[Summit] interop testing prep

Blaine Cook romeda at gmail.com
Thu Jul 3 12:54:33 CDT 2008


Does the protocol specify how this should be done, or is it just ad-hoc? I
know I ran into an issue at one point between GTalk and an ejabberd install,
where I was running two ejabberd servers for one domain, but where the
ejabberd servers were not internally clustered. The problem appeared to be
that GTalk would do a SRV lookup and try to do dialback, but wouldn't always
use the original requesting server's IP address, so the dialback would fail,
and s2s was impossible.

I think if that problem is solved, then the scalability problem is also
solved, but I'm not 100% sure.

b.

On Thu, Jul 3, 2008 at 10:03 AM, Peter Saint-Andre <stpeter at stpeter.im>
wrote:

> Mickaël Rémond wrote:
>
>> Hello,
>>
>> Le 3 juil. 08 à 18:57, Blaine Cook a écrit :
>>
>>  One test I'm really interested in seeing is "what happens when an s2s
>>> link when it's saturated?" It's the one question that I don't have really
>>> good answers for in terms of the scalability of Jabber servers, and it's a
>>> fairly real concern for some of the people I've spoken to.
>>>
>>> If you've saturated one s2s link (e.g., 100 Mbit), it should be possible
>>> to "spill over" into a second s2s link between two servers on a separate
>>> connection, giving you horizontal scalability with respect to bandwidth.
>>> Whether that actually happens is a big unknown for me.
>>>
>>
>>
>> I can talk about this. This is an implementation specific question. To my
>> knowledge only ejabberd and XCP implements something like this.
>>
>
> Well at the interop testing events we do protocol compliance testing, not
> scalability testing. But yes it's an interesting topic. :)
>
> /psa
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.jabber.org/pipermail/summit/attachments/20080703/aceeebf3/attachment.htm 


More information about the Summit mailing list