[linux-cifs-client] Question for the cifs gurus: reducing write timeouts

[wasrshi nimara]

Hi,

Any pointers on this. I can really use some expert opinion here.


On Tue, Nov 4, 2008 at 12:40 PM, wasrshi nimara <warshinimara at gmail.com>wrote:

> Hi,
>
> I am testing the CIFS code for a linux client (v 2.6.24.3) in the event of
> a network failure.
>
> I notice that the cifs_writepages calls CIFSSMBWrite2 with timeout values
> as CIFS_LONG_OP (45 seconds).
> This is how long the write function sleeps before giving up on the
> communication.
>
> The comments say 45 seconds was picked because of server oplock timeouts. I
> am looking
> at optimizing this path in the event of the client owning an exclusive
> (r/w)  oplock on the
> file. In this case wouldn't a smaller timeout like read(15 seconds)
> suffice. I want to confirm
> this change would work and not blow stuff up!
>
> I am also curious on the how the read timeout was fixed at 15 seconds? Any
> tribal
> knowledge here ....
>
> thanks
> warshi
>
-------------- next part --------------
HTML attachment scrubbed and removed