Re: [toaster] Qmail Cluster

[Bill Shupp]

Andy Abshagen wrote:

 

-----Original Message-----
From: Jason 'XenoPhage' Frisvold [mailto:[EMAIL PROTECTED] 
Sent: Thursday, December 23, 2004 8:32 AM
To: toaster@shupp.org
Subject: Re: [toaster] Qmail Cluster

Andy Abshagen wrote:
   

Bill and others,
Question on the Cluster design.  We are looking for 
     

something very similar
   

to this however we are wanting to put the mx servers at 
     

different physical
   

locations.  This causes some big questions as far as the 
     

ability to send nfs
   

traffic across the net.  Not sure it is a wise idea at all 
     

however for
   

security the traffic would be going across a vpn tunnel from 
     

site a to site
   

b.  I guess to go along with that would it be better to just 
     

have a relay
   

sitting at the second site that doesn't do a mail delivery 
     

at all just holds
   

mail with smtproutes to the main cluster/server at the main 
     

site?  That
   

would bring along its own set of problems in the fact that 
     

we want the
   

chkuser functionality in place so that we don't have the thousands of
doublebounces every day.
     

Why a separate site?  I understand you want redundancy, but 
this method 
still has a single point of failure in the NFS server...
   

A separate site for cases when the primary site is not available at all.  No
matter how redundant you make a facility there are still chances that it can
go totally dark.  Because of our services we already have multiple sites.
Also with our client base they do not want a down time at all.
 

Well, if you want to distribute the load of incoming smtp connections 
and other services, you should have the nfs client nodes in one location 
for security and performance.  But if you want backup mx machine, then 
it's feasible to use a remote location for the backup.

If you want to distribute your mail spool over multiple data centers, 
you should not use a centralized file system like this one.  You should 
rather look at something like Qmail-LDAP, where the mail spool is 
distributed over multiple systems.  You will still not have complete 
redundancy, but if one data center goes down, then part of your mail 
will still be up.

I personally think that if you are worried about downtime, just get a 
REALLY GOOD colo company, and hold them to their SLA.  You'll get better 
performance, and have a better design.

Regards,
Bill