Re: My Intention for New Suggestion for so called "www.beowulf-top500.org"

["윤재호" <yoon@xxxxxxxxxxxxxxxxxx>]

Happy lunar New Year !

Let's think about making the so called [top500 rank for beowulf].

There may be many criterion to compare the hardware or software efficiency.

Linpack test is one of them. but it has benefit to compare with the www.top500.org 

Supercomputers.
 
NPB test is O.K. 

but, In order to rank a machines, we may choose one test. and other tests will be as a 

reference for that Beowulf Computers. 

but If we use price/performance as a criterion to compare beowulf,

That criterion will be only commercially comparison, not the software or hardware efficiency.

Think about my case, I setup beowulf in the Korea University computer room. 

-- That means it costs almost nothing.

So price/performance is not a good criterion. 

Why do you think I suggest like this so called "www-beowulf-top500.org" ?

I think in order to develop our beowulf computers, I think we need fair competition.

In order to compete for the better bewoulf computers, we need compare the beowulf computers.

I think, To compare the beowulf computers with same criterion is one way to deveolp our 

beowulf computers.  

I agree Dr. Robert G. Brown to setup a kind of database for the beowulf.

but the database needs a kind comparison criterion -- Linpack test for example, -- to develop

& campare our bewoulf supercomputers.

Let's think about 10 years later.

Then, we will find our beowulf supercomputer has developed so far from fair comparison each other.

From Yoon Jae Ho
Economist
POSCO Research Institute.
Seoul, Korea.
http://ie.korea.ac.kr/~supercom/

Imagination is more important than Knowledge.  --- A. Einstein.  

P.S : I have not so much time to spend to test.

because, My job is not related beowulf supercompters until now.

My job is forecast Korea GDP, Price, Interest rate & analysize the Company result 

& economic consulting..

So busy for my job.

That means for this beowulf computers, It just hobby until now. 



-----원본 메시지-----
보낸 사람: Robert G. Brown <rgb@xxxxxxxxxxxx>
받는 사람: Douglas Eadline <deadline@xxxxxxxxxx>
참조: 윤재호 <yoon@xxxxxxxxxxxxxxxxxx>; Walter B. Ligon III <walt@xxxxxxxxxxxxxxxxxxxx>; beowulf@xxxxxxxxxxxxxxxxxxxxx <beowulf@xxxxxxxxxxxxxxxxxxxxx>
날짜: 2000년 2월 5일 토요일 오전 8:14
제목: Re: My Intention for New Suggestion for so called "www.beowulf-top500.org"


>On Fri, 4 Feb 2000, Douglas Eadline wrote:
>
>> What if a web-page of application specific benchmark data were 
>> developed. I think there are a handful of applications that many
>> people would be interested in seeing:
>> 
>> 1) how well they performed on a Beowulf
>> 2) how the machine was constructed
>> 3) What the actual cost of the machine was
>> 
>> Generating lists of real application performance
>> (both price/performance and best performance) would be quite
>> a bit more valuable than a list of 500 Beowulf machines
>> running a single code. (which bye the way could be included in
>> the suite of applications)
>> 
>> In any case, I support any effort that furthers the Beowulf
>> Community and if a group of people are empowered to put up a 
>> top 500 list then I would not want to stand in the way.
>
>I think that I'm in pretty good agreement with Doug on this.  On the one
>hand, there are a lot of things about the list that would be "fun", but
>rather than rank ordering the list on the basis of any single criterion
>(even one that contains input from many performance indicators), I'd
>rather see a list that contains data more like that which Doug suggests,
>prepared in a tabular form that can be sorted or searched according to
>preference.  I might be interested in seeing what the cost per machine
>was of all the beowulfs in the list that contained more than 64
>processors, for example.  I might then want to further sort this list by
>speed on some particular application, or (better!) on the basis of
>price/performance on the application.  Finally, I might want to be able
>to follow a link to a detailed description of the machine or its
>website.
>
>All of this could really be done on the main beowulf.org website now --
>it doesn't really need a separate site, and proliferation of associated
>sites just makes it more difficult to find things when you need them.
>The problem is one of time -- building ANY of the described sites will
>require time, and time is money (or costs money, which is the same
>thing).  I'm not volunteering because I certainly don't have time, and
>of course those folks who do volunteer their time can and should follow
>their own vision of the thing as that is the thing that will sustain
>them.
>
>Just my druthers.  I'druther have a much less "in your top-500 face"
>list/database that collected the same information that is being proposed
>and presented it in a judgement-neutral way that allows the viewer to
>choose the presentation that emphasizes what is important to THEM.  The
>design, of course, should be extensible (so new fields/values can be
>easily added) and all tools and rules for making the benchmark or other
>measurements must be distributed by the site.  Not unlike the way SPEC
>does it, if you like.
>
>   rgb
>
>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@xxxxxxxxxxxx
>
>
>
>-------------------------------------------------------------------
>To unsubscribe send a message body containing "unsubscribe"
>to beowulf-request@xxxxxxxxxxx
-------------------------------------------------------------------
To unsubscribe send a message body containing "unsubscribe"
to beowulf-request@xxxxxxxxxxx