We have a cluster of 5 computers numbered nodes 1 through 4 and a RAID machine as well. The 4 nodes are running RedHat 6.2 and the RAID is running the latest official release of Debian with a 2.4.x kernel. The computer are connected by a 3Com 100 Mbps switch and all three have a 100 Mbps Card. Up until last week, the transfer rates between all of the machines was in the range of 12 MB/s but now we are getting varying rates. Between some of the machines we get 11 MB/s but between some of the machines as low as 30 KB/s which of course is totally unacceptable if you are trying to do any kind of file transferring. All of the machines except one are able to transfer to at least one other machine at the 11 MB/s and all of the machines transfer to at least 3 machines at a rate of under 200 KB/s. All of the speeds should be at 12 MB/s. No known changes have been made to this cluster. Is a new switch the answer or is it something more basic? When starting an FTP or telnet session, there is no delay in connecting to the machine. I have read about 20-30 second delays in connecting but we are not experiencing that. Just the slow transfer rates. All of the machines have different IPs and there is no other traffic on this small cluster. Can anyone please help? Thanks.:smash:

A few suggestions:

1. Have you checked the status of the RAID sets? With some RAID 5 systems you hardly notice when a disk explodes!

2. Collision detection. If a collision detect circuit is bad, one machine may be "talking over" the others, which then have to retransmit.

If one machine appears to be in worse shape than the others, check the wiring on that one first - wiggle and bang the connectors (tech-talk) - then try replacing or swapping the cables around to see if the problem moves. Failing that, you could try swapping the network cards around.

The only sure way would be to use a network analyser, but 100Mbit hardware is not cheep. There may be some software solutions around that could be run on a "spare" machine with a 100Mbit card set to "promiscuous".

Good luck,
DaveG.

I would try taking out one machine at a time and testing the remaining 3. See if you can isolate who the culprit is.

BTW, are you using a switch or a hub for connectivity?