Grid computing from Sun
Open Source and proud of it
Sun's open source Grid Engine software currently runs over 7000 grids with an average of 47 CPUs per grid, and a steadily increasing percentage of the grids it powers run Linux, according to John Tollefsrud, Sun's Grid marketing manager.
This is not a Sun software product that gets as much press or visibility as Java, and the 21,000+ Grid Engine download total (as of November 29, 2002) hardly compares with the hundreds of thousands of downloads per month the Sun-sponsored OpenOffice project boasts. But in some ways, in the long run, Grid Engine may be as influential as these other, higher-profile projects.
There's also a commercial Grid Engine Enterprise Edition available for companies that want to spend money on the software (and on support contracts) instead using the free, open source version which, according to Tollefsrud, runs exactly the same code.
The product was originally proprietary, but Sun released it as open source two years ago. "Here's commercial grade software that used to be sold as a revenue product, now open source but still being developed by Sun," says Tollefsrud.
Sun's grid computing product is designed to be used primarily in a department-sized corporate environment. It runs a daemon on each (Unnix or Linux) computer in your network, and when you have a computationally intensive task to perform, it grabs unuused CPU cycles from the appropriate computers and uses them. You can set parameters for each task, like "Linux machines only" or "Only use this group of computers between 'X' and 'Y' hours," or whatever else suits your fancy.
The point is to make sharing computational resources across a grid not only efficient, but easy. Tollefsrud points with special pride to one user, Axyz Animation, Inc. because, he says, "They're not Disney or Lucas, just a small shop that needs to concentrate on animation, not worry about the computing behind it."
Increase Resource Utilization
Tollefsrud says Sun has about 6000 CPUs set aside for chip design, and "these CPUs run over 90% usage, while average CPU utilization is in single digits."
(The Linux computer on which this story is being written is currently averaging below 5% CPU usage.)
Obviously, the more "bang" you can get out of each CPU you own, the better. The savings from efficient CPU use are not obvious -- and may not even be measurable -- in most home or office environments, but when you get into large-scale technical computing or graphics rendering they add up quickly. According to Tollefsrud, the financial service industries are jumping all over grid computing; they have boxes that work hard at transaction-type tasks during office hours, but do little or nothing the rest of the time. Why not use that "down" time to work on portfolio analyis and other tasks that take large calculations, but don't need to be completed in real time but can wait for an overnight run?
Sorry, we don't do Windows
Grid Engine Enterprise Edition is available for Solaris and Linux. Since these are the two operating systems Sun sells, this should not come as a surprise. The open source version has been adapted by users for just about all known Unix variants, including Mac OS X. There has been some discussion about doing a Windows port -- not officially sanctioned by Sun -- but so far it is just at the "talk" stage. Right now, if you want to build a an "office network by day, supercomputer by night" computer array, you are going to have to use Linux or Unix or some combination of the two.
This is yet another argument for Linux on the desktop -- especially in financial and science-type companies.
Not quite "plug in" worldwide grid computing yet
That's a nice dream, and a lot of people have written interesting academic papers about how it's going to change the world when it finally happens, but there are a lot of steps that need to be taken between now and then. Security is one of the biggest problems, and is likely to remain one for more than a few years to come.
But this is today, and Sun's Grid Engine is here today, concentrating on department-level computational resource sharing, complete with algorithms to calculate which users are using the grid most and which ones are contributing the most resources to it, a feature needed by corporate and government bean counters who decide what percentage of which computing budget needs to come from which group or department.
Grid computing is getting easier all the time
Not long ago -- just a few years -- setting up computer clusters and grids was a rough job. Now, with software like Sun's Grid Engine and other packages available, it is not an overly intimidating process, at least for professional sysadmins.
"The people that have deployed most of these grids [using Sun Grid Engine]... they didn't have Sun come out and do it for them," says Tollefsrud. "They did it themselves. It's not that hard." © Newsforge.com.
Sponsored: Benefits from the lessons learned in HPC