Ottawa-based GridWay Computing Corp. recently launched a utility-based service in an effort to provide silicon chip design companies with the extra computing power they might require during simulation testing and verification phases of chip design.
The grid computing service enables the data-heavy customers to have access to additional computing power during peak testing periods, without having to incur capital investment costs associated with owning 20 to 40 servers.
With 40 Sun Fire V60x servers running Red Hat Linux on its grid right now, GridWay Computing has created its own server farm, said Chris Kramer, founder of GridWay Computing.
Sun Microsystems of Canada Inc. provided the server environment and Telecom Ottawa Ltd., a high bandwidth broadband data utility, ensures that the utility grid has high speed bandwidth connectivity over a 10GB Ethernet network that spans from Quebec City to Toronto, Kingston, Ont. and Cornwall, Ont.
“The sexy part about the engine we’re building is the compute capacity on demand, so we have high-end processors performing in a grid environment,” Kramer said.
Kramer said the primary focus of the utility model is to extend computing environments for chip design companies during simulating testing cycles. When a chip design company gets close to sending out a chip for fabrication, it requires a large number of servers to complete all its verification tests.
Having access to as much computing power as possible during peak periods of chip design testing is something that is attractive for Tundra Semiconductor Corp., an Ottawa-based company that makes semiconductor products for communications, server and storage markets. Tundra was the first company to start buying extra computing power from GridWay.
Until Jan. 1, 2004, Tundra had always bought, owned and managed its own computing farm, said David Lisk, vice-president of operations at Tundra.
“To design the new chips coming out, we would’ve had to double or triple the size of our compute farm,” Lisk said.
Not only are the servers costly, but they run constantly and aren’t used 24 hours a day. By having the ability to offload peak demand onto the GridWay, Lisk said Tundra can completely load its own compute farm, as well as having the advantage of double or triple the number of processors it is using to conduct testing.
“We could not afford to put in the three times the number of servers into our farm and then just let them sit idle and depreciate for eight months out of the year,” Lisk said.
The chip design phase is similar to software design, Lisk explained. There are lots of people that write chip applicable code that gets compiled and is run on these servers. Then a series of random tests are run against the software to detect as many bugs as possible.
“The more computers you have, the better chance you have at finding all the bugs,” Lisk said. “The advantage of GridWay is as the design starts to mature, you come very close to wanting to convert that software replication of the chip into a real chip. At that point, you want to run as much verification as you can. GridWay allows us to rent extra computing during those crunch periods.”
Tundra can now afford 200 to 300 per cent computing power in that critical time zone, instead of just being able to afford 100 per cent computing, he added.
“It allows us to have a better product, which is a delight to our customers because there are less bugs,” Lisk said.
So far it’s been a successful arrangement and Lisk said that while it’s still early, it could turn out to be so good that Tundra would consider reducing its direct investing in computing resources and rent grid computing time permanently.
By allowing companies to “dial-up and dial-down” computing power as they need it, those companies can avoid incurring capital acquisition costs of building a compute farm, only to do testing that is only used for 20 per cent of the farm’s lifecycle, GridWay’s Kramer said.
Pricing for computing power varies and depends on capacity. It costs about $150 per week, per processor with a one week minimum. Per minute fees are something GridWay is working on developing down the road.
An example of the cost is if a company were to use an additional 20 processors, it would cost over $120,000 in hardware to purchase. To rent the space at GridWay, it would cost $3,000.
Kramer said a typical chip lifecycle would require additional processing power twice a year, anywhere from a two week to a two month window.
Along with the high-end computing power on demand, GridWay also offers data availability and hosting services.
Kramer said he expects several more companies to come on board within the next several weeks. Also, because Ottawa, Montreal and Toronto are home to many of the country’s silicon design companies, there is a large base of potential customers.