IBM Power Connect, learn, share, and engage with IBM Power. Join / Log in
Hi experts,I used command(dd) to test the performance of an AIX client connecting to a storage device over NFS.During the test, the IOPS and bandwidth are 0. However, replacing the linux client does not cause this problem.
Leo,it would be helpful to share the command with all parameters you are using in this test.How are you monitoring IO?NFS is a cached file system, so your test, and while you monitored, may have just hit the local cache?
Hi,The command parameters are as follows, dd if=/dev/zero of=/mnt/data bs=1MI used 'topas' to check the network port bandwidth on AIX client and found that the bandwidth was zero.I also thought it might be a local cache issue, but with other storage devices the situation was different.I was wondering if there are any NFS server parameters to control the local caching on the client side?
Hi LeoAre small IO operations successful? Such as touching a file, or copying a small file?It may be a jumbo frame issue if small IO is ok, but large IO is getting blocked.
I tried small IO operations that also caused the problem.The command parameters are as follows, dd if=/dev/zero of=/mnt/data bs=8k
Hi LeoIt will be useful to know if touch and cp can write to the NFS share.touch /mnt/myfilecp /etc/motd /mntls -l /mntAlso try:ping <NFS_server>ping -s 9000 <NFS_server>
Leo please do no -a | grep tcp_sendspace and no -a | grep tcp_recvspace and no -a | grep tcp_rfc1323 settingswhat is the size of tcp_sendspace and recvspace and if rfc1323 is 1 or 0 other NFS setting what are they on your server - before you do test?# /usr/sbin/nfso -p -o nfs_socketsize=600000# /usr/sbin/nfso -p -o nfs_tcp_socketsize=600000/usr/sbin/nfso -p -o nfs_use_reserved_ports=1/usr/sbin/nfso -p -o portcheck=1/usr/sbin/nfso -p -o sockthresh=60