A long time ago, when I ran DirXML 1.1a, if a driver got behind, the *.tao file
associated with it would grow to huge proportions. The driver would then pound
its way through the file, record by record, and would finally truncate it back
down to an empty 8-byte file when it finished the last transaction in the file.

Looking at IDM2's behavior, this doesn't seem to be true anymore? I have a
couple of drivers that were down overnight and now have 200M+ *.tao files that
they are working through. But the files are shrinking, about 1M to 2M at a time.

Is the engine actually re-writing the *.tao file periodically to make it smaller
as it works its way through the transactions in it? It looks like it might be.

If it is doing what I think it's doing, I think that's why I'm seeing fairly
lousy performance today. Looking in iMonitor and watching the agent stats pages,
it looks like every few IDM transaction the database writes get really really
slow. The current writer thread gets the lock, then holds it for about 1-3
seconds, while other threads pile up behind it. When the lock is finally
released, everything takes off again for a little while, then the cycle repeats.

Is there any way to get the engine to go back to the old method of just running
through the *.tao file as fast as possible, then nuke it at the end?


---------------------------------------------------------------------------
David Gersic dgersic_@_niu.edu

I'm tired of receiving rubbish in my mailbox, so the E-mail address is
munged to foil the junkmail bots. Humans will figure it out on their own.