Using the Lock Manager for the TDB data

[Michael D. Ober]

Thanks for analyzing this and keeping the community informed of the decision
and also why the decision is made the way it is.

Mike Ober.


-----Original Message-----
From: COLLOT Jean-Yves [mailto:Jean-Yves.COLLOT at cofiroute.fr] 
Sent: Tuesday, April 22, 2003 10:35 AM
To: Samba VMS
Subject: Using the Lock Manager for the TDB data


Hi.

Some time ago, there was a couple of messages in this mailing list about the
idea of changing the TDB processing in Samba/VMS, in order to remove some or
all of the current RMS indexed files and use the VMS Lock Manager instead.

I used the PCA tool for evaluating what kind of performance enhancements we
could get, and the result is that the SMBD process spends about 5% of its
time dealing with the LOCKING.TDB file, and close to nothing for the other
TDB files.

Using the lock manager is undoubtedly an excellent idea. However, after
looking a little more into that issue, I think that it would be very
complicated to do. Changing ALL the TDB files by locks is not correct,
because some information (in SECRETS.TDB) must stay after reboot. So I
focused on the LOCKING.TDB stuff, because it's there that the performance
enhancements would be sensible.

My opinion is that would mean a complete re-writing of the LOCKING.C and
BRLOCK.C modules, and I don't even know exactly how, because as I pointed
out previously, the amount of data kept in the file is bigger than the lock
value block.

Anyway, it is impossible for me to justify to my boss the amount of work
that would be needed for doing that, the ultimate visible goal being a 5%
gain in performances, knowing that the Samba/VMS users here are now happy
with the existing performances.

The conclusion is that I will not do it, after all. 

PLEASE READ THIS IMPORTANT ETIQUETTE MESSAGE BEFORE POSTING:

http://www.catb.org/~esr/faqs/smart-questions.html