[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Sugesstions building a rather big mail system.

Am Mon, 2003-10-06 um 16.51 schrieb Theodore Knab:

Most things seem quite good - anyway a few questions/comments:

> Since you are familiar with LVS, you should have no problem setting 2
> [redundant] LVS systems up. You could balance the load between 10-20
> IMAP servers.

I would also suggest LVS as I stated in my other posting - use
keepalived to get your balancers redundant.

But 10-20 IMAP server (each of them dual Xeon)? Seems "a little bit"
In my experience the POP/IMAP server are those machines with the least
load. The need some I/O so you definitely want to use GBit for accessing
the data via network.

> You might also be able to use the same 2 LVS systems to balance your
> load between the Web-mail servers.
> <Crude Diagram>
>    [Firewall]
>        |
>        |	
>        |
> [LVS1]----[LVS2]
>   |         |
>     [Fiber Only Switch]
>    	|

Why fiber-only? Gbit copper is MUCH cheaper and as all server are
probably side-by-side in a few racks I don't see we you would need fiber

> Estimated Minimums needed for 500,000+ Email Users
> --------------------------
> 10 IMAP servers [Courier IMAP 1 [Dual Xeon 1GHz] server /200 active users]  
> 	w/ XFS filesystem and Debian Stable

As said before - 10 seem a little bit much. You can add more server to
your pool anyway.

As for the FS: XFS/ReiserFS/ext3 shouldn't matter as all files will be
stored on a SAN/NAS anyway.

> 20 Webmail Servers [Squirrel-mail 1 [Dual Xeon 1Ghz] server /100 active users]
> 	w/ XFS filesystem and Debian Stable

If you want to provide webmail too... 20 Dual-Xeon seem a little much
again as probably not all users will use HTTP to access their mail if
they can use POP/IMAP instead.

> 2  Databases Servers for authentication either [Mysql or OpenLDAP]
> 	w/ XFS filesystem and Debian Stable

Redundant setup or a clustered approach with multiple read-only and one
master-server if the DB queries become too intensive.

> 2-4 MX Gateways running either Exim or Postfix MTA and SPAMD with
> 	w/ XFS filesystem and Debian Stable Amivisd

Now that's IMHO not enough for 500k user...
If you do spam-filtering, virus-scanning and maybe even filtering
(trough procmail or something else) on your MX you'll definitely need
either some quite powerful machines or some more smaller ones.

> 2 [Fiber Channel] SAN Volumes for [MAIL storage] redundancy. 
> </Crude Diagram>

Or NAS or even "self-build" NFS server.
If you want to access your FC SAN from all your MX and POP/IMAP server
it could become "a little bit" expensive if you consider the bunch of
FC-HBA and FC-Switches you'll need...

Anyway, as I said in my other post: Hardware dimensioning is something
which will definitely consume quite a bit of time as you don't want to
have any bottlenecks nor spend huge amounts of money on something you
don't need. Maybe talk a few hours with someone (real-life, not ML) who
has build something similar (not necessarily a mail-cluster, but
something with huge amounts of data on a network) as they probably know
most of the pitfalls first-hand :-)

best regards,
Markus Oswald <moswald@iirc.at>  \ Unix and Network Administration
Graz, AUSTRIA                     \ High Availability / Cluster
Mobile: +43 676 6485415            \ System Consulting
Fax:    +43 316 428896              \ Web Development

Reply to: