mathog at mendel.bio.caltech.edu
Thu Oct 31 16:05:05 EST 2002
I plugged 5 single CPU Tyan 2466 machines into a single
TrippLite 6 Isobar Ultra. Circuit was 20A breaker, 20A wiring,
15A plug. The 5 nodes didn't blow the 12A fuse in the
surge suppressor. It ran ok for an hour with "openssl speed"
running in a loop on all nodes. These suppressors may have
too small a breaker if we add second CPUs to each node but
for now they seem to be able to handle the current. One node
was the master and it had to be up first so that the others
could boot. I turned the power on on the other 4 nodes within a
1 second interval and that didn't trip the breaker either.
I've been hunting around for alternative strips and it seems
that nobody makes a surge supressor that fits mye specs well:
1. 6 outlet (well, <8, only 5 nodes are going in per strip)
2. 15A breaker
3. >2000 Joule protection
4. Metal case
5. Mounting holes in the metal case (for a zeroU mount)
6. Surge protection failure disconnects load.
Strips either go way over (like a TrippLite or APC 12 socket
rackmount strip) or a touch under (with the surge supressor tested
above.) Intermatic makes some nice metal strips of the right
size - but no mounting holes and not enough Joules.
They make some other strips with the right electrical
characteristics but in plastic cases with no mounting holes
Ok, the 2000 Joules is a bit arbitrary. My
experience with Surge suppressors has been that those with smaller
ratings (400 Joules or less) fail quickly but the Isobar, with a 2600
Joule rating, works for many years without a problem. In fact, I've
not seen one fail yet. So lifetime
is somehow related to Joules, and I'm only guessing that 2000 is
high enough, since 2600 was ok. Maybe 1200 Joules would last as
long. I'm not going to do the experiment to find out, and good luck
finding test results for surge suppressors.
mathog at caltech.edu
Manager, Sequence Analysis Facility, Biology Division, Caltech
Beowulf mailing list, Beowulf at beowulf.org
To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf
More information about the Beowulf