[Beowulf] A Mosix Cluster or Beowulf
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.
Pere Urbón Bayes purbon at deic.uab.esThu Feb 23 03:22:54 PST 2006
- Previous message: [Beowulf] A Mosix Cluster or Beowulf
- Next message: [Beowulf] mpich2 error in 8 nodes
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Vincent Diepeveen wrote: > Performance of mosix and open ssi is of course horrible in terms of > latency between the nodes. But if the full procceses have no interconexion and the full calculation will be done on independent ucp, then the performance is quite good? isn't it? > So using either mosix or open ssi you can do only if your 100 mbit > connections between the nodes are simply not getting used for your software. In short only if your > software is doing only each 10 minutes or something a little bit of communication with other processors. > > For most software here the dual cores AMD are up to 2 times faster than > the dual core intel cpu's when using GCC compiler. > If all what you use is SSE2 then the speed difference will be quite less > than that in advantage of AMD. I'm sorry, but I could not understand what you mean with SSE2? I think that finally I'll be buying AMD, because the performance will be better. > I assume however that the scaling of 2 processes onto 1 node is 100% > anyway as they do 0 communication with each other. > With such cluster software packages that have such ugly communication > behaviour doing any form of communication > is near impossible of course. Hopefully your software isn't doing too > much i/o either. > > AMD chips of course scales a lot better than P4 dual cores. Itanium > scales in fact even better, but just slightly. The diff in scaling > between a P4 dual core and an AMD dual core is significant though. > > Get mainboards with a dual channel memory controller! Off course!! I have this all on mind. > It seems to matter really a lot in tests here for my Diep software (dual > opteron dual core). > > Vincent Regards... -- |---------------------------------------------------------------.--. |Pere Urbón Bayes JabberID: purbon at jabber.org |o_o | |Dept. of Information and Communication's Engineering |:_/ | |E.T.S. d'Enginyeria. Autonomous University of Barcelona // \ \ |08193 Bellaterra, Cerdanyola del Valles (Catalonia) (| | ) |Tel: +34935813573 QC/2005 (Combinatoric and Coding Group) /'\_ _/`\ |-----------------------------------------------------------\___)=(___/
- Previous message: [Beowulf] A Mosix Cluster or Beowulf
- Next message: [Beowulf] mpich2 error in 8 nodes
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
More information about the Beowulf mailing list
