read and write size in NFS v4

[Stephen Carville]

On Sat, Aug 16, 2008 at 1:14 PM, Yong Huang <yong321 at yahoo.com> wrote:
>> From: "Stephen Carville" <stephen.carville at gmail.com>
>> Subject: read and write size in NFS v4
>> To: "General Red Hat Linux discussion list" <redhat-list at redhat.com>
>>
>> I am migrating my NFS file server from RHEL3 to RHEL5 and moving up to
>> NFS v4 where i can.  Once question I still have is what is the maximum
>> rsize and wsize i can set for V4 automount?  The file
>> include/linux/nfsd/const.h says teh maximum is 32*1024 (32K) but some
>> google postings claim that does not apply to V4;  I've tried v4
>> automounts with values up to 1048576 and none of the machine have
>> complained.
>>
>> A part of the new server's duties will be to accept the RMAN (oracle)
>> backups so speed is important.  I am using a dedicated gigabit switch
>> with jumbo frames (9000 bytes) and I have tuned the TCP  for the
>> "private" interfaces on all the connected interfaces.  These is the
>> only parameter I am not sure what I can get away with.  I have it set
>> to 32768 for the current v3 setup.
>>
>> --
>> Stephen Carville
>
> I don't know the answer. But can you check nfsstat to see what it says the size
> is? Also, strace some read and write operations and see what the return values
> for read() and write() are. But I guess the latter method may not tell us the
> underlined driver level chunk size.

Tried that but it doesn't seem to tell me what the actual block size
is.  However I found that /proc/mounts does record the wsize and rsize
and a little experimentation led me to conclude that the default is
32768 and the max is 32768.

-- 
Stephen Carville