When using rsync and similar commands that attempt to transfer files, after a short time a file transfer hangs. It essentially makes networking useless.
You should try to log this with tcpdump to see what's causing this. This might also be a classic case of filtering ICMP destination unreachable: fragmentation needed but don't fragment bit set messages. Is there filtering done on ICMP between the endpoints?
Created attachment 5730 [details] The output of tcpdump during problem occurrence
I started tcpdump before issuing a command that hung and stopped it several minutes later. I have attatched the output to this bug report. If I leave the command run for a very long time (tens of minutes) more of the file gets transferred, but at a rate of probably less that 2kb/s. My ethernet is 100MB/s (a Toshiba 470CDT with a 3COM 10/100 PCMCIA card and a gateway 600 with a 3COM 10/100 PCI card connected via a crossover cable - it has worked for some time with both machines running RedHat 5.2 then 6.1 then 6.2 but has failed ever since I upgraded the Gateway to 7.0 - I updated the notebook first and it ran fine).
I believe that I have not altered the standard installation with regard networking (other than the IP address and /etc/hosts to add some extra hosts). So I don't believe I have filtering on but how do I tell? The amount of information that gets transferred before it slows down varies considerably. Somtimes just a few KB other times not till after tens of MB.
This looks pretty much like a classic half/full duplex problem. The symptom of it is that tranfers start going fast for about 0.5-2.0 seconds or so, and then essentially time or are reduced to 0.5-30 Kbytes/sec like with you. If this is the problem, this should also happen with other protocols, like ftp. You might want to check your Gateway system's interface configurations (/etc/modules.conf). Did the driver change at upgrade? Also, try to diagnose the problem with mii-diag and friends, available from ftp://ftp.scyld.com/pub/diag/.
I got the diag programs from the above web site. The output of vortex.diag shows the MAC settings of the notebook to be running full-duplex and the gateway to half-duplex. So it seems the above comment is correct. But how do I change this?
See /etc/modules.conf for options= line. If there is no such line, the card is in autonegotiation state (if not forced otherwise in EEPROM with NIC DOS configuration floppy). Something like: options 3c59x options=4 full_duplex=1 forces the card to 100/FD. This varies from driver to driver.
I added the line you suggested to /etc/modules.conf and things now work considerably better. Thanks!