Do virtualization and private clouds have the oomph to run demanding research workloads that currently run on high performance Linux clusters? Computer scientists want to find out.
Carnegie Mellon University is one of the first users of the Hewlett-Packard CloudStart bundle announced at VMworld 2010 today, which the university will use to provide compute resources across its school of computer science and college of engineering. As the new private cloud comes online, the goal will be to convince users to perform their research on it, rather than on the dozens of compute clusters that already exist across the university, said Dr. Greg Ganger, Jatras professor of electrical and computer engineering and director of Parallel Data Lab.
There’s some initial trepidation, Ganger said. “There’s nervousness among some groups about what they’re going to lose performance-wise by running on virtual machines,” he said.
However, its Ganger’s belief that virtualization’s performance penalty is overstated. “It doesn’t hold in all places; it holds in some places.” Among Carnegie Mellon’s tasks as a CloudStart user is to “look for when there’s going to be a loss in terms of performance or flexibility, and find ways of minimizing that.”
But before that can happen, Ganger and his team have to convince researchers to move workloads on to the new private cloud, which is based on HP BladeSystem Matrix blades and scale-out NAS storage. “There’s always work in transitioning, vs. no work in keeping it [on existing hardware].”
And unlike in the corporate world, where CIOs rule with an iron fist, Ganger can’t just order everyone on to the cloud. “We don’t have a stick; what we have is a carrot,” he said.