SpamAssassin: A Practical Guide to Integration and Configuration

Rating:
Summary: Excellent guide
Review: I have got O'Reilly's book also, but this book is much better. It covers everything from installing spamassassin to using it as a mail gateway, which means you can do your spam filtering on a separate machine and do not have to tamper with you mail server (this allows you to use spamassassin even if Exchange is your mail server).

There is extensive coverage of various techniques to improve filtering and accuracy, making spamassassin effective for your own specific needs. The chapters on rules and using online antispam services and databases are great.

I particularly liked the information on best practices, which reflects the author's experience.

This book seems to cover every aspect of spamassassin, and it does that very well.

Rating:
Summary: SURBL is the key idea
Review: SpamAssassin is a common and free antispam method that is comprehensively described by this book. McDonald is writing for a sysadmin who wants an effective method to halt spam. To be sure, he cautions that SA is not 100% effective. He gives concise explanations of how to install and run it.

The most recent version surveyed here is 3.0. And there is one key innovation in it, and also in the earlier 2.63 version. Namely what it calls SURBL = Spam URI Real time Blacklist. A powerful idea. You block a message as spam, if its body has URIs in an SURBL. It should be said that it took the SA coders a long time to recognise and implement this idea.

Another issue is how to generate an SURBL. Here, the author just says that you can get these from external sources, like Spamhaus, just like you would for an RBL. More discussion here might have been helpful.

There is little recognition in the book that the SURBL method has significant advantages over a Bayesian, the latter of which gets extensive coverage. The SUBRL method is faster in extracting URIs from a message body and then comparing against the SURBL, than in comparing the message's words to the corpus of the Bayesian. All the more so if the latter word comparision is restricted to words that are not in HTML tags and are visible to the reader.

Also, if the SUBRL has domains that are definitely considered to be spammer domains, then using it is deterministic. Unlike the stochastic general nature of SA's approach, which totes up a spam likelihood score. The deterministic aspect is far stronger.

Plus, there is no need for regular manual training, as with the Bayesian - either at the user or sysadmin level. But a casual reader will miss all this. The SURBL is given minimal description.

Rating:
Summary: Well written and up to date
Review: This excellent book is the only title I've found that covers the latest version of the SpamAssassin. It is an easy introduction
to installing, configuring, and fine tuning SpassAssassin, but it thoroughly covers all the topics I'll ever need to know about. I'm a beginner to SpamAssassin, and now feel like an expert! Even friends who already knew SpamAssassin well find themselves looking at this when they get stuck.

I found the book had plenty of coverage of SURBL. There's a whole chapter dedicated to using network tests to check URL blacklists.

Highly recommended.