On Fri, Jun 3, 2016 at 7:18 AM, Dave Howorth <[log in to unmask]> wrote:

How did you wind up with rsize/wsize of 8K? The default on Linux has been 1
megabyte for a long time.

Do you have a link for that, please? What is 'a long time'? Does it apply to all NFS versions and v3 in particular as the OP asked? Is there a list somewhere of NFS and kernel versions and defaults?

Sorry, no link. I once modified my SL6 kernel to increase the maximum to 2 meg as part of an experiment, which is why I happen to know this...

The maximum NFS block size was raised to 1 megabyte sometime before Linux 2.6.32; see http://lxr.free-electrons.com/ident?v=2.6.32;i=RPCSVC_MAXPAYLOAD for the server limit and http://lxr.free-electrons.com/ident?v=2.6.32;i=NFS_MAX_FILE_IO_SIZE for the client limit. (Note: The comments in the latter file are somewhat misleading, probably because they are out of date.)

For NFS over TCP, by default the NFS client uses the largest block size supported by itself and by the server. I do not know about UDP, but nobody who cares about NFS performance uses UDP.

All of this applies to NFSv3 and NFSv4. NFSv2 limits the block size to 8K, so don't use it.

- Pat