IBM Provides Access to Blue Gene On Demand

neutron_p writes "IBM's world renowned Blue Gene supercomputing system, the most powerful supercomputer, is now available at new Deep Computing Capacity on Demand Center in Rochester, MN. The new Center will allow customers and partners, for the first time ever, to remotely access the Blue Gene system through a highly secure and dedicated Virtual Private Network and pay only for the amount of capacity reserved. Deep Computing Capacity on Demand will service new commercial markets, such as drug discovery and product design, simulation and animation, financial and weather modeling and also a number of customers in market segments that have traditionally not been able to effectively access a supercomputer at a price within their budgets. The system enables customers to obtain a peak performance of 5.7 teraflops."

Comments

What goes around comes around.

[Ungrounded Lightning]

This is the same computing model as was used in large "computing centers" - such as those in universities - back in the 1950s-1970s:

The machine you need is too expensive to buy yourself and then leave sitting around idle most of the time (like a pencil sharpener). So an institution buys and sets one up, and you rent chunks of its time. If the demand goes up the institution gets more rent and can buy upgrades.

You get a machine fast enough to do your too-big-for-humans computing task in a short time (so YOU don't spend most of your time waiting to do YOUR next piece of work, like a pencil sharpener). You only pay for the amount you use.

Billing by CPU seconds, I/O volume, memory usage (fast and files), etc.

In the '50s you took your work to the machine, by the '60s remote terminals were becoming available, by the '70s packet-switching networks were making machines available across continents.

And also by about the '70s you were starting to see both comm and crunch becomming so cheap that, for ordinary jobs, accounting by the slice no longer made economic sense. Better use of money scattering (cheap) computers around and making them wait than only having a few and making (expensive) people wait.

Paying for comm by usage metering never caught on (too bursty, wastes human attention worrying about the effect on the bill, ...). Just buy the size of pipe you need to keep from being bottlenecked at peak load and leave it mostly idle. (You'd end up doing that by proxy anyhow - eliminate the middleman.) Client-server computing models moved institutions to a similar model for crunch and storage. General-user timesharing services gave way to networking services with unmetered shell accounts, which gave way to pure networking services, as the cheapening of computation evolved the personal terminal from a special purpose keyboard/display/comm box, first to terminal emulation on a dumb computer, then to one application on a progressively more powerful (though still small and cheap) computer functioning as a full-blown network node.

But there are still REALLY BIG jobs were the economics of a shared utility make perfect sense. IBM was once a primary provider of machines to such utilities within educational and business institutions. Now it's largely a business service provider. It seems approprate they should recognize the opportunity and use it as a way to make a profit by filling a gap at the high end of the computing market.