Red Hat Bugzilla – Bug 494381
RHEL 5.3 hangs during huge 'rcp' file copy at 95% GbE rate, 82575/igb NICs
Last modified: 2011-03-04 05:06:24 EST
Description of problem:
Using 'rcp' to copy about 3TB of data in 20 large files from one
server to another. Files reside on dedicated 'ext4' 4-way
LVM striped logical volumes. Takes 6.5 hours to complete at 95%
gigabit Ethernet utilization, but totally locked/hung RHEL 5.3
twice within the first hour. No kernel messages in 'syslogd' to
help understand the problem. No response from the console,
though network interfaces were all pingable. Had to reset
the server to recover.
Worked around the problem by booting up the F9+kernel.org
partition on the same DL160 server and copying the files.
The receiving server is a Tyan S2912 also running RHEL
5.3, but with the more established 82571/e1000e network
interfaces instead of 82575/igb NIC.
Version-Release number of selected component (if applicable):
igb 18.104.22.168 compiled from source
(Would use native 'igb.ko', but on the DL160 the RH version does
not work at all with 82575. Same version of 'igb' works under
F9 with kernel.org 22.214.171.124 kernel.)
'rcp' huge files from one server to another
Steps to Reproduce:
1. rcp /???/file remote:/???/file
file copy completes
I'm not sure I expect anyone to care about fixing this due
to our use of the upstream 'igb' driver. However it's a serious
failure of the "production" version of RH so it seems worth
reporting. Interesting that a bleeding-edge kernel has no
problem with the network file copy.
This should be fixed by now. Please reopen if you still see this problem with the latest RHEL5 kernel.