November 05, 2009, 10:21 AM — Much of the talk about cloud computing in data centers has been about enterprise workloads, but high-performance computing (HPC) customers are getting some options as well.
HPC specialist Platform Computing released some new tools Wednesday for customers looking to build a "private cloud," or a data center in which workloads can be transferred from one pool of servers to another to improve utilization rates and application performance.
Organizations doing HPC work, such as aircraft design or oil and gas exploration, often employ multiple clusters, each assigned to a different application with its own unique operating requirements, such as a particular Linux stack.
Cluster utilization rates are typically no higher than 40 percent, and customers aren't using the surplus capacity in part because their HPC applications can't be moved easily from one cluster to the next, according to Martin Harris, director of product management at Platform Computing.
The company's new product, Platform ISF Adaptive Cluster, can dynamically change the software stack on a cluster node so that workloads can be moved between clusters, allowing customers to get higher utilization. They need to also be using one of Platform's existing workload management products, Platform LSF or Platform Symphony.
"Platform LSF knows the intricate details of the workload itself, so it asks ISF Adaptive Cluster when it needs 10 more Windows nodes, for example. ISF Adaptive Cluster reaches out to the Linux cluster, repurposes those machines [as Windows nodes] and hands the IP addresses back up to LSF so the applications can use them," Harris said.
Most HPC environments aren't using virtualization yet, he said, so repurposing the node often involves rebooting the server and installing another software image locally or from the network. But by moving workloads between customers, Platform claims that customers beta testing the product have improved utilization to 80 percent or higher.
The company has worked with Microsoft to let ISF Adaptive Cluster provision nodes running Windows HPC Server 2008, as well as the main Linux distributions, Harris said. The product is available immediately, though Platform wouldn't disclose pricing.