Ian MacPhedran wrote:
Yeah, the problem occurs when smbd is very slow to start up and contention for locking the various tdb files is high. This isn`t helped by the bug in the Solaris fcntl call that causes big performance problems....Hi; Our environment consists of a server running Solaris 7, with mostly WinXP clients. Smbd is started via inetd. I am currently running Samba 2.0.7 on it, but I experienced this problem with 2.2.5 as well (and even more quickly - the first day under 2.2.5 this happened twice).The problem seems to be caused by several of the WinXP clients each making multiple connections to the server, and starting multiple smbd processes. The process table quickly fills with smbd and inetd processes and slows the server to just about a dead stop. (A crash has to be forced to get the system back in the most severe cases.) 1) Have others experienced similar problems, and how have they dealt with them?
We`ve had a couple of machines die on their feet with this in early 2.2.x samba distros and the solution we found was to hack out some of the tdb traversals that were happening on startup.
But the real way to fix this is to try and get the patch from Sun that addresses the fcntl locking problem, I`ve tested the Solaris 8 version of the patch and it helps things dramatically. If you can get a Solaris 7 version do let us know... And then stop smbd running from inetd but run it as a standalone daemon instead.
2) I think I recall an option to force all old smbds to close when a new
connection is made from the same client that started the earlier smbds.
(However, I'm not certain of this.)
- Is there an option to do this sort of house keeping? (This would help
with other problems as well.)
3) Is there anything specific to WinXP that could be doing this? (Especially if we can patch it, or do a registry edit fix, or ...)
Nope seen this with NT4 and 2K clients. Cheers, Alan
