[nycbug-talk] Mail stress test

[Pete Wright]

On Fri, Feb 25, 2005 at 01:51:11PM -0500, Steve Rieger wrote:
> Ok so nobody knows, and it might not even be possible to max out the
> concurrent connections for those protocols.
> 
> So I will take a step back and ask do you know of a way I can create 100
> concurrent connections for each of those protocols, on these servers.
> 

this still may be possible with the perl modules (i assume there
are modules for other langs as well).  for example, the Net::IMAP 
module from CPAN pretty much give you all the functionality of an
imap client. so you could setup on the server a test account with
messages in it and have to script connect, get a listing of messages
etc.  i believe the same is true for Net::POP etc.  so i guess i'm
suggesting that by using libraries like this you should be able to 
script something up.  i'm sure one of the python or php guru's here 
know of similar way's to do this.

now will this give you a good measure of how each server will handle
a live load...i dunno.

-p

> 
> 
> Thanx
> 
> 
> 
> 
> On 2/25/05 1:36 PM, "Bob Ippolito" <bob at redivi.com> wrote:
> 
> > 
> > On Feb 25, 2005, at 1:20 PM, Pete Wright wrote:
> > 
> >> On Fri, Feb 25, 2005 at 12:36:36PM -0500, Steve Rieger wrote:
> >>> Does anybody on the list know of a decent script I can run to stress
> >>> test my
> >>> smtp, pp, and imap servers,
> >>> 
> >> 
> >> hey steve,
> >> sorry i don't know any off the top of my head, altho
> >> i do know that there are ample perl modules that offer this
> >> functionality (smtp, pop and imap).  the problem may be tho
> >>  that you will end up stressing the client running the script
> >>  before you end up stressing the server.
> > 
> > The number of TCP connections that it'll accept is likely to be much,
> > much, much, much higher than the number of active connections it can
> > deal with.  Unless the application is written particularly poorly,
> > you'll hit kernel limitations (but maybe just a soft one, like the
> > per-process descriptor limit) before the application will notice.
> > 
> > Ideally you really need a benchmark that actually attempts to do
> > something with the server, like all of the http benchmarks do.  Though,
> > i can't say I know of anything for other protocols.
> > 
> > -bob
> > 
> > 
> > 
> 
> -- 
> Steve Rieger
> Office 212-804-1131
> Cell 646-335-8915
> DC 173*101254*4
> 
> 
> _______________________________________________
> % NYC*BUG talk mailing list
> http://lists.nycbug.org/mailman/listinfo/talk
> %Be sure to check out our Jobs and NYCBUG-announce lists
> %We meet the first Wednesday of the month

-- 
~~oO00Oo~~
Peter Wright
pete at nomadlogic.org
www.nomadlogic.org/~pete
917.415.9866