The FNN (Flat Neighborhood Network) paradox
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.
Patrick Geoffray patrick at myri.comWed Feb 21 08:37:30 PST 2001
- Previous message: The FNN (Flat Neighborhood Network) paradox
- Next message: The FNN (Flat Neighborhood Network) paradox
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Hi Miroslaw, Miroslaw Gawenda wrote: > The reason for this test is the potential problem when some nics access > memory and PCI bus - I think so this can't give any reasonable speedup > because 3 nics can't (probably) transmit and receive data in the same > time. If you use Fast Ethernet, 3 cards won't kill the PCI (10 MB/s per card, full-duplex => 60 MB/s). Nowadays, cheap 32 bits/33 MHz PCI can sustain 100-110 MB/s. If you use IP, the host will die first, because of memory copies in the IP stack / driver (wasting memory bus and cpu cycles) and interrupt processing. If you want to do message passing, I would recommand you some OS-bypass communication layer like GAMMA (http://www.disi.unige.it/project/gamma/). > How many nics is ideal for clusters ? One pipe per node. The size of the pipe depend of your needs (and your budget). Cheers. -- Patrick Geoffray --------------------------------------------------------------- | Myricom Inc | University of Tennessee - CS Dept | | 325 N Santa Anita Ave. | Suite 203, 1122 Volunteer Blvd. | | Arcadia, CA 91006 | Knoxville, TN 37996-3450 | | (626) 821-5555 | Tel/Fax : (865) 974-0482 | ---------------------------------------------------------------
- Previous message: The FNN (Flat Neighborhood Network) paradox
- Next message: The FNN (Flat Neighborhood Network) paradox
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
More information about the Beowulf mailing list
