INN not keeping up with incoming news

Fabien Tassin fta at sofaraway.org
Tue Jan 30 16:05:41 UTC 2001


According to Katsuhiro Kondou:
> 
> In article <Pine.LNX.3.93.1010129232154.390A-100000 at outland.tmr.com>,
> 	Bill Davidsen <davidsen at tmr.com> wrote;
> 
> } Any thoughts, tips, or bits of hope for multithreading would be
> } delightful.
> 
> As I wrote before several times, I'm rewriting incoming
> processing within innd to avoid wasteful memory copying.
> I've written the code already and now debugging.

Did you changed HISxx() code ?

>  If your
> server consumes much cpu time also at ARTclean, the patch 
> should resolve your problem as well as at ARTclean and at
> NCproc() which is invisible in innd timer.

My ARTclean() takes around 5%. I don't know for NCproc() but I can
add a timer for it (I'll probably do so). My current goal is to reduce
HISxx() usage. Looking at your own stats, your numbers are 2 orders
of magnitude lower than mines. Is your server already running your new
code ?

>  Meanwhile I
> tried to modify the code so that we can use it also as
> multithreaded, but it may be incomplete.  As to multithreading,
> r$ or someone said it'd be better to write from scatch.
> But I think I may write it based on the current code,
> although it'd take much time to accomplish due to shortage
> of my free time.

I've recently written a basic news server to benchmark innfeed. It supports
both IHAVE and CHECK/TAKETHIS protocols but has neither history nor storage
functionnalities as it is not its goal. BTW, the first implementation
was using fork() and was damn faster than the sequential approach of innd.
I'll try the multithreading way to compare. I'm now tempted to use the storage
API and history code of INN to produce an innd alternative for transit only
servers. I haven't done it yet because I'm not satisfied with INN history.
I'd like to have a kind of history daemon mixing the current INN
history and expire functionnalities. It would be nice to be able to work
at constant sizes for everything (history, spool and perhaps overview,
whatever the methods used). I like the concept behind Olaf 'thdexpire'.
By extending it, it could be possible to avoid time consuming cleaning
processes.
For now, it is a dream and I'm still working with/on INN code.

-- 
Fabien Tassin -+- fta at sofaraway.org


More information about the inn-workers mailing list