From: "Michael Fair" <[EMAIL PROTECTED]>
Date: Sat, 7 Jul 2001 13:24:04 -0700
[...]
This is pretty much what I was expecting. I would also imagine
that without a good tool migrate several thousands of mailboxes
changing the on-disk structure is not really an option. I
personally believe that we would see a performance gain by
changing the on-disk structure, but believe that can also be
delayed till some time down the road.
Without actual benchmarking, it's really impossible to say that the
performance of the alt-namespace code is a concern.
Regardless, we will not be dropping support for the old namespace or
the old hierarchy seperator any time soon.
Understanding the allocation patterns and improving them would
probably improve speed more (though the namespace stuff undoubtably
adds malloc() calls). libsasl v2 goes part of the way to reducing
random mallocs.
Minimizing disk i/o is a much bigger win. For instance, making
cyrus.cache include the entire header (not just selected portions) and
removing some of the pre-parsed gunk would probably increase speed
dramatically for Outlook/Outlook Express/Mozilla clients.
Threading/servicing multiple clients from a single process is another
obvious place for improvement.
Larry
