[Samba] IO Wait problem on a patched samba

[Yannick Bergeron]

Hi all,

Here is the context. We are running Samba in a DCE/DFS environnement on 
AIX box (4.3.3 and 5.2). We've patched the Samba source to support this 
environnement (although, we didn't created the patch, someone working 
for IBM US dit it). We are using Samba 2.2.7b and 2.2.8a.

We have a lot of Samba connections (400-600 shares everytime, 24 hours 
per day). We were running with only one server at time first and we had 
stability problem with Samba. This one was hanging after 2-3 days of 
service, so we had to dispatch the traffic on the backup server, than 
reboot the primary server, and then dispatch the traffic on the primary 
server.

Since some weeks, we are running with two primary server and one backup. 
We are stable for 10-12 days this way. But again, the Samba service is 
hanging and we must reboot it.

I'm not currently sure that it's only a Samba problem, maybe it is 
because of the modified source, or maybe it could be the use of Samba in 
our environnement.

We can observe the problem through the IO Wait of the AIX box, which 
disappear when Samba traffic is dispatched on the backup server. If we 
don't reboot the primary box and dispatch the traffic on it after the 
problem occurs, it will repeat in the next mins/hours.

The problem has appear more than one year ago, and I'm hoping that I 
could completely correct it at the end of the summer. So since the guy 
that was working on the problem will be absent for 6 or more weeks, I've 
herited of the problem. I don't have a great experience with Samba so I 
post the problem here. I'm not asking you to resolve the problem, but 
any help would be appreciated.

thanx

Yannick Bergeron
burgergold at hotmail.com
yaberger at ca.ibm.com