On 09/04/2012 03:39 PM, Jeremy Harris wrote:
Any performance numbers?

The change, by itself, shouldn't have a significant performance impact
for a reasonably sized Exim system. The main purpose of the change was
to shunt -K files off to tmpfs, so there's one less variable to
consider for performance tuning the spool filesystem.

Assuming no TLS, the outgoing path is roughly:
-H + -D (disk) -> -K (memory) -> DKIM header + sendfile -K (socket)

It would be interesting to see what benefits could be gained by
switching -D to be an SMTP on-the-wire representation of the data
portion. That would allow use of sendfile/splice further up the
outbound processing stack, DKIM signing could switch over to using
mmap, and there are perhaps optimizations for the inbound path in the
case of receiving SMTP data, queuing + header changes, and then
shipping the same data portion back out over SMTP.

Paul

--
## List details at https://lists.exim.org/mailman/listinfo/exim-dev Exim 
details at http://www.exim.org/ ##

Reply via email to