Trunking effect on latency
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.
Vann H. Walke walkev at presearch.comMon Sep 16 15:26:05 PDT 2002
- Previous message: Mpich 1.2.3 first run problem
- Next message: Distributed Computing Workshop
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
I'm hoping for some help with some network configuration questions: We're working on a 72 node, 144 processor system connected with gigabit over copper. Our current network configuration is essentially a "fat tree." The bottom layer of switches (9, Dell 10-port 5012) each have 8 nodes attached to them. These switches are attached to 3 mid-level switches using a 2 port trunk (2Gb trunk). The mid-level switches are connected to the top level switch (Dell Powerconnect 24 Port, 5224) using a 4 port turnk (4Gb). After doing some performance analysis of our application, we determined that none of the trunks were being used at anything close to 1Gb. If in this situation we were latency limited, my theory was that removing the mid-level of switches and going without the trunks would improve performance (by removing the latency added by the additional switches). However, performance with each of the smaller switches connected directly to the master switch was considerably worse. Any ideas on why this is the case? Can trunking improve latency? Is there effect on the network that I'm missing? Thanks, Vann
- Previous message: Mpich 1.2.3 first run problem
- Next message: Distributed Computing Workshop
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
More information about the Beowulf mailing list
