Your Friendly Neighborhood Supercomputer - page 2
8,000 PCs and Nothing to Do...
How this all works is pretty straightforward: customers can approach Gateway and ask that their particular application be applied to the grid. According to Gateway's Sr. Product Manager for Distributed Computing Permal Kazi, the job is then ported to the MetaProcessor server and then set loose on the Gateway grid via the agent application that runs on each Windows client in the grid.
How long this takes, Kazi explained, really depends on the application. If Gateway is dealing with a repeat job from a repeat customer, the job can be implemented almost immediately. For new, custom applications, a full port can take some time. But the up-front time is something that all such applications have to face when applied to any grid. The real benefit of this program is that it is allowing access to true supercomputing power for clientele who might not have ever thought they could get access to such capabilities before.
United Devices' VP of Marketing Paul Kirchoff described the two types of customers who will benefit the most from this service. First, there are the small- to mid-sized companies that need supercomputer processing and could not afford such an expense before. Kirchoff listed such industries as oil and gas companies, life science, geo science, and semiconductor firms as potential benefitors from this grid program.
The pricing of the program seems about right for such a market, too. Though up-front costs can vary depending on the level of porting to be done, the cost of the run-time for using this grid is $0.15 (US) per processor-hour. So, assuming all nodes were being fully utilized all at once, 8,000 processor-hours (nearly a full processor-year) would cost just $1,200 (US).
The other group of customers that will get good use out of this program, Kirchoff explained, are those companies that are all ready using their own in-house enterprise grid deployment but need a little computational "oomph" for a particular application but don't want to take the financial plunge and buy what they need for what is essentially a one-time project. Now these customers can turn to Gateway's grid instead and let the in-store PCs handle the load.
The MetaProcessor server can handle quite a bit of the management duties such a grid needs. "There can be hundreds of thousands of concurrent jobs going on at the same time," Kirchoff said. The processing loads can be scheduled all the way down to each individual node on the network.
What's more, Kazi added, if a job needs a certain threshold level of computational power (such as P4 processors that have over 2 GHz of speed) for the client agents, the MetaProcessor can filter out the job load to only the eligible client processors.
The public will be very unaware that all of this is going on, Kirchoff explained. Whenever a demo machine needs to be used, the agent will note the sudden use and release the processing power back to the human user's needs out at the store.
Kazi also indicated that in the near future, Gateway will enable technology that will allow the in-store machines to automatically boot into the Linux operating system at night, when the demonstration models are unseen by the public. This will allow grid applications to run that need a Linux client platform. This will further maximize the capabilities of the grid, he said.