Performance issue with Samba 3-0
jra at samba.org
Tue Aug 5 00:42:42 GMT 2003
On Mon, Aug 04, 2003 at 11:40:53AM -0700, Ravi Wijayaratne wrote:
> Hi All,
> We are seeing a performance degradation of Samba 3.0 with Linux 2.4.19-xfs-1.2 (Case A) compared
> Linux 2.4.19-XFS-1.1 (Case B). The details are as follows.
> 1. We are seeing a ~15% drop in netbench performance (Clients were a mix of NT/2k) performance.
> Upon examining the detailed results we see that write operations are hurting
> overall performance
> 2. From profiling information we see that for Case A there were 500000 more SMB PDUs (roughly
> about 20-25%) used for the test. The discrepancy was for the most part accounted
> for by the elevated number of trans2_query_filepath_info PDUs and the WriteX PDus.
> 3. we examined the debug level 10 logs. We see that for Case A the WriteX op sequence is
> but for Case B is
> WriteX->WriteX->WriteX .....
> So it seems that for *identical Samba 3.0 builds* the clients chose a different write protocol
> for Case A which was counter productive.
> I am curious what could have caused this.?
What is the value being returned for the allocation size in SMB_QUERY_FILE_ALLOCATION_INFO:
or SMB_FILE_ALLOCATION_INFORMATION: (trans2 calls) ?
Does it differ between xfs-1.1 and xfs-1.2 ?
This can cause the client to issue different calls.
More information about the samba-technical