[Beowulf] Multidimensional FFTs
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.
Konstantin Kudin konstantin_kudin at yahoo.comThu Mar 2 12:32:40 PST 2006
- Previous message: [Beowulf] Re: g77 limits...
- Next message: [Beowulf] Multidimensional FFTs
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
>> Anyway, the "alltoall" can be a real killer. If you want to use lots >> of cpus with really small packets, go for something like Parastation >> MPI ( http://www.parastation.com/ ). >Konstantin, > I take it that you're talking about only gigE? For which Parastation > has a novel non-TCP driver. Another strategy would be to use a > low-latency "ether-not" network. > >-- greg Greg, Thanks for the clarification! Indeed, I should have explicitly mentioned that these are troubles with gigE. Unfortunately, with shared use clusters faster interconnects do not seem to fly very well as far as the purchasing decisions go, so one has to try to squeeze out as much out of gigE as possible. By the way, any idea why gigE on Opterons with Broadcom chokes so easily with different flavours of "alltoall" operation, such as if done via "isend+irecv"? If functioning flawlessly, the scaling would have been not too bad, but there seem to be issues of some sort ... Konstantin __________________________________________________ Do You Yahoo!? Tired of spam? Yahoo! Mail has the best spam protection around http://mail.yahoo.com
- Previous message: [Beowulf] Re: g77 limits...
- Next message: [Beowulf] Multidimensional FFTs
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
More information about the Beowulf mailing list
