Tuning MailScanner

[Randy Herban]

> -----Original Message-----
> From: Tom Sevy [mailto:tsevy at EPX.COM]
> Sent: Thursday, May 22, 2003 4:25 PM
> To: MAILSCANNER at JISCMAIL.AC.UK
> Subject: Re: Tuning MailScanner
>
>
> Has anyone attempted or succeeded in implementing a load
> balanced or HA
>
> cluster of SA servers?
>

I kind of running a cluster of SA servers, but not necessarily in a load
balanced or HA way.
The REAL nice thing about handling incoming mail (which is all that mine are
doing, outbound mail is done seperately) if one mx host is down, the next mx
host will be attempted.  In a sense, the "clustering" is already done for
you.

I have done an HA cluster before, for different purposes.  I was able to get
it setup and all tests were successful, but never actually had a chance to
test in production (management is paranoid).



> -----Original Message-----
>
> From: mikea [mailto:mikea at MIKEA.ATH.CX]
>
> Sent: Thursday, May 22, 2003 5:10 PM
>
> To: MAILSCANNER at JISCMAIL.AC.UK
>
> Subject: Re: Tuning MailScanner
>
>
>
> On Thu, May 22, 2003 at 04:40:59PM -0400, Gerry Doris wrote:
>
> > I was reading the thread on running MailScanner with lower
> numbers of
>
> > children and/or reducing the number of messages/batch to be scanned.
>
> > However, I'm not sure I ever saw a concensus on what is best.
>
> >
>
> > It sounded like it was dependend on the number of messages being
> > received.
>
> > If there weren't enough messages coming in to load all the
> instances
> > of
>
> > MailScanner and only the first couple grabbed all the
> messages then it
> > was
>
> > a waste running the other instances.
>
> >
>
> > On the other hand, reducing the number of messages/batch would
> > distribute
>
> > the mail load across all instances of MailScanner and would be more
>
> > efficient.
>
> >
>
> > If this is correct then the trick is to spread the mail
> load across as
>
> > many instances of MailScanner as possible as long as you
> don't run out
> > of
>
> > server resources ie cpu load and memory.
>
> >
>
> > Sendmail is very efficient. The real problems are the virus/spam
> > scanning
>
> > especially the RBL checking.
>
> >
>
> > Is the above correct?
>
> That's pretty much my take on it.
>
> Of course, tuning, even within those very broad guidelines, will be
>
> very much a matter of tweaking one thing at a time, hoping that it
>
> was the right thing, and wondering how much of the apparent change
>
> comes from the tweaked parameter and how much from changes in the
>
> characteristics of the processed mail.
>
> I've been doing much the same thing with mainframe operating systems
>
> since 1977, and it's still a black art. I used to have a black robe
>
> with start, moons, etc., on it, which I reserved for working on really
>
> bad problems, but my management disapproved. By the time I became
>
> management, the robe was gone. *sigh*
>
> --
>
> Mike Andrews
>
> mikea at mikea.ath.cx
>
> Tired old sysadmin since 1964
>