From cclausen at uiuc.edu  Sun Jun 10 13:33:46 2007
From: cclausen at uiuc.edu (Christopher D. Clausen)
Date: Sun, 10 Jun 2007 13:33:46 -0500
Subject: [cluster-l] SMP and Standard Multi-Processing on a singlecluster
References: <7.0.1.0.2.20070507100205.00f57838@uiuc.edu><Pine.LNX.4.63.0705071055390.12721@paris.ks.uiuc.edu>
	<Pine.LNX.4.63.0705071058490.12721@paris.ks.uiuc.edu>
Message-ID: <59D4557755514742BF85E9C4DBF80CC9@CDCHOME>

Am I correct in seeing that Cluster OpenMP is a commercial product that 
isn't freely available?

Is there a cluster package that allows a fork / exec to automatially 
migrate to remote machines as well as MPI code to run on the machines? 
(Not neccessarily simultaneously.)

We've purchased 9 Intel Quad core machines for: 
http://www-illigal.ge.uiuc.edu/ and I'm wondering the best thing to run 
on them.  (Quad cores were purchased b/c the developers stated that 
their code scaled nearly linearly with the number of cores on the 
machine.)  We currently have some gentoo based openmosix clusters and 
I'd like to move away from that.

<<CDC

Jim Phillips <jim at ks.uiuc.edu> wrote:
> A little more information:
>
>   http://www.hpcwire.com/hpc/658711.html
>
>   http://www.intel.com/cd/software/products/asmo-na/eng/329023.htm
>
> -Jim
>
> On Mon, 7 May 2007, Jim Phillips wrote:
>> Cluster OpenMP sounds like what you want:
>>
>> http://access.ncsa.uiuc.edu/Briefs/06Briefs/2006-03-23NCSA_to_h.html
>>
>> I assume you can use it with whatever cluster toolkit you like.
>>
>> On Mon, 7 May 2007, Nils Oberg wrote:
>>> Hello,
>>>
>>> Is there a cluster toolkit that will allow both the cluster to act
>>> as an SMP system as well as a standard Beowulf-like cluster?  I'd
>>> like to be able to run jobs that are compiled with an OpenMP-enabled
>>> compiler as well as jobs using MPI over rsh or ssh.  (By cluster
>>> toolkit I mean something like Warewulf, Clustermatic, or OpenMOSIX.)
>>>
>>> Thanks,
>>>
>>> Nils 



From cclausen at uiuc.edu  Sun Jun 10 13:41:34 2007
From: cclausen at uiuc.edu (Christopher D. Clausen)
Date: Sun, 10 Jun 2007 13:41:34 -0500
Subject: [cluster-l] best way to use Kill-A-Watt?
Message-ID: <F9A23A9F47034D62B09DC7E86CEF617F@CDCHOME>

As suggested at the workshop, I've purchased a Kill-A-Watt meter 
(newegg.com sells them cheaper than thinkgeek.com)  What is the best way 
to get an acurate reading?  And which numbers should I be concerned 
with?  The VA number?  Is it best to take a reading over a period of 
hours? minutes?  What can I run to ensure that the CPU is running at 
full load?

<<CDC
-- 
Christopher D. Clausen 



From cclausen at uiuc.edu  Sun Jun 10 13:50:49 2007
From: cclausen at uiuc.edu (Christopher D. Clausen)
Date: Sun, 10 Jun 2007 13:50:49 -0500
Subject: [cluster-l] 4 / 5 shelf racks (non-19") ?
Message-ID: <4FA4DE9083784621928CB53188EEEB58@CDCHOME>

I know I asked about this before, but I can't find my notes...  What 
vendor did you guys use for those 4 shelf racks?  Any pointers to an 
online vendor?  Perferrably something with wheels (if that doesn't cause 
things to tip over.)

You should see the three high pyramind style stacking in this room right 
now.  Its pretty scary.

<<CDC
-- 
Christopher D. Clausen 



From tskirvin at killfile.org  Wed Jun 20 15:56:51 2007
From: tskirvin at killfile.org (Tim Skirvin)
Date: Wed, 20 Jun 2007 15:56:51 -0500
Subject: [cluster-l] best way to use Kill-A-Watt?
References: <mailman.32.1181500908.2933.cluster-l@ks.uiuc.edu>
Message-ID: <tskirvin.20070620205653$7bea@cairo.ks.uiuc.edu>

"Christopher D. Clausen" <cclausen at uiuc.edu> writes:

> As suggested at the workshop, I've purchased a Kill-A-Watt meter 
> (newegg.com sells them cheaper than thinkgeek.com)  What is the best way 
> to get an acurate reading?  And which numbers should I be concerned 
> with?  The VA number?

        The VA number is probably your best bet, yes. 

> Is it best to take a reading over a period of hours? minutes? 
        
        You're probably looking for a 'peak' usage, so a short time period
should be fine.

> What can I run to ensure that the CPU is running at full load?

        'cpuburn' is a good choice.  

                            - Tim Skirvin (tskirvin at ks.uiuc.edu)
-- 
   Theoretical and Computational    http://www.ks.uiuc.edu/~tskirvin/
Biophysics, Beckman Institute, UIUC    Senior Systems Administrator

From jim at ks.uiuc.edu  Wed Jun 20 16:40:52 2007
From: jim at ks.uiuc.edu (Jim Phillips)
Date: Wed, 20 Jun 2007 16:40:52 -0500 (CDT)
Subject: [cluster-l] best way to use Kill-A-Watt?
In-Reply-To: <tskirvin.20070620205653$7bea@cairo.ks.uiuc.edu>
References: <mailman.32.1181500908.2933.cluster-l@ks.uiuc.edu>
	<tskirvin.20070620205653$7bea@cairo.ks.uiuc.edu>
Message-ID: <Pine.LNX.4.63.0706201640440.16834@paris.ks.uiuc.edu>

On Wed, 20 Jun 2007, Tim Skirvin wrote:

> "Christopher D. Clausen" <cclausen at uiuc.edu> writes:
>
>> As suggested at the workshop, I've purchased a Kill-A-Watt meter
>> (newegg.com sells them cheaper than thinkgeek.com)  What is the best way
>> to get an acurate reading?  And which numbers should I be concerned
>> with?  The VA number?
>
>        The VA number is probably your best bet, yes.

If you care about heat output, use Watts (power).  If you care about 
overloading circuits, use Amps (current).  If you care about overloading a 
UPS, use VA (Volts * Amps without power-factor correction).

>> What can I run to ensure that the CPU is running at full load?
>
>        'cpuburn' is a good choice.

Be sure to run one copy of cpuburn per *core*.

-Jim

From tskirvin at killfile.org  Wed Jun 20 16:00:14 2007
From: tskirvin at killfile.org (Tim Skirvin)
Date: Wed, 20 Jun 2007 16:00:14 -0500
Subject: [cluster-l] 4 / 5 shelf racks (non-19") ?
References: <mailman.33.1181501616.2933.cluster-l@ks.uiuc.edu>
Message-ID: <tskirvin.20070620205757$52ab@cairo.ks.uiuc.edu>

"Christopher D. Clausen" <cclausen at uiuc.edu> writes:

> I know I asked about this before, but I can't find my notes...  What 
> vendor did you guys use for those 4 shelf racks?  

        www.buyrack.com

        I like 'em.  You're probably looking at the 'mobile wire shelving'
there.  An example:

        http://www.buyrack.com/blk_mobile_standard.asp

        The 80" high, 4-shelf units sound like what you'll want.  The
shelf size is up to your needs, of course.  It looks like they go for
$150-$250 a piece; and as I recall, shipping was bad but not prohibitive.

        You may also want to look at Sam's Club; they've got some similar
wire shelves for ~$100, which I use at home.  I doubt they've got the same
weight capacity, but they'd probably do the job, and that would eliminate
the shipping cost.

                            - Tim Skirvin (tskirvin at ks.uiuc.edu)
-- 
   Theoretical and Computational    http://www.ks.uiuc.edu/~tskirvin/
Biophysics, Beckman Institute, UIUC    Senior Systems Administrator