[Beowulf] Oldest functioning clusters
Robert G. Brown
rgb at phy.duke.edu
Mon Nov 22 17:26:48 PST 2004
On Mon, 22 Nov 2004, Roger L. Smith wrote:
>
> During a conversation at the LECCIBG at SC'04 this year, I openly wondered
> where and what the oldest still-functioning cluster system is.
>
> So, who has the oldest cluster on this list? For my curiousity, I'm not
> limiting the submissions to Intel or Linux, so it doesn't have to be a
> traditional "Beowulf" system, but it should be a system designed and used
> exclusively as a cluster, and it should still be in service of some sort
> today.
You need a few more rules. Do we count a cluster as being in continuous
service even if its hardware has rolled over several times in its
existence? If yes, brahma has been running continuously since the fall
of '96, although the last of its original systems (200 MHz PPros) have
been dead for a year or three now. I have a mini-cluster at home with
400 MHz components -- maybe six years old -- although truthfully only
one system is still powered up as part of my current home cluster as I
got tired of paying for the power and they were making funny noises and
what CAN you do with only 96 MB of memory these days? Otherwise, we
tend not to run hardware much past four years or our systems persons get
Angry -- that TCO thing as maintenance efforts and the cost of power
rise.
Besides, one can show pretty easily that one can get more work done with
a single system that costs only what the differential POWER to run a
cluster of 8 five year old systems costs -- Moore's Law is brutal. So
production clusters older than five (really three, but certainly five)
are a net loss and make negative sense. Even my home cluster gets
painful to pay for out there and it only exists for prototyping and
column writing and play. Upgrading the OS eventually is just
impossible.
rgb
Robert G. Brown http://www.phy.duke.edu/~rgb/
Duke University Dept. of Physics, Box 90305
Durham, N.C. 27708-0305
Phone: 1-919-660-2567 Fax: 919-660-2525 email:rgb at phy.duke.edu
More information about the Beowulf
mailing list