[Beowulf] longevity of 1U servers?

[David Mathog]

Mark Hahn <hahn at mcmaster.ca>

> I'm not overly fond of soprano choral works either, but they're not
> _inherent_ to 1u-ism.  I have some 1u's with centrifugal fans that are in
> their, uh, probably 7th year of service.  jeez.

Are these on the CPU heat sink or in the chassis? I am currently
disassembling a rack which used these on the CPU:

http://www.dynatron-corp.com/products/cpucooler/cpucooler_model.asp?id=86

These were horrifically load 1U units, and I had thought it was mostly
because of the many itty bitty fans in the case.  Apparently, they were
not entirely at fault. One motherboard was recycled into a scavenged
desktop case for use testing disks. That test system emits a piercing 6
KHz note which is unbearable without ear plugs.  The fan is rated just
below 50 dBa, but subjectively, it is painfully loud, as if it emits all
of that sound energy in a very narrow frequency band.  My impression is
that these CPU fans were not especially reliable - I know the fellow who
ran it had to replace several of them.  (Possibly because things were a
bit toasty inside the dual motherboard 1U cases.)

Back to the original subject: is it my imagination or do none of the
system vendors specify the L10 or MTBF ratings for the case and CPU fans
they use?  At least, I can never seem to find that info on vendor web
sites.  Buyers really should not have to tear down a test system, or
specially request the information, in order to determine the expected
component lifetimes.  Yet case, PS, and CPU fans are the most likely
things to fail in a server, with the disks bringing up a close second.

Regards,

David Mathog
mathog at caltech.edu
Manager, Sequence Analysis Facility, Biology Division, Caltech