[Samba] Samba 2.2.8a and 3.0.0

[William Jojo]

Samba Team,

How do I create simulated load on a samba server? I wish to use
feedback-directed program restructuring on AIX 5.2 to determine if
performance can be improved by rearranging the executable generated by the
IBM 6.0 compiler.

Mostly I'm determining what the upper bound on smbd's is in a
multiprogramming environment. I'm currently running about between 800 and
1000 smbd's a day on a 6-way 6H1 (RS64-III) with 24GB memory and 800 GB of
attached SSA disk for roaming profiles. All of this is being done with
2.2.8a.

I'm setting up my 3.0.0 test area on similar hardware, but wanted insight
from the experts on what you've done already to improve internal
performance as my approach is to improve cache hit ratio and translation
look-aside.

I've already tuned every piece of hardware to my abilities and even
implemented write-behind algorithms. I'm basically wringing out the towel
to see if there are any drops of peformance left to be found.

Are there any internal limitations with the TDB's, et al, where there may
be an upper bound on concurrent smbd's?

Also are there any internal performance enhancements for smbpasswd files
of >15000 entries?

We are looking at LDAP for storing passwords. What kind of improvement can
I expect when moving passwords from a flat file to a dynamic container?


On a related note, you guys rock! You are the reason we can do the above
with one server! You have saved us hundreds of thousands of US $'s on
licensing and servers alone! Your technical prowess is appreciated far
more than you know!


Bill