Pathwork Diagnostics, a molecular diagnostics company, develops high-value diagnostic tests to aid oncologists in the diagnosis of hard-to-identify cancer tumors. Pathwork chooses optimal models for its tests by using proprietary machine learning algorithms to analyze large libraries of tumor specimen profiles.
Unfortunately, the processing of these models is a highly compute-intensive task. Tens of thousands of models must be processed to ﬁnd the best model to produce the diagnostic report. While the tests are highly parallelizable, the computation can still take weeks or months using a mid-size high performance computing (HPC) resource, such as a 64-node cluster.
Pathwork turned to cloud computing as a solution. “Our challenge was a perfect ﬁt for cloud computing. We needed access to more computing capacity than we could possibly maintain internally – but only at certain peak times. When we develop and deliver a product for clinical validation we’ll have weeks where we need access to almost unlimited capacity,” states Ljubomir Buturovic, Chief Scientist, Pathwork Diagnostics.
Pathwork investigated several cloud providers before choosing Amazon Elastic Com-pute Cloud (EC2) for scalable compute power. Then, the company selected Univa UD’s UniCloud to build HPC clusters in the EC2 cloud. “We found EC2 to be the most convenient cloud solution and UniCloud to be the best HPC software infrastructure tool to use,” recalls Buturovic.
UniCloud is an extension to UniCluster, Univa UD’s product for HPC systems management. With UniCloud, organizations can provision and scale capacity on Amazon EC2 environment, expanding baseline computing resources by provisioning capacity to meet peak demand. To get started, Pathwork simply downloaded and installed the UniCloud software with the help of a How-To reference white paper and were quickly up and running with access to a full functioning HPC cluster in the EC2 cloud environment.
With Amazon EC2, Pathwork Diagnostics saved hundreds of thousands of dollars by avoiding the expenses of purchasing and maintaining HPC hardware in-house and were able to accomplish key research innovations which would otherwise have been infeasible.
“By using Amazon EC2 and UniCloud, we’re getting exactly what we need. We pay for HPC power only when we use it, using a cluster management system that’s easy to use and operate. Now, we’re able to meet our peak computational demands without investing in new hardware,” says Buturovic.
(Published February 2009)