I added a new mail_prefetch_count setting to Dovecot v2.1
(http://hg.dovecot.org/dovecot-2.1/). Its main purpose is to avoid
unnecessary latency while waiting for replies from remote server with
imapc, i.e. instead of doing (FETCH, wait, FETCH, wait, ...) it can do
(FETCH, FETCH, .., wait). For the same reason it's also going to be very
useful once I add support for key-value databases / other high latency
cloud filesystems.
It also works with Maildir, single-dbox and cydir backends by calling
posix_fadvise(POSIX_FADV_WILLNEED) for the files. This syscall is
apparently only implemented by Linux. It should tell the kernel to start
reading the files from disk to page cache. By calling this for multiple
files, it should help the kernel optimize disk I/O. For example if 10
files are prefetched, the kernel might figure out that it should read
them in a different order to avoid extra disk seeks. Perhaps it also
does something useful with NFS, I don't know.
I'm not sure if it would be a good idea to use this for mbox or
multi-dbox files, since they already have multiple messages in same file
so kernel probably already did some prefetching. With mdbox I guess it
could be useful if the message file changes..
This optimization currently works only for FETCH command (and POP3 RETR
and TOP). SEARCH/THREAD/SORT commands will hopefully implement it soon
also.
I'd be interested in knowing what the effects of this setting are if
anyone tests this some day.
