We use Alteons to do the load balancing and a pair of NetApp F810c's for
the HA NFS system. I won't include any details since its easy enough to get
the data from the guys at Netapp. We run 7 qmail/vpopmail servers with 3
mysql servers (one master and 2 slaves that are read only).
----- Original Message -----
From: "Duane Wylie" <[EMAIL PROTECTED]>
To: <[EMAIL PROTECTED]>
Sent: Monday, December 30, 2002 1:41 PM
Subject: RE: [vchkpw] vpopmail clustering examples
> Sorry 'bout sending that last message as 'Super-User'...
> We're using F5's load balancer and it's been working great for us. It has
> similar features as the Foundry switch you mentioned (and the service
> monitoring is great!).
> The main worry I have at this point is the NFS server. Even though the
> system in question is RAID 5, and is in a separate disk enclosure, I don't
> like having that one NFS server as a single point of failure.
> Is anyone using some sort of HA storage solution? If so, care to share
> Duane Wylie
> -----Original Message-----
> From: Andre Fortin [mailto:[EMAIL PROTECTED]]
> Sent: Monday, December 30, 2002 11:53 AM
> To: [EMAIL PROTECTED]
> Subject: RE: [vchkpw] vpopmail clustering examples
> We have a very similar set up. We have a third machine as the MySQL
> which doesnt directly interact as a mail server, but it makes the MySQL
> database triple-redundant without sacrificing performance.
> For load balancing/redundancy, we are using Foundry
> (http://www.foundrynetworks.com/) switches with the SLB (Server Load
> Balancing) code. Very powerful; it not only detects if the server is
> running, or listening to port 25/110, but it actually understands SMTP and
> POP3 and will make sure the SMTP and POP3 servers are responding properly
> part of its 'heartbeat' tests.. So, for example, if you had a machine that
> had a problem and would still listen to port 25 but couldnt actually
> with a '220', it would be treated as 'offline' instead of opening dead
> connections for clients.. It also allows you to load balance between X
> number of servers rather than just failing over in 'worst case scenario'