cooling

Robert G. Brown rgb at phy.duke.edu
Wed Apr 24 09:16:17 EDT 2002


On Tue, 23 Apr 2002, Jon Mitchiner wrote:

> The other consideration to have is some kind of monitoring/alerting system
> for the room.  A client has a dedicated cooling equipment for a beowulf
> cluster for 52 machines.  Recently the A/C broke one morning and they did
> not find out till the afternoon when someone walked into the small network
> room and found the room was in excess of 100 degrees.
> 
> I dont want to think about what could have happened if it happened on a
> friday evening and nobody found about it until Monday. :)

A very good idea is a thermal kill switch on the master power panels,
mentioned in my previous reply.  If room temperature hits a preset FOR
ANY REASON, all nodes go down, period.

Strategically, one would still want to monitor node and room temperature
and install alarms and automated node shutdown scripts as previously
mentioned in many discussions, but set THOSE alarms and shutdowns to go
off at (say) 25C and 30C and set the room kill switch at (say) 35C.
That way if cooling fails, first you get mail/pages/human alarms (at
25C, assuming room temperature is set to 20C and ordinarily stays there
+/- 2C), then at 30C (or when CPU temps pass a preset alarm that goes
off when ambient room temperature gets about there) nodes start shutting
themselves down, and only if the humans and shutdowns fail to control
the temperature or get the AC running again and the temperature keeps
climbing does the room kill go off.

This protects the systems against ANY POSSIBILITY that they will operate
for an extended time at a "dangerous" temperature.

Conservative people, or people with evidence that the properly
functioning room has a very stable temperature profile, could reduce the
alarm margins even further -- 35C is already quite a bit hotter than one
wants to let the ambient air reach.  30C would be better, but it doesn't
leave much room for less intrusive alarms and scripts to take effect.

   rgb

> 
> Jon Mitchiner
> 
> ----- Original Message -----
> From: "Bob Drzyzgula" <bob at drzyzgula.org>
> To: "Robert B Heckendorn" <heckendo at cs.uidaho.edu>
> Cc: <beowulf at beowulf.org>
> Sent: Tuesday, April 23, 2002 9:54 PM
> Subject: Re: cooling
> 
> 
> > If the new load requires the installation of new
> > chillers, it could indeed cost a pile-o'-money. Even
> > if each node burned electricity at 100 Watts, you
> > are looking at 50 kW of power consumption, or about
> > 170,000 BTU/hr, requiring about 14 tons of cooling to
> > remove -- your facilities folks may well be looking
> > at installing something like one or more Liebert
> > chillers such as these:
> > http://www.liebert.com/dynamic/displayproduct.asp?id=545&cycles=60Hz
> >
> > There could well be additional shortfalls in external
> > heat exchanger capacity, pipe capacity out to the
> > heat exchangers, electric power for the computers
> > and for the chillers, etc. If you don't already
> > have the raised floor space, that could also add
> > quite a bit to the cost to cool all those nodes.
> >
> > As to how we are making the A/C for our systems "affordable",
> > we do it by virtue of the HVAC budget belonging to
> > a different division, :-) although that also means
> > that we don't have *control* over that budget, and
> > when we hit the ceiling on cooling we kind of have
> > to just stop installing new equipment until the whining
> > and begging and pleading might eventually get us
> > a new chiller -- and even then we might have to give
> > up some rack space so there'd be a place to put it. :-(
> >
> > --Bob
> >
> > On Tue, Apr 23, 2002 at 05:41:35PM -0700, Robert B Heckendorn wrote:
> > >
> > > We are looking at the facilities issues in installing a beowulf on the
> > > order of 500 nodes.  What facilities is telling us is that it is going
> > > to almost cost us more to buy the cooling for the machine than to buy
> > > machine itself.  How are people making the air conditioning for their
> > > machines affordable?  Have we miscalculated the HVAC loads?  Are we
> > > being over charged?
> > >
> > > thanks for any guidance.
> > >
> > > --
> > > | Robert Heckendorn                        | We may not be the only
> > > | heckendo at cs.uidaho.edu                   | species on the planet but
> > > | http://www.cs.uidaho.edu/~heckendo       | we sure do act like it.
> > > | CS Dept, University of Idaho             |
> > > | Moscow, Idaho, USA   83844-1010          |
> > > _______________________________________________
> > > Beowulf mailing list, Beowulf at beowulf.org
> > > To change your subscription (digest mode or unsubscribe) visit
> http://www.beowulf.org/mailman/listinfo/beowulf
> > _______________________________________________
> > Beowulf mailing list, Beowulf at beowulf.org
> > To change your subscription (digest mode or unsubscribe) visit
> http://www.beowulf.org/mailman/listinfo/beowulf
> >
> 
> 
> _______________________________________________
> Beowulf mailing list, Beowulf at beowulf.org
> To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf
> 

-- 
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



_______________________________________________
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 mailing list