filter_innd.pl question (MT)

[Weber, Markus]

Russ wrote:
> innd is single-threaded because writing threaded code properly is a
> lot more complex and INN dates from 1991.  :)

"Properly" is the right wording here ... Not sure if it's worth cleaning
up INN to get it MT safe. Beside some added functionality the only
reason
I've come along so far is when you get CPU bound and have a few spare
CPUs
to burn ... however, it might require quite some more flow redesign to
get
it "fine grained".

We took the 2.4.1 base and "coded" (well, you can't call that properly
co-
ded) some MT stuff in to have ARTparse run in parallel as our box got
over-
loaded @ 175mbps incoming to find the newlines ... it "kind of works but
not too stable, sometimes puts some fancy output in the logs ... still
ok
for the time spent on it". As the post stuff wasn't a bottleneck we've
protected that whole part with a single mutex ... to keep it easy.

However, don't most people nowadays buy "cheaper" x86 boxes with just
one
or two "blinding fast" CPUs? Would the added functionality justify a
code redesign or better find other working solutions?

I remember MT on the wishlist for some upcoming releases -was it 2.5?-
but it looks like those wishes are gone. Does anyone knows what happend
to
them?

Markus

--=20
[E] Markus Weber <fvd at de.kpn-eurorings.net> [IRC] FvD [P] <fvd at icmp.de>