Supercomputers have long been available, but only to the largest of enterprises. Even in those environments though, computing capabilities are often underutilized and it footprint often enormous.
Last year, IBM sought to level the supercomputing playing field by bringing its capabilities to the masses. It opened the Deep Computing Capacity on Demand Center in Poughkeepsie, New York. Now, rather than shelling out millions of dollars for a supercomputer that is equally expensive to maintain, enterprises can purchase a virtual segment of a machine for considerably less.
Today, the Center claims 20 customers ranging in size from a 25-user pharmaceutical firm to a large financial services firm. Customers are equally geographically dispersed, in locations from North America to London to Australia, Mark Solomon, program director Deep Computing Capacity on Demand told ServerWatch.
Most customers come from vertical industries, like gas & oil, pharmaceutical, life sciences, digital animation, and financial services industries that benefit from supercomputing capabilities for specific tasks at specific times, Solomon said.
The Center "allows customers to do things they've never done before," Solomon said, as well as the ability to "do more than they ever could before." He sees three basic advantages for enterprises that have access to a supercomputer: Customers can take on ventures they were unable to before; in-house resources are freed up for other tasks; and project cycles times are shortened, resulting in cost savings for the enterprise.
Other, less tangible, benefits of renting supercomputing power are that customers need not worry about managing the rapid life cycle of their supercomputing resources or of overloading their data center real estate, Solomon said.
The Deep Computing on Demand facility launched with 512 systems in June 2003. Today, it has nearly 2,400 xSeries servers, Opteron-based servers, and BladeCenter systems. Maxed out of space, IBM last week launched a second Deep Computing Center in Montpellier, France.
Although Solomon admitted that geography makes little difference from a technological perspective (case in point, a U.K.-based customer that began using the supercomputer services this week is initially using the Poughkeepsie facility), from a sales and marketing perspective regionally dispersed Centers are beneficial. Solomon added that geographically distributed Centers are also advantageous from a disaster recovery and security perspective.
The new Deep Computing on Demand Center will offer customers access to pSeries AIX or Linux servers and xSeries systems running Linux or Windows with related disk storage. Workloads will be shared between the Montpellier and Poughkeepsie, Solomon said.
Enterprises that sign on to use the supercomputing center receive a VPN connection good for one year. Data is uploaded from their site to a virtual Linux cluster on a secure storage device on IBM's premises. Customers pay only for the time spent using the supercomputer; during that time, they are the only entity using the resource. This maximizes security and performance, Solomon said.
IBM is already planning to implement grid technologies to enable resource sharing within and across the two Deep Computing Capacity on Demand Centers. According to Big Blue, a grid environment will lend itself to a new pay-for-use model to complement the existing reservation model, and thus offer enterprises additional convenience and flexibility to access and pay for supercomputing power. Solomon noted that one customer has already implemented a grid-like infrastructure on its own, splitting the workload between its servers and the Poughkeepsie Center.
Wilsonville, Oregon-based Mentor Graphics is the first customer to sign on for the IBM's new computing center. It will power up from the pSeries 690 systems in Montpellier, as well as eServer xSeries systems based in Poughkeepsie. Mentor is also testing its Calibre design to silicon software platform on Opteron-based eServer systems in the Poughkeepsie facility.
IBM is considering other sites for additional Deep Computing on Demand Centers, though none are planned at this time.