makehistory / architecture change
rra at stanford.edu
Mon Oct 23 05:41:10 UTC 2006
Bill Davidsen <davidsen at tmr.com> writes:
> I have often wanted a more steady state way to do this which would
> eliminate the whole concept of batches. Fortunately I did this so
> infrequently that I never got to doing anything about it. Large batches
> definitely help performance, but the whole makehistory process is so
> slow on a moderate size spool that it makes only minimal
> difference. Even making history without overview is pretty painful.
What takes forever is touching each message on disk; once you've read the
message in to get the history information, also generating the overview
information is trivial and only incurs an extra write (which can be punted
to another overchan process).
I don't think makehistory is ever going to be fast; that's why it's so
important to try to make it unnecessary as often as possible.
Russ Allbery (rra at stanford.edu) <http://www.eyrie.org/~eagle/>
Please send questions to the list rather than mailing me directly.
<http://www.eyrie.org/~eagle/faqs/questions.html> explains why.
More information about the inn-workers