Take the 2-minute tour ×
Server Fault is a question and answer site for professional system and network administrators. It's 100% free, no registration required.

I'm re-implementing an overloaded Postfix cluster, which beyond its many problems, has major efficiency concerns. The cluster is a dedicated MX Gateway, so mailboxes are not a consideration. Efficiency really is a significant concern, as we handle a substantial amount of mail, and the cluster is on its knees. I/O isn't our bottleneck, and statistics have confirmed this conclusion.

Currently I'm considering how attachments are processed, and weighing the methods. I assume once I know the proper method, I could locate the feature specific projects.

I've still got a few questions, even after reading the Postfix docs.

  1. Postfix Milters: Are they only available before-queue? The docs mention some limitations, are Postfix Milters able to do things like strip certain attachments from an email? What about scanning for viruses, and removing them as well?

  2. Before-queue Content Filters, and After-Queue Content Filters: The docs also reference using an SMTP proxy, to an internal address, is that just a specific method to utilize Content Filters? Would an alternate method use something like an external program/script? But both would be different ways to setup Content Filers?

  3. Which is more appropriate for virus scanning, and stripping infected files? Same question for stripping banned attachment types? What about spam detection methods, such as DCC? With performance/efficiency being a significant concern, which method is most ideal? The docs describe all the methods horrifically!

I really appreciate any expert suggestions!

share|improve this question
How many users are you supporting? –  ewwhite Feb 17 '13 at 23:14
@ewwhite: It's actually just used as the MX gateway, so assuming a mail is accepted, it's often just relayed to an MS Exchange cluster. The users access all their mailboxes from MS Exchange servers. The actual number of possible address is in the thousands, but the real damage is from the volume of incoming mail. –  TechZilla Feb 18 '13 at 19:33
@ewwite: Additionally this explains why I/O has not become a bottleneck, but we're taking a pounding on mail processing. With my proposed stack implementation, I expect a significant reduction of post-queue processing, as most spam should have theoretically been rejected. Although it's still definitely a concern, with some basic efficient design, this should be mitigated. –  TechZilla Feb 18 '13 at 19:40
I'll be honest... I offload this type of processing to purpose-built appliances or virtual machines on hardware that I specify... Barracuda is the solution I normally go with. Are you scanning attachments with something like ClamAV? –  ewwhite Feb 18 '13 at 19:45
@ewwhite: Actually the horrific implementation, which I'm replacing, was previously using ClamAV. Although when I had to do emergency efficiency improvements, to buy time for my proposal, I replaced ClamAV with F-Prot. ClamAV was the worst bottleneck, but after being replaced, many other bottlenecks became apparent. It's actually shocking to compare ClamAV's engine, to engines known for being efficient, it's extremely significant. I did successfully buy myself some time, load reduced to about half of its worst average. –  TechZilla Feb 18 '13 at 19:54
show 1 more comment

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Browse other questions tagged or ask your own question.