Load testing with Tsung showing low simultaneous users *FIXED*

As part of an evaluation at work, we're running a simple Tsung script against various XMPP servers in order to pull together some figures to inform our final choice. My colleague is currently testing Openfire, and I'm working on ejabberd. We're sharing the same Tsung script (more or less) and getting differing results.

When the script is run against Openfire, the report is showing figures for simultaneous users exceeding 10000. However, when I run the script against an ejabberd instance, I'm not able to push the simultaneous count much above 100. I've configured the ejabberd instance to use internal, then external authorization using MySQL with no difference. I've run it on a 2.4Ghz CoreDuo machine with 3GB of RAM, and a VM running on a MacBookPro (yea, I know, how *enterprise* is that), both running the latest release of Ubuntu. I'm using both the Erlang(12b) and ejabberd(2.0.3) versions from the Ubuntu repositories. I've been running in a single node configuration up to this point. Nothing I tweak or change is effecting the number reported.

The issue for me is that both the ejabberd and Openfire tests are being run on the same hardware with the same script but producting very different results. I know it must be down to my configuration on the ejabberd side. I can produce any portion of the ejabberd configuration or Tsung script for anyone who's interested.

Thanks,
Gareth

Update: Turns out that having thought I'd updated the file handle limit on the machine, the change hadn't stuck for the ejabberd user. Looking at much better figures now.

Figures

Hi Gareth,

What kind of better figures did you get with ejabberd? I'd really love to know as I am planning out hardware needs for a major ejabberd deployment.

Hope to hear from you!

Karthik

Yeah, really need to

Yeah, really need to know.

But I highly doubt that user is going to return. That was a 2009 post.
I'm a bit late I guess.

Syndicate content