In the past few months, we’ve made a couple of behind-the-scenes improvements to our mail systems that have improved reliability for our customers.
- Redundant outgoing mail delivery
- Automatic accepting of replies
Redundant outgoing mail delivery
One of the hardest things about running an outgoing mail server is dealing with obscure anti-spam “blocklists”. Our strong anti-spam policy, outbound virus scanning, and rate-limiting systems do a good job of preventing outgoing spam and keeping our servers off major lists — but occasionally we hear about more obscure blocklist errors, too.
For example, one of our customers recently sent mail that was returned because one of our mail servers was on a list we’d never heard of. We found that the list tries to include every server that sends any kind of marketing mail, even when the messages have been explicitly requested and aren’t spam. As you’d expect, the list includes servers run by large companies you deal with every day.
Fortunately, few people would use such a list to block mail. In fact, the list Web site explicitly tells people not to do so (you’re supposed to just use it to sort mail, marking messages from large companies as lower priority than personal mail). Unfortunately, some people think more mail-blocking lists are always better (which isn’t true, as Al Iverson’s excellent reviews explain).
To solve this rare-but-frustrating problem for our customers, we’ve added a feature that other companies don’t have. If an outgoing e-mail message from our servers is rejected, we try sending the message again from our secondary data center before “bouncing” it.
So every message you send is tried twice, if necessary, from two completely independent servers / IP address ranges / data centers / AS numbers. Since most lists block the IP address the mail is coming from, the chance that both attempts will be blocked by an obscure list is low.
This should virtually eliminate the chance that your messages will be returned because of an obscure list entry.
Automatic whitelisting of replies
We try to block as much incoming spam as possible — in fact, we’re blocking a higher percentage of it all the time. (Unfortunately, spammers just keep sending more and more, so the number of spam messages our customers receive remains about the same, unfortunately. Keeping it roughly the same is an accomplishment in itself.)
As we block more and more spam, we need to be careful that we don’t accidentally block a real message. One of the ways we do this is by “whitelisting” — allowing certain messages past our filters even if they look like spam.
We’ve always done some whitelisting, but for the last few months, our mail system has automatically remembered who you send mail to, then whitelisted the replies. Any reply someone sends to you won’t be blocked, no matter how much it looks like spam. (The only exception is certain kinds of forgeries flagged by SPF — that makes sure that if you send a message to PayPal, for example, we won’t incorrectly whitelist forged PayPal messages.)
This has advantages over the “add to your Webmail address book” whitelisting offered by some companies, because it works even if you use a mail program like Outlook on your own computer.