Tulane University is home to the Center for Computational Science (CCS), a unique facility designed to provide computational resources for research projects across many disciplines. The Center provides an infrastructure for investigators interested in computational science to exchange ideas, produce research, and establish new collaborations.

The Challenge
One of these collaborative efforts involves a team of researchers performing computational simulations of multi-scale models in biological systems. This research is funded through individual research grants from the National Institute of Health (NIH), National Science Foundation (NSF), and National Aeronautics and Space Administration (NASA), as well as a Center grant from the NIH entitled "Biocomputing: Integrating Molecular/Organ-Level Function."

"Since computationally demanding simulations of 3D models of bioelectric phenomena, fluid-structure interactions, and molecular dynamics simulation are important to these research projects, it was important for the CCS to obtain a scalable high-performance computing system that could be easily shared among several different research groups," said Dr. Donald Gaver, director of the CCS.

For example, Dr. Natalia Trayanova, professor of biomedical engineering, studies cardiac defibrillation - the application of a strong electric shock to the heart to restore cardiac rhythm and prevent sudden cardiac death. Despite the critical role that defibrillation therapy plays in saving human life, it is only known what happens on the surface of the heart during these electric shocks. By using computer models, Dr. Trayanova is able to produce a 3D simulation of a heart to better understand what happens during a shock within the cardiac walls. This improved understanding is expected to lead to new advancements and optimization of the clinical procedure of cardiac defibrillation.

"Computer models and 3D simulations are imperative to my research to better understand how electric shock affects the heart," said Dr. Trayanova. "Therefore, it is crucial that the computing system we use be capable of handling compute-intensive workloads and simulations."

Tulane's existing biomedical server was a shared memory machine, but the cost of upgrading and maintaining the system was already quite costly - Tulane knew expanding this existing system would be too expensive.

"To achieve successful, accurate results in a timely manner for our researchers' projects, we needed a computing system that was fast, reliable, but affordable," said Rene Salmon, senior systems analyst for CCS. "Since the outcome of so many projects would depend on the reliability and ease-of-use of the system, we had to seriously consider what type of computing system would benefit our users the most."

Investigating Linux Clusters
Tulane started looking for computing alternatives that could fit within their budget and provide the speed and reliability that was crucial for ensuring successful research results. After investigating several alternatives, Tulane became increasingly interested in Linux clusters. This distributed computing platform seemed to provide the price/performance ratio Tulane was interested in achieving. However, the researchers at the CCS were worried about the difficulty of migrating to a Linux cluster.

"With a Linux cluster we could afford many more CPUs, which allowed us to run much larger simulations and get faster results at a fraction of the cost," said Salmon. "However, we were concerned about what would be involved in migrating our codes to a new system. Luckily, our researches were willing to adapt their codes to run on a Linux cluster."

The scalability of Linux clusters was another attractive feature for Tulane. As researchers received more grant money from various sources, they could add more compute nodes to the cluster and increase its computing power.

"The great thing about the scalability of Linux clusters is that it allows our researchers to pool their resources so they can have access to more CPUs and computing power," said Salmon.

Managing the Linux Cluster
Since the Linux cluster would be running multiple programs for different users, it was imperative that the computing system Tulane chose be easy to use and manage. Tulane was drawn to the comprehensive management tools offered by Linux Networx, especially the Icebox management appliance and LinuxBIOS.

"Since we knew the cluster would be running multiple programs for several projects simultaneously, how we would manage the cluster became a big concern," said Salmon. "After looking at Linux Networx's management tools, we were very impressed with the capabilities of Icebox. The serial terminal server and a remote-controlled power distribution were also critical for ensuring a productive, scalable system."

Linux Networx's expertise with LinuxBIOS, an open source BIOS alternative, was another management feature that interested Tulane. LinuxBIOS performs the same basic functions as commercial BIOS only 10-20 times faster. LinuxBIOS initializes the hardware, checks for valid memory, and begins loading the operating system in about three seconds. Most commercial BIOS require about 30-60 seconds to perform the same tasks. In addition, LinuxBIOS can be configured and accessed from within the Linux operating system. This means changes to the BIOS can be made remotely to a single node or to all the nodes in a cluster system.

"LinuxBIOS was important to us for scalability reasons. We really liked the fact that we could make changes to the BIOS on the entire cluster from Linux with just a few commands," said Salmon.

With expertise in LinuxBIOS, comprehensive management tools, and powerful computing architecture, Linux Networx became Tulane's choice for their first cluster implementation.

"We wanted a Linux cluster vendor that could provide us with an easy-to-manage cluster that could scale and grow easily. Linux Networx did this for us," said Salmon.

Implementing a Linux Networx Cluster System
Starting with a 20-processor Evolocity system, Tulane researchers quickly noticed the benefits of Linux Networx cluster technology and have scaled up their cluster four times as additional grant money has been received. Their Evolocity cluster now totals over 80 AMD Opteron processors.

"The scalability of the Evolocity cluster has proved to be an essential feature as we've quadrupled the cluster's power since we first installed it," said Salmon. "Scaling the cluster has also been a fairly easy process as programs and codes were easily migrated."

Results
Tulane researchers quickly noticed the benefits and potential of Linux cluster technology. In the past, researchers were limited to running small simulations, now they can run much larger, complex simulations than was possible before.

"Running large, detailed simulations is essential to our research efforts as it allows us to better understand biomedical functions," said Dr. Trayanova. "Additionally, a quick turnaround time for running simulations is essential as there are several research groups that need time on the cluster to complete their simulations."

Most important, with the power and capabilities of the Linux Networx system, Tulane researchers can improve the quality of their research and achieve much more focused and valid results than they were able to accomplish previously.

"By completing these jobs in a timely manner, researchers can get results faster, which helps advance the entire mission of the biocomputing program and allows us to do better research," said Dr. Trayanova.