nfsstat -s output paramter dupchecks
Franz Stibitzhofer
realy means ????
Is a quite high value relative to the amount of nfs calls good or bad ?
Tom Weaver
Franz Stibitzhofer <stib...@avl.com> wrote:
>Can anyone tell me what the dupchecks value in the nfsstat -s output
>realy means ????
Its the number of times that the NFS server checked a request against the
dupcache. NFS traditionally uses UDP for sending requests between client and
server. With UDP, its entirely possible that packets could get lost. So, if
the client hasn't gotten an answer in a while, it sends another request.
It could happen that a request is not lost, but simply delayed a bit (or the
server delayed a bit) and that both the request and its retry get through.
Now, some requests - such as "write record N" - are idempotent - it doesn't
matter how many times they are done, the results are the same. (This is true
for a write of the same data). Others - such as "close file" are not: the
second close will fail since the file is already closed.
To help deal with this, the NFS server maintains a cache of non-idempotent
requests. When a new, non-idempotent requests comes in, its checked against
the cache and ignored if it looks like a retry of one that was just done.
This keeps a lot of temporary network glitches from showing up as application
level failures.
>Is a quite high value relative to the amount of nfs calls good or bad ?
I doubt if its either good or bad; it just says something about the type of
requests that are being made. Now, if the dupreqs value is high, that would
suggest a problem in either the network or the NFS server.
--
______________
Tom Weaver (512) 838 8277, T/L 678-8277
tvweaver@[austin|us|vnet].ibm.com or TVWEAVER at [RTP|RALVM17]
Franz Stibitzhofer
> In article <382824BF...@avl.com>,
> Franz Stibitzhofer <stib...@avl.com> wrote:
> >Can anyone tell me what the dupchecks value in the nfsstat -s output
> >realy means ????
>
> Thanks a lot Tom for this explanation. It was very usefull.