<html>
<body>
Thanks for the results, and the link. In section 6.7 of the NAS
Parallel Benchmark
(<a href="http://www.nas.nasa.gov/News/Techreports/1996/PDF/nas-96-010.pdf">
NPB 2.1 Results Report, NAS-95-010</a> (PDF-213KB) on MPI, I found a
discussion of the Clustered-SMP issues discussed so far in this
thread. Its interesting that these issues discussed twelve years
ago are coming around again. La plus ca change..., I
suppose.<br><br>
In addition, there is a table of results in that section for an SGI Power
Challenge Array showing that idling processors on a given node and using
more nodes improves the speed per processor across four different code
kernels and two different problem sizes. This doesn't tell us how a
hybrid MP/MT application would work within a 4 core 2 CPU node, but it
does hint that memory contention can be just as nasty a problem as high
latency message transmission.<br><br>
<br>
Mike<br><br>
At 12:52 PM 12/10/2007, you wrote:<br>
<blockquote type=cite class=cite cite="">Some people had asked for more
details:<br><br>
NAS suite version 3.2.1<br>
Test class was: B<br>
Units are Mops (Million operations per second)<br>
see the NAS docs for more information<br><br>
--<br>
Doug<br><br>
<br>
> I like answering these types of questions with numbers,<br>
> so in my Sept 2007 Linux magazine column (which should<br>
> be showing up on the website soon) I did the following.<br>
><br>
> Downloaded the latest NAS benchmarks written in both<br>
> OpenMP and MPI. Ran them both on an 8 core Clovertown<br>
> (dual socket) system (multiple times) and reported<br>
> the following results:<br>
><br>
> Test
OpenMP
MPI<br>
> gcc/gfortran
4.2 LAM 7.1.2<br>
> ------------------------------------<br>
> CG
790.6
739.1<br>
> EP
166.5
162.8<br>
> FT
3535.9
2090.8<br>
> IS
51.1
122.5<br>
> LU
5620.5
5168.8<br>
> MG
1616.0
2046.2<br>
><br>
> My conclusion, it was a draw of sorts.<br>
> The article was basically looking at the<br>
> lazy assumption that threads (OpenMP) are<br>
> always better than MPI on a SMP machine.<br>
><br>
> I'm going to re-run the tests using Harpertowns<br>
> real soon, maybe try other compilers and MPI<br>
> versions. It is easy to do. You can get the code here:<br>
><br>
>
<a href="http://www.nas.nasa.gov/Resources/Software/npb.html" eudora="autourl">
http://www.nas.nasa.gov/Resources/Software/npb.html</a><br>
><br>
> --<br>
> Doug<br>
><br>
><br>
><br>
><br>
><br>
><br>
><br>
><br>
><br>
>> On this list there is almost unanimous agreement that MPI is the
way to<br>
>> go<br>
>> for parallelism and that combining multi-threading (MT) and<br>
>> message-passing<br>
>> (MP) is not even worth it, just sticking to MP is all that is
necessary.<br>
>><br>
>> However, in real-life most are talking and investing in MT while
very<br>
>> few<br>
>> are interested in MP. I also just read on the blog of Arch
Robison " TBB<br>
>> perhaps gives up a little performance short of optimal so you
don't have<br>
>> to<br>
>> write message-passing " (here:<br>
>>
<a href="http://softwareblogs.intel.com/2007/11/17/supercomputing-07-computer-environment-and-evolution/" eudora="autourl">
http://softwareblogs.intel.com/2007/11/17/supercomputing-07-computer-environment-and-evolution/</a>
<br>
>> )<br>
>><br>
>> How come there is almost unanimous agreement in the
beowulf-community<br>
>> while<br>
>> the rest is almost unanimous convinced of the opposite ? Are we
just<br>
>> tapping<br>
>> ourselves on the back or is MP not sufficiently dissiminated or
... ?<br>
>><br>
>> toon<br>
>><br>
>><br>
>><br>
>> _______________________________________________<br>
>> Beowulf mailing list, Beowulf@beowulf.org<br>
>> To change your subscription (digest mode or unsubscribe)
visit<br>
>>
<a href="http://www.beowulf.org/mailman/listinfo/beowulf" eudora="autourl">
http://www.beowulf.org/mailman/listinfo/beowulf</a><br>
>><br>
>><br>
>> <br>
>><br>
><br>
><br>
> --<br>
> Doug<br>
> _______________________________________________<br>
> Beowulf mailing list, Beowulf@beowulf.org<br>
> To change your subscription (digest mode or unsubscribe) visit<br>
>
<a href="http://www.beowulf.org/mailman/listinfo/beowulf" eudora="autourl">
http://www.beowulf.org/mailman/listinfo/beowulf</a><br>
><br>
> !DSPAM:475c325f61251246014193!<br>
><br><br>
<br>
--<br>
Doug<br>
_______________________________________________<br>
Beowulf mailing list, Beowulf@beowulf.org<br>
To change your subscription (digest mode or unsubscribe) visit
<a href="http://www.beowulf.org/mailman/listinfo/beowulf" eudora="autourl">
http://www.beowulf.org/mailman/listinfo/beowulf</a></blockquote>
!DSPAM:475f2a0321726491211187!
</body>
</html>