From mboxrd@z Thu Jan  1 00:00:00 1970
From: corey@lod.com (Corey Lindsly)
Date: Wed, 20 Sep 2017 11:51:22 -0700 (PDT)
Subject: [TUHS] Who is running their own mail server and what do you run?
In-Reply-To: <20170920162502.GN25650@mcvoy.com>
Message-ID: <20170920185122.EB54D410F@lod.com>


> 
> I tried running my own server on mcvoy.com but eventually gave up, the
> spam filtering was a non-ending task.
> 
> If someone has a plug and chug setup for MX I'd love to try it.
> 
> Thanks,
> 
> --lm
> 

We provide email service for a few hundred (mostly legacy) customers. The 
essential elements of the system are:

Linux (HA pair)
Postfix
SpamAssassin
ClamAV
RBLs (Spamhaus)
MySQL database
Maia Mailguard

I spend maybe five minutes a week maintaining the system. The most 
important part is Maia Mailguard. Incoming email is scored by SpamAssassin 
(higher = more likely to be spam). Above some threshold, which can be 
tuned up or down by the user, the mail is tossed into their spam 
quarantine. The user is able to browse this quarantine. False positives 
can be freed-up and delivered as usual and the sender is whitelisted. 
True positives are confirmed and contribute to training the filter. After 
a little while the system gets pretty good at automatically 
distinguishing between spam and ham.

The important point of this system is that it puts the user in control. 
Too much spam? Turn the spam threshold setting down a little lower. Too 
many false positives? Turn that knob a bit higher. Expecting an important 
email? Whitelist the sender or check your spam quarantine. As a mail 
admin, my goal is to give the users the tools to manage their own email 
effectively according to their requirements. I don't want my phone 
ringing every time some user wants to complain that he gets too much spam.

The system has been a success. Complaints from the users have dropped to 
nearly zero. I spend very little time thinking about or dealing with the 
mail system. Stats follow.


As of: 2017-09-20 11:38:53 PDT 

Efficiency 99.35% False Positive 0.50% False Negative 0.15%
Sensitivity 99.84% PPV 99.45% Specificity 94.00% NPV 98.16%


--corey