Supercomputer power goes to the cloud

The computing power of supercomputers is no longer limited to organizations which can afford to run their own high-performance computing (HPC) labs as cloud-delivered HPC-as-a-service (HaaS) gain wider acceptance in the market, according to market players.

In an e-mail interview with ZDNet Asia, Sandra Lue, director of IBM Asean's cloud solutions, said the company believes demand for HaaS will grow as enterprises realize the delivery model can yield the same benefits that drove the adoption of cloud computing. These include the efficient and speedy utilization of resources, pay-per-use model for standardized services and the ability to use resources only when needed, thus, removing the need to purchase dedicated siloed HPC environments for each user group, Lue said.

Tom Coull, senior vice president and general manager of on-demand computing from Penguin Computing, concurred. The high cost and effort required to bring an on-premise HPC cluster online is encouraging organizations to look at cloud-based HPC systems, especially those with intermittent computing or storage needs, Coull explained.

He added that a company's need to reduce IT expenditure and minimize capex (capital expenditure) further justify turning to the cloud for HPC capabilities.

The San Francisco-based company started offering its HaaS, named Penguin Computing on Demand (POD), commercially in January 2010 and has processed over 5 million jobs since its debut. Its HaaS customers hail from various countries including United States, Canada, Australia, the United Kingdom, Germany, Japan, Korea and Belgium, said Coull.

Mechanics behind HaaS
Noting that HaaS works the same way as cloud computing, Lue explained: "When cloud service providers (CSP) start providing HaaS, users will be able to request or purchase HPC resources just as today's cloud users request for basic infrastructure from the CSPs.

"In a typical scenario, a CSP may create several standard predefined HPC clusters based on user needs and demand, and make them available via a catalog. The users will them simply select from the catalog and request the CSP to provision those clusters for the required duration," she said.

At Penguin Computing, Coull said it is "quite simple" for customers to buy HaaS computing power. "A customer signs up for POD using our online form, then works with our HPC support staff to create their compute environment," he said. "After that, they log in and run their jobs as needed. They can send their data to POD over our high-speed Internet connection or directly on disk using our POD Caddy service [via courier]."

Penguin Computing's HaaS customers come from a range of verticals including bioinformatics, discrete manufacturing, defense, geographic information services, academia and entertainment, he added. These companies use the service to support compute and storage resources during peak needs, he explained, while others tap the offering to completely replace their on-premise systems if it proves more cost-effective or convenient to outsource their HPC requirement.

S'pore university builds private HPC cloud
In Singapore, Nanyang Technological University (NTU) and IBM announced in late-May that they were in the pilot phase of creating a cloud-based technology platform to allow researchers and students at the local tertiary institution to utilize the university's HPC environment. The pilot phase of the private HPC cloud is expected to complete in March 2012, said Son Huynh, HPC cloud project executive at IBM, in an e-mail.

While NTU's HaaS will be limited to its own community, Lue told ZDNet Asia that Big Blue hopes to expand its private HPC cloud services to other verticals such as the oil and gas industry for exploration and seismic research, the automobile and aerospace industry for engineering simulation and modeling, and the financial trading industry for data analytics and business intelligence.

In an e-mail interview, Melvin Hwee Soh, NTU's system manager and HPC cloud project manager, underscored the benefits a cloud-delivered HPC platform can offer over on-premise HPC clusters. While the traditional HPC model is highly optimized, he noted that users may be limited to a single fixed set of environment and have to adapt to that environment.

However, with HPC on a cloud platform, this limitation can be overcome by allowing users to have the flexibility to quickly provision any type of HPC environment they need or an environment that may be most suitable for the users to work on their problem sets, Soh added.

IBM's Son added that based on actual results from previous cloud implementations, organizations can expect their HPC utilization to increase by 75 to 90 percent, average provisioning time to reduce by 50 to 75 percent and performance benchmark to improve 2 to 4 times, compared with existing on-premise HPC.

That said, he noted that the cloud platform is still an emerging and rapidly evolving technology. Therefore, implementing a private HaaS platform should be approached in the same way as any cloud implementation and would require a well defined scope, objectives, success criteria, resource planning and skill development, Son said.