On Tue, Aug 26, 2014 at 9:50 PM, Elvir Bahtijaragic <el...@bahtech.com>
wrote:

>  Hi Waqas,
>
>
>  Actually the project was added about 3 months ago (
> https://github.com/Kixeye/kixmpp/graphs/commit-activity).
>
>
>  This project has an interesting history. Initially it started as an
> efficient load testing client for testing Openfire - we are able to
> simulate 15000 clients on a xlarge instance on AWS. Other XMPP clients were
> unable to handle a large number of XMPP clients on a single instance, i.e.
> Smack had tons of memory leaks and Tigase was extremely inefficient in
> processing stanzas. Soon after load testing Openfire we realized that it
> performs extremely poorly under load and is completely unstable in a
> clustered environment. So we dropped Openfire and implemented our own XMPP
> server - using high-performance computing design patterns. Our requirement
> was to reach 500,000 simulated chatting clients - but we never actually hit
> a limit in our load tests. Based on our load testing parameters: a million
> simulated clients caused the server to output 230,000 message stanzas per
> second with a load average of 6 on a 32 core box using only 30% memory on
> average.
>
>
>  Thanks! We will add more documentation soon. Currently it's meant to be
> used as a library and enable any Java application to embed a XMPP client or
> XMPP server.
>
>
>  Regards,
>
> Elvir B.
>
That's very informative. The performance and memory usage numbers are
impressive. It would be great if you could publish your load testing tools
as well, so other server vendors can compare. I know I'd like to benchmark
Prosody, the server I work on.

Unless I missed it, the server is missing server-to-server federation. Is
this something you plan to add?

Again, great job on the project and thanks for releasing this. I like the
code quite a bit. My biggest complaint so far is mixed tabs and spaces in
many of the files :)

--
Waqas Hussain
_______________________________________________
JDev mailing list
Info: http://mail.jabber.org/mailman/listinfo/jdev
Unsubscribe: jdev-unsubscr...@jabber.org
_______________________________________________

Reply via email to