From: SERTKAYA BARIS (e1097757@ceng.metu.edu.tr)
Date: Thu 23 Mar 2000 - 11:56:13 EET
guzel bir beowulf vs IBM SP2 karsilastirmasi :)
Forwarded message:
> From owner-beowulf@beowulf.org Thu Mar 23 04:24 EET 2000
> Reply-To: <jrforsythe@uswest.net>
> From: "James R Forsythe" <jrforsythe@pop.clsp.uswest.net>
> To: "'Chris'" <charwel@chthry.chem.lsu.edu>,
> "'beuwulf list'" <beowulf@cesdis1.gsfc.nasa.gov>
> Subject: RE: cost comparison to ibm sp2...
> Date: Wed, 22 Mar 2000 17:38:01 -0700
> Message-ID: <000401bf9460$0c2229c0$0100a8c0@jimhome.usafa.af.mil>
> MIME-Version: 1.0
> Content-Transfer-Encoding: 7bit
> X-Priority: 3 (Normal)
> X-MSMail-Priority: Normal
> X-Mailer: Microsoft Outlook 8.5, Build 4.71.2173.0
> In-reply-to: <Pine.LNX.4.10.10003221429280.17437-100000@chthry.chem.lsu.edu>
> Importance: Normal
> X-MimeOLE: Produced By Microsoft MimeOLE V4.72.3110.3
> Sender: owner-beowulf@beowulf.org
> Precedence: bulk
> Content-Type: text/plain;
> charset="iso-8859-1"
> Content-Length: 3932
>
>
> About 6 months ago we priced a 30 processor beowulf vs. a 30 processor SP2.
> The cost was $50k compared to slightly over 1/2 million. The same was true
> for a 30 processor Origin2000. We went with the beowulf (duh), and found
> that the PIII-MhZ chips were as fast as an SP2 processor (195Mhz?).
> Fortunately for us, our code scaled at near 100% efficiency on 100Bt, so the
> beowulf gave us over a 10:1 price/performance ratio. If your code scaled
> poorly, you may need a faster backplane than 100Bt, which would increase the
> cost some, but the price performance ratio would still be very good. I'd be
> happy to let you know who sold us the beowulf in a private e-mail (I don't
> want to be accused of advertising since I'm in the Air Force). You can
> check out the benchmarks at:
>
> http://www.usafa.af.mil/dfan
> Click on the computational fluid dynamics link - you need Internet Explorer
> unfortunately. The SP2 benchmarks aren't on there - but a SP3 is.
>
> I don't have better price info for you on the SP2, since we never seriously
> considered the SP2 - we don't have that kind of money.
>
> Jim Forsythe
> Air Force Academy
>
> > -----Original Message-----
> > From: owner-beowulf@beowulf.org [mailto:owner-beowulf@beowulf.org]On
> > Behalf Of Chris
> > Sent: Wednesday, March 22, 2000 2:27 PM
> > To: beuwulf list
> > Subject: cost comparison to ibm sp2...
> >
> >
> > hello beowulf list!
> >
> > i'd like to enlist your help if you are willing...
> >
> > i am trying to make a price/performance or cost comparison between
> > a) a small, modest beowulf cluster for our chemistry dept.
> > and
> > b) an ibm sp2 type cluster in the "high performance
> > computing" area of the
> > universities computing services dept.
> >
> > i.e. compare:
> >
> > a_$/a_p and b_$/b_p
> >
> > for performance: (a_p and b_p) i should be able to run some
> > quantum chem
> > programs i normally use.
> >
> > for price: i know the price of a (a_$), but not b (b_$).
> > (at the moment i only mean hardware price not ttl cost of ownership).
> >
> > is there anyone with a few moments to spare and rough knowledge of ibm
> > parts pricing (or perhaps a catalog) that would hazard a
> > ball-park figure
> > for b? the hardware description is at:
> > http://lsumvs.sncc.lsu.edu/ocs/hpc/hardware.html
> >
> > i tried to rougly condense:
> > Adapters
> > Machine Arch Proc Mem Disk ATM/Mbps,Switch,SSA,Eth
> > # Type /MHz /MB /GB
> > ______________________________________________________________
> > ____________________________
> > 4 7015-R24 PWR2 71.5 256 4.5 155,HiPS Type
> > 2,SSA Enh.,Integrated
> > 7 9076-SP2 Thin PWR2 67 512 4.5
> > 155,HiPS Type 2,SSA Enh.,Integrated
> > 2 9076-SP2 Wide PWR2 77 512 2*4.5
> > 155,HiPS Type 2,SSA Enh.,H-Perf. LAN
> > 1 9076-SP2 Wide P2SC 135 256 4.5
> > 155,HiPS Type 2,SSA Enh.,H-Perf. LAN
> > 15 9076-260 2*PPC3 201 1024 2*9.1
> > 155,SP,switch,SSA Enh. RAID,10/100Mbps
> > 8 PPS 6015 PPC ??? 32 1.1 ---,---,---,IBM ISA
> > -----
> > 37 ttl
> >
> > <aside>
> > objectionably enough they term this ibm/sp2 cluster a beowulf
> > despite it using aix
> > and ibm hardware, see
> > http://lsumvs.sncc.lsu.edu/ocs/hpc/userguide.ht> ml.
> >
> > i have a
> > serious concern that people may pick up on the
> > term beowulf and
> > spin it to mean something far different than commodity-off-the-shelf
> > parts, COTS, & open source os. i realize there are some
> > subtle and good
> > distinctions that list members have made about the definition
> > of beowulf,
> > but i mean something far less subtle. i would be very interested in
> > others opinion about this. i'd also be interested in knowing how many
> > people would object to (b) being called a beowulf.
> > </aside>
> >
> >
> > thanks!
> > chris
> >
> > chris@chthry.chem.lsu.edu
> >
> >
> > -------------------------------------------------------------------
> > To unsubscribe send a message body containing "unsubscribe"
> > to beowulf-request@beowulf.org
> >
>
> -------------------------------------------------------------------
> To unsubscribe send a message body containing "unsubscribe"
> to beowulf-request@beowulf.org
>
Listeden cikmak icin:
unsub linux
mesajini listeci@bilkent.edu.tr'a gonderiniz.
Lutfen Listeci icin MIME / HTML / Turkce Aksan kullanmayin.
Liste arsivinin adresi: http://listweb.bilkent.edu.tr/
This archive was generated by hypermail 2b29 : Thu 23 Mar 2000 - 11:56:27 EET