Ten to 1 expansion of SMB requests to fcntl system calls?

[Volker Lendecke]

On Wed, May 22, 2013 at 09:52:14PM -0700, Richard Sharpe wrote:
> On Wed, May 22, 2013 at 9:50 PM, Volker Lendecke
> <Volker.Lendecke at sernet.de> wrote:
> > On Wed, May 22, 2013 at 09:04:50PM -0700, Richard Sharpe wrote:
> >> Hi folks,
> >>
> >> I have been running some dbench runs using the standard client.txt
> >> file and using dtrace to measure system calls made by the smbd.
> >>
> >> For ~97,000 requests (SMB1 requests) I see 1,288,252 fcntl syscalls.
> >> This seems excessive. Where might those be coming from?
> >>
> >> This is a FreeBSD 8.0 system.
> >
> > What Samba version?
> 
> 3.6.12 plus some patches from 3.6.13 and your patch to remove stat
> from the non-sendfile read path.

Also, this is exactly the reason why I've spent so much time
on the tdb mutexes. Under Linux, they are much faster than
fcntl locks.

Last time I looked, FreeBSD did not have robust mutexes. But
maybe you could try to find a different IPC mechanism on
FreeBSD that offers semantics that are close enough. We need
mutual exclusion and automatic cleanup if a process dies.
With the tdb mutex work I think the tdb locking is
understood well enough now that we could possibly adapt to a
different locking scheme with reasonable effort.

Volker

-- 
SerNet GmbH, Bahnhofsallee 1b, 37081 Göttingen
phone: +49-551-370000-0, fax: +49-551-370000-9
AG Göttingen, HRB 2816, GF: Dr. Johannes Loxen
http://www.sernet.de, mailto:kontakt at sernet.de