[Samba] Troubleshooting poor (small) random read performance -- serverid.tdb?

[Ray Van Dolson]

On Fri, Jun 20, 2014 at 08:37:18AM +0200, Volker Lendecke wrote:
> On Thu, Jun 19, 2014 at 02:48:52PM -0700, Ray Van Dolson wrote:
> > We're using strict locking = no at the global level and locking = no at
> > the share level.  The latter option is where we get good performance
> > improvements for this workload.
> 
> So it's really byte range locks that bite you. Can you take a network
> trace of one client? It will reveal your data, it will be huge, but it
> might provide valuable input for us to reproduce and eventually fix the
> issue with brlocks for you in a proper way. "locking=no" is not a good
> idea except for special use cases :-)
> 
> Volker

I believe I can share a trace -- will confirm.  Should I open a bug for
this against 4.1.8 for better tracking?

I may be misunderstanding some of the risks of using locking=no, but at
least with our read only workload (share is set to read only -- and we
could even set the underlying file system to read only to minimize risk
from some local process mucking things up), we felt we were pretty
safe.

Ray