Hacker News Reader

Comment by hackuser | Hacker News Reader

> This is not a good suggestion on their part, and has long been proven ineffective:

Agreed; that is bad advice.

I tell people: If you think of a trick then the attackers, who have expertise and think about these issues all the time, have thought of it long ago and have written it into their password-cracking software. That applies to visual substitutions (such as GoToMyPC recommended), phonetic substitutions (e.g., AmeriKa), patterns on your keyboard, etc.

In a more technical sense, that applies to tricks that can be defeated with an algorithm and affordable computing resources.

View on HN

But if the trick you are using results in sufficient entropy, it shouldn't matter that the enemy has thought of it too. I'm still a fan of linking common words together as this results in easily-memorized passwords with very high entropy. "Catrunningfishhostagelaptop" is a good password these days. it is easily remember but also difficult to brute force even if you know the trick by which it was constructed.

(And yes, by putting it out there I realize it is now a bad password.)

Couldn't I write a bruteforcer that instead of 5 * letters, tried 5 * dictionary words and get your password easily?

There are a lot more words in the dictionary than letters.

The question is whether people are all going to pull words from a very small pool or form sentences to cut the search space (adjective noun adverb verb noun). A search space of 2000^5 (2000 most common words without punctuation) is still about 20 times larger than 80^8 (upper/lower/number/symbols)

Five or six words is a huge amount of entropy. In reality there are other rules in my example than just using words. It starts with a capital letter and includes a conjugated verb (running) both of which increase complexity considerably. And it of a random length, having taken its length from my selection of words rather than from any fixed policy. Few services accept random lengths but imho all should.

There is a great scene in DoctorWho where a door password is a series of concepts (spoken mentally) that one thinks about in a particular order. Fiction, but the person writing that scene knew a thing or two about password complexity.

> It starts with a capital letter and includes a conjugated verb (running) both of which increase complexity considerably.

They reduce complexity: It reduces the number of possibilities on the first character, and the attacker knows that one word must be a (conjugated? present participle?) verb yet, when otherwise it could be anything.

Those are rules in the user's head. We assume that the enemy only knows that the user is chaining words.

That might not be true for an enemy who has access to some of your other passwords already, but assuming it is I still don't see how the strategy above increases complexity. At best, the complexity remains the same.

> strategy above increases complexity

It doesn't, but it does make it easier to remember :)

The number of possible letters is much, much smaller than the number of possible dictionary words. The latter would take many orders of magnitude longer to brute force, which would make it unfeasible.

The way to compute the amount of possible brute-force combinations:

For a normal 5 letter (alpha-numeric only) password:

36 factorial, which is (36 * 36 * 36 * 36 * 36) = 60 million possible combinations.

However, for a 5-word password, the calculation is as follows:

Assuming the number of possible words in the English language is 1.025109 million, then:

1,025,109 factorial, which is (1,025,109 * 1,025,109 * 1,025,109 * 1,025,109 * 1,025,109) = A really big number.

So, granted, the amount of possible words is much smaller than that. But the attacker can't know that for sure. To put the number above in scale, just 69 factorial is equivalent to 1.711224524 * 10^98.

I guess you meant 36 to the power of 5, instead of factorial. 36! is a much bigger number than 60 million; same goes for 1,025,109 a few lines below

I agree with your general point, but I have just a couple of mathematical points:

> For a normal 5 letter (alpha-numeric only) password: > 36

With the "normal" recommendation of numbers, uppercase and lowercase letters, and punctuation, it's up around 100.

> factorial

The calculation is X^Y, where X is the number of possibilities per element, and Y is the number of elements (letters or words in these examples). So using only only lowercase letters (26) in an 8 character password would be 26^8. It's not factorial because you can reuse characters (e.g., your password could be X92m-sXp/)

> the amount of possible words is much smaller than that. But the attacker can't know that for sure

They will search the more likely words first. I read someplace that 10,000 words covers most people's vocabulary. That's still 100 times better than numbers + uppercase + lowercase + punctuation.

>"The calculation is X^Y, where X is the number of possibilities per element, and Y is the number of elements (letters or words in these examples). So using only only lowercase letters (26) in an 8 character password would be 26^8. It's not factorial because you can reuse characters (e.g., your password could be X92m-sXp/)"

You're absolutely right, not sure why I got ahead of myself with using factorial so incorrectly.

You were probably thinking about the fact that picking, in order, 5 elements taken from a set of 1000 elements yields (1000 factorial) / (995 factorial) possible combinations.

Too late to edit my post. I made a bit of a mistake referring to it as factorial.

Technically true perhaps, but from my experience the word selection is always way narrower than that, like usually from maybe the 2000-5000 most common words.

Still a big number, but not astronomically big.

It's big enough to matter. 2000^5 ~= 60^9 so it's as strong as a 9 random character password which is usually regarded as strong enough, and far better than a typical 1 word with 3 or 4 special characters mixed in.

You're making the incorrect assumption that all letters are equally likely, and all English words are equally likely.

I'd guess the words used in that password fall into the top 10k words, easily. No need to check words like pediatrician, when the password contains the word cat.

^This shouldn't be downvoted. It is a perfectly valid question.

Are there any good sources on what sort of generative trickery a good cracking tool will use?

I've noticed that the OSX Keychain 'memorable' password suggestions are almost always /(\w+\d+[[::punc::]]+\w+)/, which is enough of a pattern to probably make an explicit generator for which will match much faster than a naive brute-force.

> Are there any good sources on what sort of generative trickery a good cracking tool will use?

Two answers:

1) It's not worth trying to keep up with, unless that's your specialty. There are people who spend days thinking up attacks; how much time will you need to spend to keep up with them?

Far more time-effective than trying to anticipate all the specific attacks is to use provably secure defenses: Passwords that are mathematically too expensive to brute force, and which cannot be guessed by other means (e.g., by knowing personal information about you). That shifts the weak link from your password to the security of your OS, network, and applications, which probably have a couple of holes in them.

2) This will give you an idea of the scope of possible attacks. Note that it's 9 years old: https://www.schneier.com/blog/archives/2007/01/choosing_secu...

Note that what you're really looking for is the amount of entropy in the encoding. So for many 'memorable' encodings, the password does need to be longer vs. an alphanumeric+symbols password, but still encodes the same number of bits of entropy.

Diceware[1] is a nice example of how such a system works, and how you can generate a strong 'memorable' password by hand (and dice ;-). In short, there's a dictionary of short words, letter combinations, and numbers that form the "symbols" of the generated password. A random input (here, rolls of dice) are used to select a series of symbols. You could just as well create your own table and generate the password from that, but if you've got enough randomness as input then it doesn't so much matter what encoding you use. In fact, the general assumption is that you should assume your attacker knows your encoding.

[1] http://world.std.com/~reinhold/diceware.html

This is what I tell people about the + notation in their email addresses. Spammers know what it's for, and can easily strip it out if they wish to conceal their address source.

Reply on news.ycombinator.com