|
|
Log in / Subscribe / Register

10,000 Linux servers hit by malware (ars technica)

10,000 Linux servers hit by malware (ars technica)

Posted Mar 20, 2014 15:08 UTC (Thu) by Wol (subscriber, #4433)
In reply to: 10,000 Linux servers hit by malware (ars technica) by creemj
Parent article: 10,000 Linux servers hit by malware (ars technica)

Eggsackerly.

The word "average" is one of the most abused words in the heavily abused realm of statistics. When someone quotes "the average" at me, my immediate reaction is "which average? What misdirection are you trying to get past me now?".

My instant reaction on seeing the post pointing this out was to smile at the report's author being outed as being clueless to what he was writing ...

Cheers,
Wol


to post comments

10,000 Linux servers hit by malware (ars technica)

Posted Mar 21, 2014 3:37 UTC (Fri) by plaxx (guest, #53703) [Link] (5 responses)

I'm afraid that we were not being clueless but instead we had a language barrier issue. The ones who wrote the report, myself included, are Canadian French with French being our tongue language. In our language the mean is 'moyenne' but 'moyenne' when writing or speaking in English is much more commonly translated into average than into mean (probably because mean is precise whereas average is not). We were never really exposed to the difference I guess..

Anyway, we simply have no way to differentiate average and mean in French and we've made a mistake. We are sorry. We'll remember this one.

Oh and by the way, we really did mean the mean (no pun intended).

10,000 Linux servers hit by malware (ars technica)

Posted Mar 21, 2014 18:39 UTC (Fri) by Wol (subscriber, #4433) [Link]

Sorry to impugn you. I have a low tolerance for native speakers who abuse the language, but a high tolerance for foreigners (who almost always speak my language far better than I speak theirs!). I didn't realise this was written by non-native-speakers.

That said, "average" is *not* a word to use when you wish to precise about statistics :-)

Cheers,
Wol

10,000 Linux servers hit by malware (ars technica)

Posted Mar 22, 2014 3:09 UTC (Sat) by creemj (subscriber, #56061) [Link] (3 responses)

When I read the statistics on passwords (other than wondering about 'average') I immediately thought that measure such as mean, median, and whether they contain numbers in addition to letters, etc., are not particularly great measures of password security. Even if a password is 10 (or more) characters long with both letters and digits, if all those characters are either 'a's or '1's then that is a useless password. Indeed, knowing that a password is required to contain at least one capital and at least one digit provides information to reduce the search space (if searching all possible passwords).

Some measure of randomness would be interesting to see. So I suggest the Shannon entropy, s = \sum_i p_i \log p_i, where the p_i is the probability of the ith character appearing in a password, be calculated over all passwords, and the mean Shannon entropy of the passwords be calculated.

(If it is hard to determine the probability of a character appearing in a password then maybe start with the probabilities of a character occurring in a natural language for some appropriate natural language. And modify that a bit by some known transformations, e.g. that a one is often substituted for an el biases ones up and els down.)

If the Shannon entropy is small then the passwords, even though reasonably long, are predictable, thus easily cracked. If the Shannon entropy is large then the passwords have much randomness thus are difficult to predict. What constitutes small or large depends on the base of the logarithm chosen. If base 2 is used then that estimates the minimum number of bits required to express the password. Compare that against the actual number of bits (say 6 or so bits to represent upper and lower case, digits and a couple of symbols) per character to represent the password.

Now that would give a much better measure of password security (assuming a reasonable assignment of prior probabilities to password character selection in the calculation).

10,000 Linux servers hit by malware (ars technica)

Posted Mar 22, 2014 9:45 UTC (Sat) by Jandar (subscriber, #85683) [Link] (2 responses)

This doesn't take into account the order of characters. The password "password" is much weaker than "ospswadr". Even if I make every of my passwords to consist of the same 255 characters (0x01-0xFF) and only reorder them randomly, the resulting strength is more than enough.

10,000 Linux servers hit by malware (ars technica)

Posted Mar 22, 2014 23:07 UTC (Sat) by creemj (subscriber, #56061) [Link] (1 responses)

You are correct. I haven't taken account of the order of characters and that does matter. But, note, that is only a fault in the calculation of the p_i; using entropy is *not* the fault here, as entropy, when properly calculated, is a measure of manifest disorder.

The difference between what I proposed and what should be proposed is analogous to the difference between calculating the entropy of an ideal gas (in which particles do not interact together) and the entropy of a real gas (in which particles interact together if sufficiently close). Thus fixes to the calculation of entropy could be that one should be using p_{i,i-1}, that is, the probability of the ith character occurring given the (i-1)th character. That may be enough to deal with the "password" versus "ospswadr" example, but probably not enough to deal with the preponderance of passwords that consist of a series of letters followed by two or three digits. Thus the p_i may need to be also dependent on the position in the password.

This is now at the point where one needs an extensive password database to calculate the p_i (and the p_{i,i-1}, etc.) but, apparently, some extensive password databases have been posted to the Internet (admittedly by unscrupulous sources), thus calculating these probabilities would appear to be not out of the question.

10,000 Linux servers hit by malware (ars technica)

Posted Mar 22, 2014 23:56 UTC (Sat) by Jandar (subscriber, #85683) [Link]

If there were a global counter used for all passwords on the world, the passwords computed as echo $((++n)) | sha512sum, the entropy based on previous probabilities regarding neighboring characters and so on would be perfect without the merest real disorder.

The problem with such formal tests is the detection of hidden order an attacker knows about.

But that's nitpicking ;-). Any test which prevents most of stupid passwords is welcome.


Copyright © 2026, Eklektix, Inc.
Comments and public postings are copyrighted by their creators.
Linux is a registered trademark of Linus Torvalds