[linux-cifs-client] Re: debugging dbench to track performance

[Steve French]

Jeremy Allison wrote:
> On Sat, Jul 15, 2006 at 01:04:44PM -0500, Steve French wrote:
>   
>> Running with very current cifs against very current SAMBA_3 branch 
>> (server) - cifs mount over loopback interface
>>
>> I am getting puzzling behavior of dbench again - with "dbench 70" (ie 70 
>> processes) dbench never gets out of warmup mode (my guess is that a 
>> request or two gets lost on the server or is so slow that dbench thinks 
>> that it has not reached steady state for warmup - but I am not I am 
>> convinced that I understand dbench's warmup phase properly).   Running 
>> it with fewer processes dbench gets through its warmup fine but is 
>> performing pretty poorly (hopefully because of the memory leak in Samba 
>> that jra mentioned - but I am not convinced that that is the issue).   
>> Also note that "cannot remove directory" errors in the cleanup phase - 
>> any ideas how to debug those (there is presumably too much data to catch 
>> this on a sniffer without affecting timing) ie catch the problem earlier 
>>     
>
> With the latest SAMBA_3_0 server code and the modified cifsfs I
> can complete a dbench 70 very quickly. Here's the output.
>
> jeremy at linux:/mnt> dbench 70
> 70 clients started
> ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++**********************************************************************
> Throughput 146.972 MB/sec (NB=183.715 MB/sec  1469.72 MBit/sec)  70 procs
>
> I'm not getting any errors (on the command line or in /var/log/messages).
>
> Here's a timed version :
>
> jeremy at linux:/mnt> time dbench 70
> 70 clients started
> ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++**********************************************************************
> Throughput 305.796 MB/sec (NB=382.245 MB/sec  3057.96 MBit/sec)  70 procs
>
> real    0m30.230s
> user    1m57.136s
> sys     0m1.794s
>
> Does that look reasonable ? (Not run dbench much before...). This is
> with dbench 3.04.
>
> Jeremy.
>
>   
If that is right - that would be quite impressive performance (similar 
to local dbench performance on my laptop - albeit one with a slow disk) 
- sanity check by running the same thing against a local directory would 
be helpful (just to make sure, if you have not already done, that it is 
hitting the network mount rather than a local  target).   Based on the 
output you posted though I am not convinced that you are running tdbench 
3.04  since with default build and run options, dbench 3.04 prints 
something like e.g.

stevef at smf-t41p:~/Desktop/dbench-3.04> ./dbench 70
dbench version 3.04 - Copyright Andrew Tridgell 1999-2004

Running for 600 seconds with load '/usr/local/share/client.txt' and 
minimum warmup 120 secs
70 clients started
  70       153   217.56 MB/sec  warmup   1 sec
  70       178   130.99 MB/sec  warmup   2 sec
  70       543   139.89 MB/sec  warmup   3 sec
  70       817   156.32 MB/sec  warmup   4 sec
  70       830   129.48 MB/sec  warmup   5 sec
etc.

I downloaded dbench from 
http://samba.org/ftp/tridge/dbench/dbench-3.04.tar.gz but as that is  a 
year or two old, ie January 2005, (although it appears to be the newest 
version I could find) - you may have found a newer version somehow.   
Where did you get your version from?