<DIV>Yes, the "serial" FFTW seems to run faster than the one (downloaded from netlib.org) that I used in my MPI code. Since I need to do a few things between FFT's, so I just wrote my own "MPI" version (by taking advantage of my particular data structure) without figuring out how to do it, perhaps more neatly :-), using FFTW. I did not use FFT provided by commercial vendors because I need to port the code across various platforms. Having source codes and being able to recompile is very important to me.</DIV>
<DIV> </DIV>
<DIV>Thanks for the hints of Warewulf; that's something I did not know of. Will have to some homework. :-)</DIV>
<DIV> </DIV>
<DIV><BR><BR><B><I>Tim Mattox <tmattox@gmail.com></I></B> wrote:</DIV>
<BLOCKQUOTE class=replbq style="PADDING-LEFT: 5px; MARGIN-LEFT: 5px; BORDER-LEFT: #1010ff 2px solid">On Wed, 01 Sep 2004 21:46:25 -0400, Joe Landman<BR><LANDMAN@SCALABLEINFORMATICS.COM>wrote:<BR>> SC Huang wrote:<BR>> <BR>> > Thanks, Jeff.<BR>> ><BR>> > 1. It is not frequency domain modeling. The FFT's, together with a<BR>> > multigrid method, are used in solving a Poisson equation (for each<BR>> > time step) in the solution procedure.<BR><BR>You should also take a look at FFTW http://www.fftw.org/ but unfortunately<BR>their version 3.0 doesn't yet do MPI. The older 2.1.5 version does MPI.<BR><BR>[snip]<BR>> > 4. Thanks for the suggestions on the diskless or other file systems. I<BR>> > will discuss that with my group members.<BR>> <BR>> <BR>> This is an interesting way to go if your IO can be non-local, or if you<BR>> just need local scratch space. It makes building clusters quite<BR>> fast/easy.<BR><BR>You can use
Warewulf to boot the nodes from ramdisks (via PXE or<BR>Etherboot), and then use local disks for swap and scratch storage<BR>(possibly as part of a PVFS/GFS/Luster/"did I forget one?"/GPFS file system).<BR>See http://warewulf-cluster.org/ for more details. Disclaimer: Warewulf<BR>was such a great system I became one of it's developers.<BR><BR>[snip]<BR>> > Also, I heard of the name "channel bonding" here and there. Is that<BR>> > some kind of network connection method for cluster (to use standard<BR>> > switches to achieve faster data transfer rate)? Can someone briefly<BR>> > talk about it, or point me to some website that I can read about it? I<BR>> > did some google search about it but the materials are too technical to<BR>> > me. :( Is it useful for a cluster of about 30-40 nodes?<BR>> <BR>> <BR>> There are plusses and minuses to channel bonding. Search back through<BR>> the archives for details. I am not sure if the issues with
latency and<BR>> out of order packet send/receive have been addressed in 2.6.<BR><BR>I'm sure others can comment from experience, but my impression<BR>has been that, at least for the time being, channel bonding with GigE<BR>isn't particularly helpful, and can slow things down in some situations.<BR>I'd hoped to have tested this myself already, but other research priorities<BR>have been higher.<BR><BR>Good luck on your cluster design choices.<BR>-- <BR>Tim Mattox - tmattox@gmail.com - http://homepage.mac.com/tmattox/<BR></BLOCKQUOTE><p>__________________________________________________<br>Do You Yahoo!?<br>Tired of spam? Yahoo! Mail has the best spam protection around <br>http://mail.yahoo.com