Mike Jackson wrote: > Not that I want to threadjack, but this prompts the questions: Is there > an advantage to running Razor over Pyzor? And, do you gain anything by > running both simultaneously? (Other than increased system load, that is.)
Well, Razor's implemented in Perl, so SA can just call the code via the Perl module. It uses several different engines to generate hashes, so it's better at catching polymorphic spam, including one that focuses on embedded URLs (which makes it somewhat SURBL-like). It's more robust than Pyzor, and currently maintained. The client uses the Artistic License, but there are some limitations on using the service (though most people probably won't run into them), and the server design is closed. (I believe there was some talk about developing a caching server, which would have been a commercial product, but I don't remember whether it went anywhere.) Pyzor's implemented in Python, so SA has to actually call out to the pyzor script. So there's the two levels of overhead (calling an executable and firing up Python). The Pyzor client hasn't been updated in several years, and includes some long-standing known bugs with message handling such that it can crash on certain types of malformed input. (A bad transfer encoding header is all it takes.) It only has one engine, and I believe the user base is smaller. (Of course, that cuts both ways -- less spam gets submitted, but fewer false positives get submitted too.) However, both the client and server are 100% free, and you can run your own Pyzor server if you want to. I haven't run the numbers in at least a year, but the last time I compared results there was a significant difference in which spam was caught by each service. There was a large overlap, of course, but there were enough messages that were only caught by Pyzor or only caught by Razor that it was worth calling both if you could afford the increased load. -- Kelson Vibber SpeedGate Communications <www.speed.net>
