Steve Roylance wrote:
hi
I have 2 boxes side by side both have gigabit ethernet connected to a switch on the desk. The source is an i3-550 with /home mounted from an md0-array of 4*1TB drives in a raid5 config and the sink is an E2200 with /home on 1 disk and an nfs mount point
issue command mv Fedora-18-x64.iso from source sink. The transfer starts at 350Mb/s for about 200MB then drops to 500kb/s for the rest.
This is not necessarily the network. One disk or the other might be buggered. I would start by isolating each component and testing it, e.g. dd if=foo.iso of=/dev/null # on source dd if=/dev/zero count=1G of=delete-me # on sink ssh sink 'nc -l -p 12345 | dd of=/dev/null' & ssh source 'dd if=/dev/zero count=1G | nc sink:12345' Oh, you said you were doing the transfer over NFS, not ssh or netcat. NFS has trouble with lots of little files, but a big ISO shouldn't give it any grief. It could also be something else on that switch (I'm thinking particularly of cycles, that STP would stop), or that the switch or one of the NICs is buggered and can't sustain a meaningful speed. Check mii-tool or ethtool on both while the transfer is running and slow; check nfs read and write block sizes (though that's very unlikely to help much with the numbers you've given).