AOL Email Issues

It has been noticed by many of our Gatorlink users that from time to time AOL refuses to accept mail from the ufl.edu domain. This document is a discussion of the problem that describes how the impass is reached and what OSG has done to alleviate the problem.

AOL uses two sets of heuristics to decide when to refuse email from a system.

The first is the rate of messages -- that is the number of attempted message transactions per unit time. They're less than forthcoming about the threshold at which they block, but experimentation shows it to be around 200 messages/hour. They apply this rate limit by IP address. So it is possible for only one or two of the three computers that talk SMTP to the outside world with the name smtp.ufl.edu to be blocked. Our central servers are currently white-listed with AOL, so these rate limits do not apply to them.

AOL sometimes applies the rate limit to the "second Received: line" (their words). In practice they apply the rate limit to almost all of the non-white-listed Received: headers. This means that messages going through our servers may be blocked because of the machine from which they were sent, or any machine they traversed. It is suspected that some of the problems we've had point to departmental mail servers being blocked, even though they use the central relay.

There was an attempt to develop and maintain a white-list (with AOL's help) for our campus departmental mail servers, but that effort is pointless. As you can imagine, the return on time investment there is essentially zero.

The second, completely independent, metric AOL uses to blocks messages is customer complaints. When an AOL customer clicks the "This is Junk" button for a message it counts against our mail servers. A substantial portion of the traffic we send to AOL is messages from non-University sources forwarded through users' Gatorlink accounts. When one of the forwarded Gatorlink addresses gets actual spam and the AOL customer clicks it so, we are penalized. There are no white-lists for this method.

Something like 15%-20% of the complaints we get back are legitimate University business, for which a complaint should not have been generated. The students deciding to complain about Wednesday Update is a good example. AOL, of course, has no way to determine this.

We've set up a feedback-loop with AOL, and we get a copy of every message about which there is a complaint. When we started rejecting messages with a spam score of 10 or higher, we reduced the number of complaints by more than 50%. When we started using the Spamhaus Real-time black-list service, we reduced the complaints by an additional 20%.

Using the RBL we've been blocked much less of the time, while reducing the amount of spam we accept ourselves significantly.

The newest development: according to the Higher Education Email Administrators mailing list -- of which the AOL Spam Czar is a member. The denizens of the list are trying to convince AOL that they should create a list of schools whose spam scores they trust. The idea is that AOL will use our spam score (and maybe their customer's preference) to pre-file most spam message into their Junk folder. We'd then get a free pass on the complaint volume. We have high hopes for this deal, but it's too early to tell much.

So that's "The AOL Problem" in a vary large nut-shell. OSG will be happy to answer any questions you might pose about the specifics of the situation or our methods.