[Beowulf] Gigabit PCI-E NIC
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.
Mark Hahn hahn at physics.mcmaster.caSat Sep 9 07:49:16 PDT 2006
- Previous message: [Beowulf] Gigabit PCI-E NIC
- Next message: [Beowulf] Gigabit PCI-E NIC
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
> As these computers will already have Broadcom NIC, should the second NIC > also be Broadcom? From a performance/stability point of view, does it you need to question your assumption that you need two nics. why do you think so? when will a node have concurrent heavy file and MPI traffic? its possible, I suppose, but I don't see that on my clusters. > NIC? I believe Intel GB NIC is already in the kernel and is a > "favorite" amount clusters so I probably could not go wrong with Intel. there are some intel-o-philes, though I suspect they got that way by being burned by transient problems with other nics. we certainly have a large number of BCM57xx's in our systems, and I've never had any problem specific to them. we've also got some intel-based systems, and they work fine as well. I don't believe there are measurable differences in performance or overhead. it would be interesting to try a really tight comparison (e1000 and tg3 cards, both presumably 1x PCI-e; all my hardware is integrated in either xeon or opteron systems, and thus not comparable...) regards, mark hahn.
- Previous message: [Beowulf] Gigabit PCI-E NIC
- Next message: [Beowulf] Gigabit PCI-E NIC
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
More information about the Beowulf mailing list
