sun cluster and even a Beowulf setup. “All were quite expensive, and cost is an issue for our center. Interoperability with our applications is also a big factor.”
When a Microsoft team visited the foundation a year ago to discuss the option of Windows® Compute Cluster Server (CCS) 2003, Holmes knew the cluster would not only provide the computing power needed, but extend HPC to users not comfortable with traditional HPC platform environments.
The 32-node cluster, which went live this past September, runs on a HP Opteron™-based 64-bit dual core platform and requires the skills of one full time network administrator. Holmes says it has boosted research productivity considerably and promises even more given the “headroom” built into the cluster. “What once took a week to do can now be done in an hour or two,” he notes, adding that his team is gaining unexpected benefits as well.
“We’re not just processing more data, we can now ask questions that were impossible before, and that has opened up exciting new areas of research. We’re benefiting from computational predictions, and our shared computing capacity is now on the terabytes level.”
Expanding HPC capabilities
Queen’s University in Belfast has significant investments in UNIX- and Linux-based clusters. While the University has successfully deployed HPC systems and tools, leveraging the resources requires significant training.
Research teams under Peter Hamilton, Professor of Bio-Imaging and Informatics, and Danny Crookes, Professor of Computing Engineering, were sharing a cluster for cancer cell scan imaging. They wanted to speed up image processing time, and be able to use a variety of third-party software not available on the traditional HPC platforms.
“Our objective was to improve our ability to diagnose cancer earlier and on a much more reliable basis using computer vision of microscopic images. It’s very complex analysis and the images are enormous in size. This can only be done effectively with HPC,” explains Hamilton.
Cluster implementation
Queen’s University installed Windows CCS in January 2006 on an 8-node cluster of HP ProLiant DL140 servers, each with 4 gigabytes of memory. These clusters are connected to the University’s storage area network (SAN), which is shared with the Linux-based clusters. The network architecture is built to a standard HP design so it can be expanded as needed.
After six months of testing, the team installed the distributed MATLAB application, and also ported the 32-bit version of ANSYS CFX, a computational fluid dynamics software application. The Queen’s University cluster project group is now meeting with other University departments on how the Windows-based cluster can serve their needs as well.
The professors say in addition to the computational benefits, they expect the use of the cluster to free up other computer resources throughout the University. Having a centrally managed cluster also means little duplication of support costs.
“Today’s Linux-based clusters require considerable knowledge and experience to assemble the right parts, get them working together, and then keep up with ongoing patches. Microsoft’s traditional strength in delivering well integrated enterprise platforms will make it easier for many more companies to use HPC,” says Don Rule, Microsoft's Platform Strategy Advisor for Bioinformatics.
“We have a new way to handle information and data. We have a shared computing capacity that provides collaboration, which is vital to our industry,” Holmes says. “The cluster advancement has increased our ability to do research that we hope will improve the life and economy of Chile.”
|
Printable version
