I have noticed a significant decrease in performance on NFS in 2.6.0-test7 and later. I have tested kernels 2.4.22-xfs and 2.6.0-test[4-10], kernels 2.6.0-test[4-6] all gave the same high performance and kernels 2.6.0-test[7-10] all gave the same low performance. I have saved testresults with kernels 2.4.22-xfs, 2.6.0-test6, 2.6.0-test7 and 2.6.0-test10. The time it takes to copy a 100 million byte file from my file server (which runs on RedHat 9 with a 2.4.19-xfs kernel) is 3 to 4 times longer while running 2.6.0-test[7-10] than 2.6.0-test[4-6]. I have tested this by creating a 100 million byte file on the fileserver with dd: dd if=/dev/zero of=100000000.bytes bs=1000000 count=100 Then I have timed the time it takes to cat the file to /dev/null on my workstation: /usr/bin/time cat /mnt/home/fredrik/100000000.bytes > /dev/null I have also run netstat -s and nfsstat right after the above command and put the output in a file test.txt, attached to this mail. I have verified that the problem isn't with the network card by using netcat (nc) to transfer to file. The relevant line in my fstab is: maggie:/home /mnt/home nfs defaults 0 0 I have tried different rsize and wsize options to mount with no success. The output for the test.txt file is generated by a small script that runs: uname -r /usr/bin/time cat /mnt/home/fredrik/100000000.bytes > /dev/null netstat -s nfsstat The tests for all kernels are on a freshly booted kernel. mount --version gives: mount: mount-2.11y The only thing that stands out to me (not a kernel hacker) is that for 2.6.0-test[7,10] and 2.4.22-xfs netstat -s reports a large number of packets that it failed to reassemble, while 2.6.0-test6 reports none. Also 2.6.0-test[7,10] and 2.4.22-xfs nfsstat reports a large number of resends while 2.6.0-test6 reports no resends. Although 2.4.22-xfs and 2.6.0-test[7,10] both reports about the same number of resends and failed reassembled packets, 2.4.22-xfs is more then twice as fast to transfer the file than 2.6.0-test[7,10]. Anyone can give a good explanation? If you need more details and/or other information, feel free to ask.