Mitchell Bingemann | July 22, 2008
case study | Monash University
MONASH University wanted to become one of the leading biomedical research facilities in the world but a shortage of computer processing power meant it was lagging its international peers.
Infrastructure and major IT projects manager Adrian Ling decided cluster computing was the way to go for Monash
Two years ago, the university's supply of high-powered computing systems was fragmented and researchers were often forced to hire outside computing facilities for their work.
As in most universities, funding was scarce and with no supercomputer at its disposal, Monash University needed to take action to ensure it kept in touch with the world's leading research facilities.
Infrastructure and major IT projects manager Adrian Ling decided cluster computing was the way to achieve this.
"Like all successful universities we have a push to bring world class researchers here, but to do that we needed to start providing better computer facilities," Ling says.
"We always knew a computing cluster would best suit our needs and had even attempted a couple makeshift clusters with desktop PCs in the labs, but it was really a Mickey Mouse operation."
After consulting with five vendors, the university settled on a high-performance computing cluster based on Dell PowerEdge blade servers.
One of the key considerations for future-proofing the cluster was scalability, so Ling and his team constructed it with off-the-shelf hardware.
"The use of commodity based enterprise hardware has provided the advantage of being able to purchase much more CPU power per dollar than is available in custom-built cluster systems," Ling says.
It has also allowed the university's cluster to be a true plug-and-play system, enabling it to tie different hardware generations together in one system without any modifications to the original system.
"As money becomes available we can simply purchase more blades of whatever the current technology offering is and plug them into our system," Ling says.
When the cluster was first constructed in 2006 it consisted of 15 blade servers and 30TB of disk storage running on Red Hat Enterprise Linux.
Just 18 months and 343,500 experiments later, the cluster has tripled in size and makes use of five blade enclosures to accommodate its servers, which total more than 240 CPU cores in parallel operation.
"We wanted to keep it small at first so we wouldn't run it off the rails, but overnight our needs pretty much doubled," Ling says.
"In fact, the hardest bit about deploying the cluster has been keeping up with the demand for its use."
Due to the high work demands of the university's researchers, Ling and his team were unable to run the cluster through a development cycle and were pushed straight into a production environment.
"Our researchers desperately needed the cluster so we were always conscious there was a tight time line."
Skipping the cluster's development stage meant Ling and his team became experts at tweaking the performance of the cluster.
"We quickly learnt what would work and what wouldn't," he says.
"We flew by the seat of our pants when building it, but I would do it all again because it was fun and successful."
The cluster now consists of about 40 blade servers and 100TB of storage capacity, and is as popular as ever.
In fact, it's so popular with the university researchers that Ling hasn't had a chance to run tests that gauge the cluster's raw computing power. One of the main culprits to tying up much of the cluster's time is the lead researcher of the university's Drug Discovery Laboratory, Professor Patrick Sexton. The Drug Discovery Laboratory and Professor Sexton use the cluster to construct complex three-dimensional models of protein structures, which aid in the development of therapeutic drugs.
Before the cluster's time the university lacked infrastructure powerful enough to process the high volumes of data needed to construct these complex 3D models.
Now, with the cluster in place, Sexton and his peers are able to advance their research to the point that Monash University now has one of the leading biomedical research facilities in the world.
In fact, Sexton and his colleagues have been so successful that they have secured a $6.5 million program grant from the National Health and Medical Research Council to support their research.
"The analysis we do is critically dependent on parallel processing power," he says.
"Now that the cluster has more than 200 CPU cores we can do that.
"In many instances it makes our work achievable in a week or so, where previously it would take much, much longer."
Now that the university has achieved its spot as one of the world's leading biomedical research facilities, what's next for the cluster?
"You need to be first in biomedical research. You don't get prizes for second or third," Sexton says.
"So as long as we have researchers coming to Monash, our cluster will continue to grow."