[reSIProcate] Load Testing Using Resiprocate

[Dennis Dupont]

Some clarifying questions: 
 - are you using dum or just resip?
 - which version are you running?
 - do you have a stack trace?
 - have you run your code under valgrind to look for memory corruption?
 - do you mean 10 call setups/second per user (1000/second) or just
10/second overall spread over 100 users?
 
I have tested my B2BUA under loads of up to 60 calls/second on a dual
Xeon LINUX system. In fact, I actually 
ran 3 instances of my program on this machine, each at 50 calls/second
with a load balancer in front of them.
So I had a total load of 150 CSPS on the system.  We split the program
up only to have less calls affected
by a software failure.  This configuration ran for several days with
over 10 million calls.  I never even had any 
swapping and ran all within just over 1GB of physical memory (2 GB on
the system).  My system is built with 
0.9.0-5019.  However, I am not using dum, so I cannot comment on that
part of the software.
 
Before I achieved that stability I also had crashes similar to what you
describe.  It was almost always in the
stack, but I was pretty sure I was stomping on the stack.  Sure enough,
I used valgrind to find that I was 
writing to released memory.  Since resip uses the heap intensely and my
application only uses it rarely,
resip was the most likely victim of my defects.
 

-----Original Message-----
From: Asheesh Joshi [mailto:asjoshi at varaha.com] 
Sent: Thursday, September 29, 2005 12:41 AM
To: resiprocate-devel at list.sipfoundry.org
Subject: [reSIProcate] Load Testing Using Resiprocate


Hi All,
 
    Over the past one week I am trying to get my B2BUA using the resip
put under load test. There are
some issues  I am seeing with the memory creeping up slowly and slowly. 
 
    My team is still trying to find out is it our UAS / UAC state
machines that is slowly eating up the
memory. Meanwhile I thought I will write in this mail and seek
suggestions/information from you guys
on past observed behavior of Resip under load testings.
 
    We have a tool generatting load at 10 calls per second for 100
users... each call cycle being defined by
INVITE, 100 trying, 180 ringing, 200 Ok, ACK, ( Pause ), BYE, ACK.
 
    The B2BUA is running in a Linux machine and we monitor the memory
using "top" command
and I see memory rising .1% to .5 % after each batch of 100 calls.
Eventually after 12,000 calls
the system crashes !
 
    I wanted to ask the forum that do we have any statistics on RESIP
being tested under high
traffic for long periods ? And has any kind of memory leak testing being
done on Resip. Can
somebody suggest me some good tools ?
 
    Any help will be appreciated.
 
-cheers
Asheesh 
 

 
Asheesh Joshi
SCA-Voice Development.
Varaha Systems India Pvt Ltd.
 <http://www.varaha.com/> www.varaha.com
mail:  <mailto:asjoshi at varaha.com> asjoshi at varaha.com
 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.resiprocate.org/pipermail/resiprocate-devel/attachments/20050929/9b8ec2f3/attachment.htm>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: VScompanyLogo.gif
Type: image/gif
Size: 4644 bytes
Desc: VScompanyLogo.gif
URL: <http://list.resiprocate.org/pipermail/resiprocate-devel/attachments/20050929/9b8ec2f3/attachment.gif>