The Ohm Project: Frequently Asked Questions |
Why is your cluster named Ohm?
Our cluster and project are named after Georg Simon Ohm, the 19th-century mathematician and physicist who discovered Ohm's Law. (The machines in Calvin's Department of Computer Science are named after individuals who made important contributions to Computing and related disciplines.) It has been rumored that Ohm is also an acronym for our hypercube multiprocessor.
What hardware/software does Ohm use?
For Ohm's hardware, see our hardware configuration
page.
Originally,
Scyld OS
provided most of Ohm's system software,
which includes cluster-management utilities,
as well as parallel-processing libraries like
MPI
and PVM.
When Scyld changed the documentation and pricing policy,
we switched to a customized version of
Redhat Linux.
Near the end of the project, we switched to a customized version of
Gentoo Linux,
to simplify administration, updates, and maintenance,
as well as to ensure that all of our installed software
was optimized as much as possible for our hardware.
Students write their own application software,
and faculty researchers may use open-source or commercial software
(e.g., Gaussian,
gridMathematica),
depending on their research needs.
What communication topology does Ohm use?
Ohm's communication topology will vary over the three years of the project.
In the first year, Ohm's topology will be
a "star" topology with a gigabit switch at the center of the star.
In the second year, Ohm's topology will be
a "ring-star hybrid", with the central switch being augmented by
the additional links needed to form a ring.
In the third year, Ohm's topology will be
a "hypercube-star hybrid", with the central switch being augmented by
the additional links needed to form a hypercube.
Our
Ohm Performance Applet provides a summary of our performance/price findings
for each of these three topologies using 2001 prices
(and lets the user enter today's prices to see how the performance/price ratio
has changed).
What is a hypercube?
Check out our Hypercube
Applet page for a quick overview.
How much did Ohm cost to build?
This is hard to answer, because Ohm is evolving over the
three years of the project.
We spent just over $76,000 the first year to build Ohm in its
initial configuration.
Our basic hardware configuration
was not all that expensive, so what cost so much?
In a nutshell, infrastructure, and "doing it right":
To some extent, some of these expenses were luxuries,
and could have been avoided if we were willing to have a slower,
less-convenient-to-use, and more-error-prone cluster.
However, taking the long view, spending the extra $$ at the
outset should pay off down the road, as upgrading Ohm
will not be as expensive as would be the case otherwise.
In year 2, we bought an additional thirty-four Syskonnect NICs
($450 each), plus the necessary cross-over cables
to add a ring atop our star topology.
In year 3, we bought an additional thirty-four Syskonnect NICs (~$430 each),
plus the necessary cross-over cables
to extend the ring atop our star topology into a hypercube.
With leftover funds, we purchased a
1.44TB IDE-to-SCSI RAID array
and a dedicated 2-CPU file-server node to manage it from
Western Scientific
for roughly $7000,
plus two more Syskonnect gigabit NICs to connect the new file server
to our network.
How fast is Ohm?
With 1 master and 16 slaves running the Linpack benchmark,
Ohm's performance has been measured at 10.4 Gigaflops.
What kind of problems is Ohm being used to solve?
Ohm is being used to solve problems from a variety of disciplines, including:
Back to the
Ohm Home Page
This page maintained by
Joel C. Adams.