System: AIX 5.2 ML-07, Samba 3.0.21a & 3.0.21b, Workstations: Win XP Pro
SP2.
(OpenLDAP 2.3.11, BDB 4.3)
Users are in LDAP pdb. Stand alone DC, no winbindd.
This system was upgraded from 3.0.20 to 3.0.21a (and later 3.0.21b). I'm
looking for a scenario where under heavy load there could be locking
contention within Samba
Under heavy load, context switches go through the roof and we get hardly any
work done. Trussing shows a lot of locks against what appear to be tdb
files. (I'm still narrowing down which ones).
Now this could be an AIX problem, ML-08 was just released, but stopping smbd
and restarting clears the problem for awhile (read the next day).
I'm recompiling 3.0.20b and 3.0.20 to see if the problem goes away. This
system is production and never showed any symptoms like this in development.
The other thing is all the WS's hitting this server are SP2. We have another
server with *exact* same setup and XP boxes with only SP1 hitting it and not
quite the same degree of load - no problems.
I'm hoping this blurb will shake loose any thoughts on changes since 3.0.20
since a log level 10 is practically impossible with the number of machines
hitting it at once.
I do have a PMR open with IBM and so far they've only suggested APARs for
GPFS (which we're not using).
Cheers,
Bill
