----- Original Message ----- 
From: "Davide Libenzi" <[EMAIL PROTECTED]>
Newsgroups: mlist.xmail
Sent: Friday, May 07, 2004 12:39 AM
Subject: [xmail] Re: Embeded perl filters for XMail


> On Thu, 6 May 2004, Beau E. Cox wrote:
> 
> > On Thursday 06 May 2004 11:57 am, Davide Libenzi wrote:
> > > On Thu, 6 May 2004, Beau E. Cox wrote:
> > > > I'm really not trying to BS anyone. OK, I'll make a new filter:
> > > > 'mod_xmail', that tcp's to a server 'mod_xmaild' which will
> > > > implement my perl embedding. To use mod_xmail, simply change the
> > > > filter's .tab from:
> > > >
> > > > "/where/ever/my_filter.pl"[TAB]"@@WHATEVER"[TAB]...
> > > >
> > > > to:
> > > >
> > > > "/where/ever/mod_email"[TAB]"/where/ever/my_filter.pl"[TAB]"@@WHATEVER"[T
> > > >AB]...
> > > >
> > > > I will get back to everyone soon with timings.
> > >
> > > Ouch, since it seemed to me that someone already implemented my old idea
> > > ...
> > >
> > > http://www.saltstorm.net/depo/scopee/
> > >
> > >
> > >
> > > - Davide
> > >
> > 
> > Yep. Bye (I had fun anyway ;) ).
> 
> Did anyone actually tried to compare performances?
> 

I did some Scopee benchmarking back in late 2002 to get
a hunch on what to expect from a daemonized perl interpreter
using sockets as means of comm.

The tests involved the whole chain of events, synchronously
running 100, 1000 and 10000 request. The script on the perl-side
did not do more than a 'stat' a static file though, so there
was no overhead which isn't really a real-life situation.
anyways, the best averages I got doing: 

for 0 to N do:
 (spawn client (machine A)) -> socket -> perl exec(machine B) -> socket -> client
done

....was:

where A == B (client/server: Win32/Duron 1Ghz)
~60 requests/sec

where A != B (client: Linux/K6-2 450MHz, Server: Win2K/Duron 1GHz)
~53 requests/sec


Forking off a perlscript with a set of modules the traditional way, is an expensive
op and I think you are lucky to even achieve a number of 10 execs/sec. Running
SpamAssassin this way would really be a gruesome experience. I'd figure maybe
0.5 execs / sec tops. Picking a pile of pre-compiled perl bytecode off the shelf
is a huge boost, since this is the big bottleneck.

the full benchmark details here:
<http://www.saltstorm.net/depo/scopee/benchmarks.txt>

/thomas.



> 
> - Davide
> 

-
To unsubscribe from this list: send the line "unsubscribe xmail" in
the body of a message to [EMAIL PROTECTED]
For general help: send the line "help" in the body of a message to
[EMAIL PROTECTED]

Reply via email to