Myrinet vs. Dolphin
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.
Jason Holmes jholmes at psu.eduThu Feb 8 06:25:29 PST 2001
- Previous message: Myrinet vs. Dolphin
- Next message: Gigabit (1000base-T) NIC performance?
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Sergiusz Jarczyk wrote: > > On Wed, 7 Feb 2001, Bill Broadley wrote: > [...] > > You mention 4 1m cables? Why 4? The dolphin card has 2 ports requiring > > 2 cables no? Or can a single card be in 2 loops, i.e. a matrix where > > you have loops in the x and y directions. > > > > I've looked through the dolphin site and haven't found a good > > description of the physical interface/possible topologies. > > > > No, Dolphin cards (known as Wulfkit - packaged with SCALI Software > Platform) has 4 ports (two on board and two on doughter card). Possible > topologies include 2D Torus and ring. In Torus (which is better than ring > and allow to build clusters with some fault tolerance) each card is > connected with four other (two in X and two in Y directions). Not to mention that the newer cards will have 4 ports all on one card (no daughterboard). The 64/66 cards will be like this, at least (I'm told these just went into production). I'm not sure if the 32bit cards with 4 ports that were shown to me when Dolphin came to visit are going to be available or if they were just concept cards. There was also talk about maybe doing 4 port cards with a 4 port daughterboard to allow for even more interesting topologies (though this was also just concept talk). -- Jason Holmes
- Previous message: Myrinet vs. Dolphin
- Next message: Gigabit (1000base-T) NIC performance?
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
More information about the Beowulf mailing list
