frequency/size distribution of SMBs

[David Collier-Brown]

Steven French <sfrench at us.ibm.com> wrote:
> I want to analyze whether the default small SMB buffer size for the Linux
> cifs client, which is currently set at 256 bytes, is reasonable.   [...]
> What I noticed running the connectathon "nfs" (posix
> file API) tests from cifs client to Samba was that increasing the cutoff
> from 256 bytes to 1024 bytes changed very little, most SMBs appear to be
> either smaller than 256 bytes or much larger (e.g. close to the negotiated
> buffer size, or at least bigger than 1K).
> 
> Has anyone plotted a frequency distribution of SMBs by size?   This would
> be helpful in setting buffer sizes optimally.   I did not see an easy way
> to do this in ethereal (it would be easier to code profiling code in smbd
> or cifs.ko for this than messing with ethereal it seemed).

	Do we have a Solaris 10 build machine?
	dtrace happily provides that kind of information.

	This would be best done inside the Samba team,
	although I could theoretically ask one of our folks
	in benchmarking when (s)he is next going to be
	testing Samba...

--dave
-- 
David Collier-Brown,      | Always do right. This will gratify
Sun Microsystems, Toronto | some people and astonish the rest
davecb at canada.sun.com     |                      -- Mark Twain
(416) 263-5733 (x65733)   |