[Beowulf] Re: More cores/More processors/More nodes?
Many of your questions may have already been answered in earlier discussions or in the FAQ. The search results page will indicate current discussions as well as past list serves, articles, and papers.
Håkon Bugge Hakon.Bugge at scali.comThu Oct 5 05:45:08 PDT 2006
- Previous message: [Beowulf] Intel MPI 2.0 mpdboot and large clusters, slow tostart up, sometimes not at all
- Next message: [Beowulf] Re: More cores/More processors/More nodes?
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
At 04:47 03.10.2006, beowulf-request at beowulf.org wrote: Greg Lindahl wrote: > On Thu, Sep 28, 2006 at 07:06:14PM +0100, Peter Wainwright wrote: > > >> What (in your opinion) is the right tradeoff between more cores, >> more processors and more individual compute nodes? > > $/performance. > > Once you have your code written into pure MPI form, then you can run > on any of the above alternatives. Then you can simply work out > the price for various things, and make a guess at the performance. > Run a few benchmarks to check your guesses. On two-socket, dual core Opteron systems, we often see that throughput is enhanced by using two times two processes per node. For example, if you're running jobs with 16 processes on an eight-node cluster, 8x2 plus 8x2 is faster than 4x4 and 4x4 (nnodesxppn). But gain depends on flexibility of process-to-processor mapping of the MPI used. Some apps like to use as few sockets as possible, whereas others like the opposite ;-) Håkon
- Previous message: [Beowulf] Intel MPI 2.0 mpdboot and large clusters, slow tostart up, sometimes not at all
- Next message: [Beowulf] Re: More cores/More processors/More nodes?
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
More information about the Beowulf mailing list
