Samba Performance.

Anthony Ord samba at rollingthunder.demon.co.uk
Fri Feb 26 18:12:00 GMT 1999


At the moment, I have a samba server, and performance seems very poor.

I have done some trials on it, and determined the following.

Copying a 10Mb file: 70 secs
Typing a 10Mb file: 18 secs

No surprises so far?

I cranked up the debug level (using sigusr) and made a log file of
when I "copy"ied this file, and when I "type"d it. 

Interestingly, the time taken "type"ing the file went up to 38secs
while "copy"ing it remained at 70 secs.

I then stopped the logging and used strace to monitor the smbd
process. The times remained at 38secs and 70 secs.

Second surprise, the strace file for the "copy" is only 98k, while the
strace file for the "type" is 1.3Mb. This is despite the "type" file
covering half the elapsed time of the "copy" file. 

It seems that samba works completely differently internally when doing
the "type" as when it does the "copy", despite the "official" smbd
generated logs not looking that different. The main difference (as far
as I can see) is the "write" system call, which is taking 100 times
longer to  move 16 times larger data, when the "copy" command is used.

Is anyone interested in using these logs to investigate samba's
performance problems? I'm out of my depth right about here. This is
not a critical server, so I can perform experiments on it.

Regards

Anthony
-- 
-----------------------------------------
| And when our worlds                   |
| They fall apart                       |
| When the walls come tumbling in       |
| Though we may deserve it              |
| It will be worth it  - Depeche Mode   |
-----------------------------------------


More information about the samba mailing list