Slashdot Mirror
ISPs Starting To Charge for 'Guaranteed' Email Delivery
Presto Vivace writes "Under the guise of fighting spam, five of the largest Internet service providers in the U.S. plan to start charging businesses for guaranteed delivery of their e-mails. In other words, with regular service we may or may not deliver your email. If you want it delivered, you will have to pay deluxe. 'According to Goodmail, seven U.S. ISPs now use CertifedEmail, accounting for 60 percent of the U.S. population. Goodmail--which takes up to 50 percent of the revenue generated by the plan--will for now approve only mail sent by companies and organizations that have been operational for a year or more. Ordinary users can still apply to be white-listed by individual ISPs, which effectively provides the same trusted status.'"
How does it fight spam if the spammer can ask to be whitelisted, or if the spammer can pose as or actually be a business operating for more than a year? Lame.
"Well, assuming an user pays for the e-mail account, isn't this a breach of contract and false advertising? By "providing an e-mail account", it can be assumed no real mail is ever meant to be knowingly dropped.
Declaring those who haven't paid the protection racket as not "real mail" is not really something that I would envision as something which would pass a non-bribed judge."
Guess what, this is exactly how the USPS works. They are not responsible for making sure the mail is delivered unless you pay more for it, like certified mail, etc.
How do I know? I was told this in so many words when I had mail lost and complained to the Post Office about it.
It was basically, "if you want to make sure it gets there, have it insured, otherwise..."
Yeah, nice little racket the USPS has too!
I think part of the problem is that spam filters are generally broken and don't work that well. Part of the problem is that no one has seriously thought about how crappy the approach is. The other part of the problem is that their is little or no personal ownership of the filtering of spam.
When the ISP/customer have no relationship on identification of what is spam the ISP has to aim really high and take the approach that anything that is obviously spam is not delivered and everything else is. The net effect is the ISP might not deliver porn spam, but they'll deliver many other things with impunity. If there was a more aggressive involvement of the customer/consumer of the email then you could better tune the filters to match each user better.
SpamAssassin is the worse offender. It's origination was to do static regex checks and add points for each hit. And when you were done, the points put you either IN or OUT. But in order for SA to work you have to tune the number of points added for each regex test. And this is constantly changing. But for it to work, you have to be constantly monitoring the results. No one does this on a consistent basis.
A critical drawback with their approach is the constant game of catch-up they have to play in order to get the filtering to work correctly and then someone has to run some update script to hopefully get everything working correctly. Again, this has to be done continually like the tuning or it will start to fail.
Bayesian filters offered a great alternative but they quickly turned into their own problems. SA uses Bayes, but it's not effective because of the lack of feedback from the consumer (in most cases). It's also prone to over-rides by their own auto-whitelisting. Convenient, but deadly. Where Bayes lacks goes back to the original problems of non-customized feedback and involvement. It's very inconvenient to try and set up something like bogofilter to run for every individual in a group of 1000's so the mail admin makes one file for everyone thereby generalizing the statistics and making them less effective because they have to be good enough for everyone but not so good they remove any of the really serious spam.
And yes, SA does user specific Bayes filtering. I used it for three months and it sucked. It was not a very effective spam filtering system even with user specific bayesian filtering included. It's also getting pretty darn slow. Slow enough to become a consideration.
DSpam is effective, customized, and slower than molasses in january. It will also lose email. But YMMV and I don't really care to hear about how great it is. I lost a lot of email and a lot of money as the result of it. Perhaps some day they can get their act together, but there will always be a severe performance penalty for CRM114. But Bayesian filtering can still compete with CRM statistical success with 100X performance increase.
So what do you do about spam filtering?
The technology exists to effectively and efficiently filter spam. But that's not the problem. The technology that is used today is relatively lame because there are shortcomings abound that prevent a good solution for someone really large (like an ISP).
The problem is to redefine how the consumer is going to own their own spam filtering effectiveness. No more auto-whitelist. No more auto-blacklist, No more auto-update of Bayesian tokens. All of these can be carefully manipulated to taint the statistics and allow delivery in droves. The consumer must take ownership of their mailbox in the same manner that they are expected to take ownership of their credit card information on the internet.