[isf-wifidog] WiFi Dog auth server mirroring

[Chandi Bernier]

There seems to be a software called PGCluster that does it. It's a
multi-master replication/load balancing solution.

It's open-source but there is not much documentation on the web. The main
programmer answers questions on the mailing list. 

http://pgcluster.projects.postgresql.org/

Chandi Bernier


-----Original Message-----
From: wifidog-bounces at listes.ilesansfil.org
[mailto:wifidog-bounces at listes.ilesansfil.org] On Behalf Of Benoit Grégoire
Sent: 19 août, 2005 12:54
To: wifidog at listes.ilesansfil.org
Cc: Rein Petersen; Neil Lee; Patrick Dinnen; Phillip Smith
Subject: Re: [isf-wifidog] WiFi Dog auth server mirroring

On August 19, 2005 10:24 am, Patrick Dinnen wrote:
> Hi
> There has been some discussion in Wireless Toronto about the potential
> usefulness of having a mirrored/redundant auth server setup, so there
> isn't the single point of failure for the system.
>
> Does anyone have any experience and pointers on setting up something
> like this? I gather that there may be an option to setup the system so
> that each gateway router knows about an alternate auth server, in case
> of network/server problems with the first. Obviously there would be a
> requirement to sync the WIFi Dog DB between the two machines somehow.
>
> Any thoughts or instructions from anyone who has looked at this?

As Mina already told you, redundant servers are already supported on the 
gateway side, and already usefull if you have redundant connectivity to your

main auth server.

However, having truely redundant auth servers is problematic.  The main 
problem is that you need multi-master replication of the database.  There 
have been a number of changes in the schema to avoid colisions in this 
scenario.  Unfortunately, there still isn't any free software solution to 
enable this for either Postgres or MySql.  So we are waiting for the
releaase 
of Sloony II.  

The next best would be to use Sloony I, which allow transfering the Master 
function to the slave, and back again.  Unfortunately, this would require 
extensive and fragile application logic so the BOTH servers can
independently 
detect that they are down.


-- 
Benoit Grégoire, http://benoitg.coeus.ca/