What you should be doing is using bcrypt, tuned to take several hundred milliseconds per crypt. /usr/share/dict/words --- a pitiful dictionary --- with even the most basic set of permutations will take weeks to test a single password under this regime. Bcrypt is free and available for every major web platform. I didn't write it; the guys who did, David Mazieres and Niels Provos, are much smarter than me.

Computer science has solved the brute force problem on passwords, and we should stop dithering and adopt the solution.

Seen from a security perspective this adds to the overall security of the site, however most users simply find an easy password that fits the bill. blink182, 123456qwerty, qwerty! - which is almost as easy to bruteforce or keep in a rainbow table as normal dictionaries.

You should be aware, however, that password requirements detract from the user experience. When users pick a password for a new account they are at the most crucial part of a website: The sign-up. This is where you can lose people that were otherwise ready to sign up.

So it's a trade-off. If your site holds credit card numbers, or other vital information a certain minimum requirement for passwords is certainly in place, but if your site is something like HN, where no valuable information is held, you should probably just let users pick whatever password they like. Worst case they will lose a few comments and some karma.

I think the worse case is actually substantially worse -- call it an identity escalation attack. I think a huge number of Internet users are vulnerable to it:

a) Use any means you please to compromise the password on a low-security site, such as HN.

b) Use the same login credentials at other low-security sites, such as Facebook and Twitter. These compromises are valuable in their own right, but what you are really looking for is the email address.

c) After you have their email address, compromise it using variations on the same password you already know.

d) You can now read their email. Search for "statement", and identify which bank they bank with. Go to bank and try logging in with their email and password. That probably just worked. If not, click "lost password", supply email address and their security question (you have their Facebook already or can in seconds, how hard will it be to find their mother's maiden name or middle school?), and enjoy.

e) For added fun, look for other high-value accounts whose security is premised on the email user being authorized, such as brokerages (a tough target due to strong institutional controls, like banks) and GoDaddy (a successful compromise would cost me FAR more than what I have in my bank account).

The worst part? You can automate large portions of this attack (given a list of usernames and password, writing a script to try them at a dozen or a hundred popular services is child's play) and outsource the rest right up until the "compromise bank account" portion.

"The first is to create 'canary' accounts. Create accounts that have e-mail addresses, like "something-really-long-xyz-123@gmail.com". This account is not going to get any spam e-mail. When it does get its first spam, you'll know that it came from your database."

In the event that your canary account receives a spam message, you will have gained one piece of data: that at some point in the months since you installed the account, it is likely that someone exploited a problem that exposed a table column in your database.

Now what?

You don't know what the bug was. You don't know how far-reaching the damage is. You don't know when the event occurred. There is, in fact, almost nothing actionable about this information at all.

You might be telling yourself two things:

* That, upon learning of the possible breach, you will lock down the security of your code and your site. Of course, think for two more seconds and realize this is ass-backwards; if you can reasonably lock down your site, you might as well do it before you burn your whole database.

* That, upon learning of the breach, you can get all your users to change passwords. Which, of course, you will never do, because you will be taking an enormous reputational hit over circumstantial evidence, and for the purposes of seeding your database with a couple thousand new passwords for the same attacker to steal.

People like this "canary" and "honey token" stuff because it's fun. But it's not valuable.

I'm guessing that after that call, you'd go through your logs and attempt to find out how the site was cracked. Then, if you find some anomalous behavior, you might be able to isolate the component that was being exploited and fix it. Maybe you wouldn't find it, but maybe you would.

Here, let me help all of you with web apps out right now: you have probably been compromised.

Go do something about it. Then come back and tell me what you did.

Not sure how much that would help either though.

One thing I've also debated about before is whether to use a per-site or a per-user salt for passwords. From this article, it sounds like a per-site hash is enough to defeat the rainbow table services.

None of this matters if someone is trying to crack your site in particular, but sometimes you just have to be faster than the slowest gazelle in the herd.