Dual CPU nodes?
Robert G. Brown
rgb at phy.duke.edu
Mon Oct 21 10:41:38 PDT 2002
On 21 Oct 2002, Dean Johnson wrote:
> On Mon, 2002-10-21 at 10:28, Robert G. Brown wrote:
> > A beefy UPS is not a bad thing to
> > consider if you can afford it.
> Actually, for all but the most critical applications ("critical" as
> defined by you), even a crappy little one is a *really* good idea. Most
> places have a good enough power grid that having a UPS to save you from
> the first second of a power outage will satisfy 99% of stability needs.
> Its the little brownouts and drops that, IMHO, make up most of the
> problems. Actually the biggest problem with power outages is trying to
> get my heart back into a regular rhythm after suddenly having a pile of
> UPSes squeal underfoot. ;-)
Agreed, for precisely these reasons. By "beefy" I just meant big enough
to support all your nodes for anywhere from a few seconds to a few
minutes. In Durham on the commercial grid we actually have the
one-second outages that crash your system every two to four months
(yuk!). We ALSO have five minute to an hour long outages due to e.g.
thunderstorms once or twice a year (and every few years have a hour to
days down due to hurricanes or serious ice storms:-(.
In my home cluster, I use UPS that a) can provide the requisite current
with a nice fat margin for b) fifteen to thirty minutes, at least on the
server(s). That gives me time to shut the server and some of the nodes
down "nicely" if the power fails to come back up in the five or ten
minutes it takes me to get to my desk and initiate a shutdown. I've
been thinking of getting the el cheapo "five minute" UPS's to handle
nodes that double as workstations in my kids' rooms (some are as cheap
as $50 these days) to handle the 1 second outages and do a better job of
power conditioning at about the same cost as a high end surge strip.
Of course, UPS are getting a lot cheaper than they were some years ago.
It is pretty easy to get UPS that can hold order of 0.5-1 KW-hour for a
couple hundred bucks. I could probably leave my core systems running
through most of the longer outages with one of those, if I could stand
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