For those who need enormous amounts of computing power, the fact that corporate CPUs remain idle the vast majority of the time is always a difficult cross to bear.
The idea of linking systems to increase aggregate computing power has been around for many years but recently the technology has improved, allowing for easier implementation.
Sun Microsystems Inc. recently purchased GridWare Inc. to add compute farming to its list of offerings. The technology will also eventually be available for other platforms. Sun is renaming the technology the SunT Grid Engine.
Sun’s angle is slightly different from traditional offerings, according to Peter Jeffcock, product line manager in the Internet, desktop and server group at Sun in Menlo Park, Calif.
“It is free. So we have changed the economic rules,” he said. “Secondly, it is supported by a major system vendor or a partner of a major system vendor,” Jeffcock added.
“That is why we did this acquisition, because we wanted to change those rules so everybody could [create a compute farm],” he explained.
Tony Iams, senior analyst with D.H. Brown Associates in Port Chester, N.Y.,
sees another twist to Sun’s offering.
“They are slightly altering that positioning in trying to make [SunT] more of a free-form client that will run on more than just dedicated servers,” he said.
Iams said it expands the scope of CPUs available to tie into the system without requiring them to be dedicated to the compute farm.
“The whole idea is to take advantage of spare cycles that you have on workstations or PCs or pretty much any computing device that isn’t being used for some period of time,” he added.
Simple architecture
The SunT system maintains a list of available resources and a list of jobs, and matches the two up and keeps the system running efficiently, Jeffcock said. Architecturally, it is split into three components.
The execution host sits on every system that is part of the farm. The client software is designed to submit jobs to the system. Often the computer that an engineer is working at is also available to be used in the compute farm. So when he or she is not using the system, its CPU power is available to others. Since computers tend to be more occupied during the day, power sharing tends to be more prevalent at night.
The final part of the solution is the scheduler. There is only one of these and it works as the intermediary between the multiple clients and the multiple execution hosts, Jeffcock explained.
Today if you could count up the number of CPUs in the Sun technical computing market that are running this type of software you would probably find it is two to five per cent, Jeffcock explained.
“One of the primary reasons that people haven’t adopted [compute farm technology], as much as we would like them to, is cost. This stuff has either been open sourced, and difficult for them to figure out, or expensive. So what we are trying to do is change the rules here.”
The technology is designed for those who need intense computing power, not run of the mill. But surprisingly there is a wide array of potential uses, according to Jeffcock.
“Technical computing and people who do computationally intensive stuff,” is probably the biggest group Jeffcock said. This includes mechanical CAD rendering and mechanical computer-aided engineering in manufacturing and electronic design.
There is also a large potential market in digital content creation and rendering, he said. This includes everything from the entertainment industry and science and research to weather forecasting and simulating nuclear explosions.
Easy to install
If your environment is set up to farm, there is no special cabling or set-up required, Jeffcock said.
“At this level, I have to say, it is not rocket science,” he said. “It is not complex to set it up and make it work, the complexity will come if you really want to tune the heck out of this thing,”
Ira Chayut, director of verification at Cradle Technologies Inc. in Fremont, Calif., agrees it is easy to use, though he has been using the version shipped prior to Sun’s acquisition of GridWare. He said it has some advantages over Markham, Ont.-based Platform Computing Corp.’s rival LSF offering.
“It is very much lighter and easier to play with than LSF, and it is much more responsive,” Chayut said. He said that in the past he has used the LSF offering and brew-your-own solutions, but said they were always too expensive, either in terms of money or time.
Creating his own solutions were very costly in terms of time and LSF required, in his estimation, one full-time person to manage the system, Chayut added.
Cradle is using GridWare to conduct thousands of regression tests, which are run against the company’s latest hardware designs.
They have 20 CPUs on Sun Ultra 60s arranged in dual CPU mode, he explained.
Sun’s version of GridWare (www.sun.com/gridware) is now available for free download for the Solaris system and will be available on other platforms in the new year.
