Enterprise scalability

[Jeff Earickson]

Question here...  Are the 40 boxes all NFS mounting one common /var/mail
filesystem?  Or are the mail spools spread across 40 machines?  If NFS,
isn't the machine with the shared NFS filesystem a chokepoint in your
setup?  If the mail spools are spread across 40 machines, how do you
determine which MX takes email for what user/machine combo?

Jeff Earickson
Colby College

On Thu, 26 Feb 2004, Michael Baird wrote:

> Date: Thu, 26 Feb 2004 08:43:12 -0500
> From: Michael Baird <mike at TC3NET.COM>
> Reply-To: MailScanner mailing list <MAILSCANNER at JISCMAIL.AC.UK>
> To: MAILSCANNER at JISCMAIL.AC.UK
> Subject: Re: Enterprise scalability
>
> It's per blade, they do about a 1GB a day it appears, average message
> size is 6.2k according to my stats. As I said I'm using a NFS share over
> 10/100, each blade has 1GB of mem. Your sytem is significantly more
> powerful then any of my blades, SCSI local disk and a P4 Xeon.
>
> Regards
> MIKE
>
>
> On Thu, 2004-02-26 at 05:56, Randal, Phil wrote:
> > Is that 200,000 per blade?  What sort of volume in GB are we talking about?
> >
> > Just curious.
> >
> > We're handling about 8600 messages a day, 560MB on a single P4 Xeon 2.4GHz,
> > 1GB RAM, tmpfs and hardware mirrored SCSI disks with a load average around
> > 0.5 (box is also running squid for around 800 users).
> >
> > Cheers,
> >
> > Phil
> >
> > ---------------------------------------------
> > Phil Randal
> > Network Engineer
> > Herefordshire Council
> > Hereford, UK
> >
> > > -----Original Message-----
> > > From: MailScanner mailing list [mailto:MAILSCANNER at JISCMAIL.AC.UK]On
> > > Behalf Of Michael Baird
> > > Sent: 25 February 2004 21:41
> > > To: MAILSCANNER at JISCMAIL.AC.UK
> > > Subject: Re: Enterprise scalability
> > >
> > >
> > > You would need a lot of mighty boxes to handle that kind of volume. I
> > > use blades, so when my volume gets to a certain level, I just image in
> > > another one, and mx to it as well (to a centralized NFS spool). My
> > > blades are PIII-1200, I can handle without delay running
> > > mailscanner/spamassassin, and using tmpfs for the queue.in 200,000 per
> > > day, I'm using McAfee to do virus scanning as well, the machines only
> > > handle inbound mail, no outbound relay is allowed.
> > >
> > > Regards
> > > MIKE
> > >
> > > > If I were implementing in this type of environment, I would break it
> > > > up into more manageable chunks.  First, figure out roughly how many
> > > > messages are processed each day.  If you are expecting 500,000 users
> > > > who will receive on average 75 messages per day you are looking at
> > > > about 37,500,000 messages per day (that's a lot of mail).  You can
> > > > build boxes fairly cheaply for handling a fraction of that mail, say
> > > > 1,000,000 messages per day.  Get yourself 40 boxes, some load
> > > > balancing tools, a way to manage the configuration files easily and
> > > > you are in business.  There were some threads within the
> > > past 3 months
> > > > about average load with hardware descriptions that you will find
> > > > somewhat helpful.
> > > >
> > > > > -----Original Message-----
> > > > > From: Forrest Aldrich [mailto:forrie at FORRIE.COM]
> > > > > Sent: Wednesday, February 25, 2004 1:01 PM
> > > > > To: MAILSCANNER at JISCMAIL.AC.UK
> > > > > Subject: Enterprise scalability
> > > > >
> > > > >
> > > > > I'm looking to evaluate a scalable scanning solution - the
> > > > > tune of 100's of thousands of users - and I wonder if anyone
> > > > > here can share their successes (and nightmares) with regard
> > > > > to MailScanner and its auxiliary
> > > > > tools (SA is another worry).   I'm looking into Qmail at
> > > > > first, as we've
> > > > > a need for virtual mailboxes (5 per user), etc.
> > > > >
> > > > > I'm concerned about how perl might behave in this type of
> > > > environment.
> > > > >
> > > > >
> > > > > Thanks.
> > > > >
> > > >
> > >
> >
>