lør, 14 05 2011 kl. 12:11 -0300, skrev Henrique de Moraes Holschuh: > On Sat, 14 May 2011, Camaleón wrote: > > > I am running a mailserver with exim, courier-pop, courier-imap and > > > squirrelmail. I has been running "fine" for about 18 months now. But how > > > can I assure my self and my customers that I deliver a good quality > > > mail-server? > > > > You can endorse a SLA (Service Level Agreement) that ensures and metrics > > some basic aspects of your service. > > Yes. Assuming you're going to do it professionaly (and not > semi-professionaly, or whatever): > > 1. Measure the *SERVICE*, which OUGHT to be different from measuring a > server (because really, you need at least two in active/active or > active/passive servers in an high-availability cluster to deliver something > worth paying for). > > 2. Measure the standard metrics: > a) service uptime (pingdom.com, custom scripts) > - this is not just SERVER uptime. > - service availability > b) service performance metrics > - delivery latency (incoming and outgoing. Outliers are not > a problem. Average delay to deliver 90% of the email, is). > - rejects (incoming and outgoing) > - spam/virii blocking > > 3. Measure per-user metrics > a) incoming/outgoing bandwitdh usage > b) storage quota > > 4. Have a data recovery and backup strategy, provide an "undelete" service > (your users will demand it), and test them often (or you're toast at the > first failure). > > 5. Use the high-availability cluster capabilities to do rolling upgrades, an > outdated server is a compromised server... and this also makes sure the > HA is working well (which also means you NEED to do the fail over in a > monitored way to make sure it won't go up in flames if it fails. E.g. > whatever you do, don't use a filesystem that will not tolerate > active/active usage even if normally you prefer to operate in > active/passive mode. > > You will need scripts. munim, nagios, cacti, net-snmp can help you a lot to > monitor the servers. queuegraph and mailgraph can help you monitor the MTA > performance. You will have to write scripts to monitor service latency > (send email, measure time until it is delivered to a system pop account, > measure time until it is delivered at a remote account in gmail, hotmail, > yahoo, other providers). > > You will need to enroll on yahoo/hotmail/etc spam feedback loops, and keep > strict monitoring of all RBLs, and take a lot of care with the reputation of > your outgoing servers and email domain. You will need a way to temporarily > switch the outgoing server to a different IP if you get blacklisted (but do > so only after you *HAVE* fixed the issue that resulted in the blacklisting). > > The list goes on and on. SPAM, and the "email attestation" industry > fomented by the big email services like hotmail and yahoo, are causing a big > problem. > > -- > "One disk to rule them all, One disk to find them. One disk to bring > them all and in the darkness grind them. In the Land of Redmond > where the shadows lie." -- The Silicon Valley Tarot > Henrique Holschuh > >
Hi all, Thank you for you fine answers. I got a lot of inspiration. I know now that some kind of HA cluster is my target, but the first steps will be more monitoring with nagios, munin and pingdom. Also I will write some kind of SLA that my users can use to track my level of service. Thanks again, and more suggestions are always welcome! :-) Regards Lars Nielsen www.lfweb.dk -- To UNSUBSCRIBE, email to debian-user-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org Archive: http://lists.debian.org/1305408242.2439.12.ca...@mp.fullrate.dk
