[Dovecot] new configure-option --disable-index

[Marten Lehmann]

Hello,

I experimented with dovecot the whole evening yesterday and the 
mailserver worked fine since there haven't been that much users 
connecting to it. I'm using dovecot 1.0.alpha5 as pop3 and pop3s server. 
The maildirs are mounted through NFS and since there is no nfslockd 
running I disabled mmap in the config and told dovecot to use dotlock. 
Also I experienced, that dovecot was much faster when using one 
login-process per connection instead of sharing them throughout several 
connections. Sharings seems to be a greater performance problem 
(bottleneck) than additional fork()s.

However, under normal load at business time the server load increased to 
a value that made it impossible to download emails without timeouts, so 
we switched back to tcpserver/qmail-pop3d for pop3 and only kept dovecot 
for doing the pop3s part.

Dovecot is really fine, but it still isn't NFS ready. Especially when 
using dovecot as pop3 only this index-stuff must be disengageable. What 
is the use of generating a complex index file for hundreds of messages, 
that will be deleted in the next step after downloading them? But the 
index is also a problem when using IMAP over NFS. Access to the files is 
too slow, so the connections keep hanging around and the i/o waiting 
value is raised to a maximum which slows down the whole machine. It 
simply doesn't make sense to build and update an index every time the 
maildir is accessed (which is very cpu and timeconsuming), while search 
operations (the only moment when the index is actually used and not just 
heavily updated) don't affect even 1% of all IMAP calls (and such a 
command doesn't even exist for pop3).

I would be much obliged if you could create an option called 
"--disable-index" in the configure-script or something like 
"disable-index = yes" in dovecot.conf that makes dovecot ignoring 
existing index files and doesn't care for related operations, so NFS 
should work much faster and IMAP search would simply parse the files 
directly instead of looking in the index.

I would really appriciate this very much and I'm absolutely sure, that a 
lot of people beside me are waiting for this.

Regards
Marten