There’s no shortage of computing power in the computer science
department at Yale University, in New Haven, Conn. Along with
assorted high-powered workstations and other computers, the
department has two Linux-based Beowulf clusters. But taking
advantage of the computing power in those clusters hasn’t always
been easy.
Until recently, students and professors who wanted to harness
the processing power of the Beowulf clusters — one of which uses
20 Dell servers, the other 20 IBM Netfinity systems — had to
modify the programs they were using to run in parallel. That, says
Prof. Martin Schultz, of Yale’s computer science department,
requires having access to the program’s source code, and using a
low-level tool such as Message Passing Interface (MPI) to modify it
— not always a simple task. “MPI has over three hundred commands,”
Schultz says. “It’s not easy to master.”
Thanks to Turbolinux’ Enfuzion, which Yale is now running on
both Beowulf clusters, that’s no longer the case. Enfuzion, says
Schultz, is easy to use, because it takes care of distributing
programs across the various nodes in the cluster, so they run
without having to be modified. Researchers and students at Yale can
now take advantage of the super-computing level of power provided
by the clusters, without having to do any special programming.
The importance of that “can not be overestimated,” says Schultz.
“If scientists have to spend time making changes to their software
program, it can take a long time to get their research up and
running. When you’re competing with other researchers, you don’t
want a system that takes a long time to use, you want to crank out
results.”
Now that it’s possible to run programs on the Beowulf clusters
without spending lots of time to modify them for parallel
computing, Schultz plans to assign projects involving cluster
computing to his classes. Several dozen other professors and
graduate students, both in and out of the computer science
department, will be using Enfuzion to run programs as well.
Eventually, Schultz expects lots of departments on campus to use
the clusters. For example, the Yale medical school and the
university’s biology departments do a lot of
computationally-intensive research. “I expect Enfuzion will play a
big role there,” Schultz says. Since Enfuzion can easily manage
clusters consisting of thousands of computers, Schultz anticipates
no problems expanding the clusters to handle the load.
But the clusters may not have to grow at all. Yale is taking
advantage of Enfuzion’s unique ability to harness idle CPU cycles
on computers outside the cluster. The university will use Enfuzion
to distribute programs to 20 other computers, which sit on the
desks of students and professors in the department. Enfuzion’s load
balancing facilities will automatically run programs on these
systems when they’re not otherwise in use. “If you don’t use them,
these CPU cycles disappear,” say Schultz. “It’s not a resource you
can store up. But with Enfuzion, these machines will be in use
nights and weekends, or whenever someone is not actually sitting at
the keyboard using them. It lets us make maximum use of our
computing resources.”
About Turbolinux EnFuzion
EnFuzion clusters all available computing resources on a
corporate network to create a powerful “virtual supercomputer” and,
as a result, allows companies to reduce time and costs associated
with computationally demanding data processing jobs. Traditionally,
these jobs – such as complex financial calculations – have been
handled by expensive high-end servers. With the growing need to
process increasing volumes of complex jobs in a shorter time
period, the cost of traditional solutions becomes prohibitive. To
learn more about EnFuzion please visit www.turbolinux.com.