NFS client - ASYNC write

[Matt Sears]

Does anyone know how to get an NFS client to send a WRITE with the stable bit set to ASYNC?

I've exported the share using ASYNC, but it seems there is something else I'm missing. I mounted the share and used vi to write a character to a file. Using Wireshark, I can see that the client is sending FILE_SYNC rather than ASYNC.

[Dave Olker]

Hi Matt,

Have you tried using other applications to send WRITE data to the server other than vi? It could be the way vi is issuing the write that causes the FILE_SYNC semantics. I know there are different ways to write data on other platforms, like mmap(), that cause synchronous semantics.

Do you see the same behavior with something like cp? Also, are you using default NFS mount options? If not, what options are you using to mount the filesystem?

Regards,

Dave

[Matt Sears]

Dave, thank you for your response. I had tried vi and emacs, and when I tried cp it too gave used WRITE with FILE_SYNC. I then tried dd. I used dd if=/dev/full of=/nfs_mounted_file bs=1024 count=100. This caused the WRITE request to be called with the stable bit set to UNSTABLE. I guess my other WRITE requests were just too small.

[Dave Olker]

What NFS mount options are you using for this filesystem? Can you do an "nfsstat -m" and cut/paste the output here so I can see exactly which options you're using?

Also, what OS is running on the NFS client and server?

Dave

[Matt Sears]

> nfsstat -m
/mnt/1 from e4s01:/mnt/1
Flags: rw,vers=3,rsize=32768,wsize=32768,hard,proto=tcp,timeo=600,retrans=2,sec=sys,addr=e4s01

Client OS: CENTOS5 (linux kernel 2.6.18-128)
Server OS: RHEL5 (linux kernel 2.6.18-92)

I may not have been clear with my previous post. I was able to force the desired condition by sending a large WRITE request. My guess is that the other requests were just too small.