[Semi-OT] Advice on large webmail setup

[am.lists]

Another alternative that would be Linux (as opposed to somewhat
proprietary HP/HP-UX implementation)

I know that you have "no further information" -- but based purely on
webhits (which we can interpret as some ratio of sending vs receiving
mail, image (gui element) retrievals, etc.),

Doing purely bogus linear math, 40M webhits / 600K users = only 67
webhits per user per month. That number could be attained by all of
the users hitting the site at exactly the same time, or perfectly
evenly distributed over the course of the month. Having run this stuff
myself, I would say that probably 10% of the users make up 90% of the
traffic. Maybe 20/80, but it's still some curve that looks like this.

That being said, let's take a 20% estimate. 40MM/mo * 80% = 32MM just
for that 20% of the users. 20% of the users = 120K users. Let's say
that the 20% of the users are working the 10-ish hour day x 5
days/week and concentrate that usage during that time.   Further
reducing this down, I would estimate a number that says you need to
support approximately 2750 gets/posts per second on your web front
ends.

Now. How to get 2750 gets/posts per second.  Apache surely can achieve
this but you have to spread it out. I, like Res, am a fan of Foundry's
products. But I'm also a fan of the F5 line too. The F5 Big-IP stuff
is quite comparable to Foundry in terms of functionality, but I favor
the F5 on management interface.

My setup would look like this:

"Hardware load balancer out front"
Webmail farm:  10 x Dual-Proc Intel/AMD-style boxes, Linux (CentOS 4
or RHEL 4). As for the software, I'm not an expert in Web GUI for
mail. I've used Squirrel, Horde, and a slew of commercial ones.
Whatever is suitable.

---Firewall to protect the frontends from the backends---
Backends:

Centralized MySQL on a quad-cpu box, 16GB memory.
Qmail/Mailscanner/SA: distributed across multiple boxes... but with
centralized storage
Storage: If you have 600K users with 10MB each, that pulls just under
6TB of data. The HP MSA stuff can give you that kind of storage in a
mini-SAN (or NAS if you like) configuration. This could readily assign
enough storage to both the SQL and Mail Store file systems.

Angelo