On 12.7.2007, at 10.07, Arkadiusz Miskiewicz wrote:
I also did a bit of benchmarking. v1.1's performance improvements are looking pretty great, it seems to be twice as fast as v1.0.
Could you somehow test system usage resources, too? That will be
probably problematic to measure.
It is problematic to measure. Feel free to figure out how to do that. :)
For example being twice fast and eating four time more resources
isn't a great win. That bit me when I wanted to switch from tpop3d to dovecot for
pop3 (already done similar switch for imap). It looks like dovecot does
much more I/O (even when delivery is used at MTA level so indexes are
available) which caused huge problems (unsolved unfortunately).
Yes, Dovecot's hasn't been a very good POP3 server. With v1.1 there
are several improvements that should make it a lot better, although
maybe still not perfect.
Although I'm guessing your "much more I/O" means that Dovecot reads
the message contents to calculate the messages' correct virtual size,
while tpop3d violates POP3 spec by returning physical message sizes.
Also Dovecot 1.0's deliver doesn't help with this, because it doesn't
write the virtual size to cache file.