[ejabberd] Cluster and load balancing - help needed with the hardware architecture

f.m.gomes at controlrede.com f.m.gomes at controlrede.com
Sun Jul 2 01:04:01 MSD 2006


Thanks for your reply!
Please correct me if I'm wrong, but what VRRP does is router redundancy, but 
at a single moment, only one router has all the connected clients passing 
through it, so it can be a bottleneck... In my first configuration, I also 
have a redundant load balancer, using IPVS and UltraMonkey, if one of the 
load balancers fail, the other gets its job done, but my doubt is that if we 
build an ejabberd cluster in order to support many clients connected, and 
the connection goes through a single machine (load balancer) before getting 
to the cluster, that machine can be the bottleneck, specially if we have 
permanent connections, as is normally the case with IM clients, not so much 
the case for web clients. I'm refering to scalability, the redundancy seems 
to be guaranteed, with my design.

Fernando

----- Original Message ----- 
From: "Alexandre LISSY" <alexandrelissy at free.fr>
To: <f.m.gomes at controlrede.com>; <ejabberd at jabber.ru>
Sent: Saturday, July 01, 2006 9:21 PM
Subject: Re: [ejabberd] Cluster and load balancing - help needed with the 
hardware architecture


> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> I know that VRRP can solve the loadbalancer dependency, but never 
> experienced it.
>
> f.m.gomes at controlrede.com wrote:
>> I'm designing the hardware architecture for a ejabberd cluster, and I
>> need your advice about it. We are planning to have a redundant load
>> balancer (using heartbeat and IPVS) connected to the internet (a single
>> public IP address), and on a second interface (Gigabit ethernet) we will
>> have the ejabberd cluster. Is this the usual setup for ejabberd clusters?
>> I never built a ejabberd cluster in the past, so I'm using the 'normal'
>> clustering architecture for web applications. Since the IM clients have
>> almost permanent connections to the ejabberd server, won't this
>> architecture could be an issue to the load balancer themselfs? A round
>> robin DNS can be preferable, since the connections are done directly to
>> each ejabberd server in the cluster, and not through the load balancer?
>> In this case each ejabberd server must have a public IP address, I need
>> a redundant DNS for the cluster, but the permanent connections shouldn't
>> cause problems as in the previous setup...
>> Any 'expert' tips for this situation?
>>
>> Thanks
>>
>> Fernando
>>
>> ----- Original Message ----- From: "numE" <numE at onionizer.de>
>> To: <ejabberd at jabber.ru>
>> Sent: Friday, June 30, 2006 9:32 PM
>> Subject: Re: [ejabberd] Patch nudge
>>
>>
>>>
>>>> Are many people running on the SVN version currently ?
>>>
>>> we, too :)
>>>
>>>
>>> _______________________________________________
>>> ejabberd mailing list
>>> ejabberd at jabber.ru
>>> http://lists.jabber.ru/mailman/listinfo/ejabberd
>>>
>>>
>>>
>>>
>>> -- 
>>> No virus found in this incoming message.
>>> Checked by AVG Free Edition.
>>> Version: 7.1.394 / Virus Database: 268.9.7/379 - Release Date: 
>>> 29-06-2006
>>>
>>>
>>
>> _______________________________________________
>> ejabberd mailing list
>> ejabberd at jabber.ru
>> http://lists.jabber.ru/mailman/listinfo/ejabberd
>>
>>
>>
> -----BEGIN PGP SIGNATURE-----
> Version: GnuPG v1.4.2 (MingW32)
> Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org
>
> iD8DBQFEptlgfP+GwT1GJ3ERArzeAKCQPTThtKPoLq7e9kj8olmA/4AIjgCfZKgc
> sQ5dz+yKWX0p5sKi+Gz/O+Y=
> =XsQR
> -----END PGP SIGNATURE-----
>
>
>
>
>
> -- 
> No virus found in this incoming message.
> Checked by AVG Free Edition.
> Version: 7.1.394 / Virus Database: 268.9.8/380 - Release Date: 30-06-2006
>
> 



More information about the ejabberd mailing list