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.
Jared Hodge jared_hodge at iat.utexas.eduWed Jan 24 07:24:24 PST 2001
- Previous message: U of Delaware 132 node 1 GHz Athlon cluster
- Next message: Myrinet vs. Dolphin
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
I was wondering if anyone has experience with Dolphin Interconnect Solutions (http://www.dolphinics.com)? We currently have Myrinet (www.myri.com) on our cluster and it seems to be working reasonably well. I have several questions though. First, we are considering upgrading our cluster with more nodes, and myrinet seems to have a scalability problem in the average cluster size area. We have 8 nodes now, and can go up to 16 nodes with our current switch, but past that, Myricom says we would need one additional switch to get up to 18 nodes and two additional switches to get up to 24. Ok, we just went from spending $5000 on JUST SWITCHES to $15000, and we didn't even get close to the 32 nodes we were thinking about. I know they have there $40,000 Millionare-in-a-box, I mean network-in-a-box "solution", but really who wants to waste that much money (if I just need even 32 nodes). Not to mention that Myrinet prices have not come down in... ah, ever as far as I can tell. Dolphin's switchless technology appears inviting in this area, but I don't know if it has really been proven in the Beowulf area. I also don't know how much Dolphin equipment costs (I just sent a request in for pricing). From the specs that I've seen, I think Dolphin is supposed to be faster as far as latency, but slower as far as bandwidth, but I rarely trust specs from manufacturers, and was wondering if any of you have run comparison tests. From what I've seen though, Myrinet uses 64/32 bit, 66/33 mhz PCI, where Dolphin uses 64/32 bit, 33 mhz only PCI, so I this would lead me to believe that Dolphin is a lot closer to saturating their PCI interface, while Myrinet has and can continue to improve their performance (not that this affects those of us who already own Myrinet hardware). Anyone have any experience or comments. -- Jared Hodge Institute for Advanced Technology The University of Texas at Austin 3925 W. Braker Lane, Suite 400 Austin, Texas 78759 Phone: 512-232-4460 FAX: 512-471-9096 Email: Jared_Hodge at iat.utexas.edu
- Previous message: U of Delaware 132 node 1 GHz Athlon cluster
- Next message: Myrinet vs. Dolphin
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
More information about the Beowulf mailing list
