FRAMINGHAM (11/24/2003) - At the local pool hall, you pay for as many tables as you need to accommodate your party crowd. And you pay only for the time you use the tables.
Similarly, with new high-performance computing services from IBM Corp. and a joint offering from Gateway Inc. and United Devices Inc., life science companies are charged only for the number of processors used and the time required to complete a computation or simulation. And when the in silico project is completed, you can simply walk away. New Twist on an Old Service Utility computing offers life science companies an alternative to traditional outsourced IT services. Rather than locking companies into rigid fixed-price, long-term contracts, new offerings from IBM and United Devices tout a pay-as-you go, pay-per-use model.
IBM's Deep Computing on-demand service and the Gateway Processing On Demand service, which uses United Devices' Grid MP platform, offer biotechs and pharmas an alternative to purchasing powerful computer equipment or using a traditional outsourced data center.
The new services are realizations of a concept that has been around for many years: utility computing. The term has several meanings within the IT community, but here, utility computing is defined as a service -- analogous to utilities such as electricity or water -- where a company pays as it goes for its necessary computing power.
The IBM service "falls into an inter-domain (for companies) not contemplating purchase of their own systems or using an outsource service that (implies) a long-term commitment," says David Turek, vice president of IBM Deep Computing.
Turek notes that the service is ideal for "applications that need to be executed one time or with some degree of periodicity." In particular, "the service is geared to discontinuous events -- typically married to things of great urgency," he says.
The need for speed is certainly what drew the medical technology company Electro-Optical Sciences (EOS) to the IBM service. EOS is seeking FDA approval for a new melanoma early-detection diagnostic tool called MelaFind. The device, which resembles the nozzle of a showerhead, performs a multi-wavelength light scan of a patient's skin. Images are collected and securely transmitted over the Internet to the IBM Deep Computing on-demand center, in Poughkeepsie, N.Y. There, an EOS pattern-recognition program compares the scanned images to a database of more than 20,000 high-resolution images of skin lesions.
EOS designed MelaFind to be used by doctors and nurses, ideally in such a way that would give patients their results while they waited in the office. As a relatively small company, EOS lacks the high-performance computing power required to process images in a timely manner. "The (on-demand) supercomputing power enables us to quickly train the pattern-recognition algorithms to obtain a reliable diagnosis," says Marek Elbaum, EOS CEO.
In EOS' case, the utility service provides short, occasional bursts of supercomputing power. If EOS were to install its own systems, the computers would sit idly by most of the time.
The American Diabetes Association (ADA) is taking advantage of a similar service provided by Gateway/United Devices to complement its existing computational capacity. This allows the ADA to ramp up research projects quickly without investing in additional computer equipment.
The ADA uses the service to run Archimedes software to model things such as determining how minor lifestyle changes can reduce a diabetic's risk of complications, or how heart-attack rates are reduced by lowering blood glucose levels. The ADA had previously teamed with other groups to perform simulations (see "Entelos Gives Its Computing Mix a Stir," June 2002 Bio-IT World, page 32). But "renting supercomputer time to (run Archimedes) can cost hundreds of thousands of dollars," says Richard Kahn, ADA's chief scientific officer. The Gateway/United Devices service offers the ADA an efficient alternative.
Industry analysts say that both services have a similar pricing structure. There's a fee to establish an account and start the software running, and an hourly rate per central processing unit (CPU) used.
IBM will not disclose exact pricing for its service. United Devices charges a one-time fee of about US$1,000 to set up an account. It provides an optional service that helps a company port its application to the grid. "But you can also do it yourself," says Paul Kirchoff, vice president of marketing at United Devices. The company will give its customers a tool to port the application.
Once the program is ported, United Devices charges 30 cents per GHz hour (the computers use 1GHz processors, so this rate is equivalent to 30 cents per CPU hour). Hourly blocks can be purchased at a slightly discounted rate. If a company runs a simulation on 1,000 computers (assuming one CPU per computer), and the job takes six hours to complete, that would cost about $1,800.
To put that hourly cost into perspective, the Virginia Bioinformatics Institute at Virginia Tech charges researchers about 21 cents per CPU hour to run programs on its Linux cluster. And a network manager (who wished not to be identified) at a Cambridge, Mass., life science company said that a rate of about 25 cents per CPU hour would be appropriate to offset the power and cooling costs to run his data center, IT labor charges to manage and administer the servers in its clusters, and annual fees such as warranties and service contracts.
The IBM and Gateway/United Devices services will not likely replace traditional outsourced data centers (which require long-term contracts whether the systems are used or not), nor deter companies from buying their own computing facilities. However, they do represent an interesting alternative that, in some scenarios, makes economic sense.
IT managers should bear that in mind the next time they feel behind the eight ball in need of extra computing power.