posted by
LaminatorX
on Saturday May 31 2014, @11:48AM
[Skip to comment(s)]
from the Ogettingsmaller dept.
from the Ogettingsmaller dept.
A class of Discrete Logarithm Problems (DLPs) used in a 128bit crypto scheme allegedly cracked in two hours (paper). It can also be formulated as breaking 128bit secure supersingular binary curves or how to solve discrete logarithms fast. Findings to be presented at the IACR Crypto 2014 conference.
This discussion has been archived.
No new comments can be posted.
128bit Discrete Logarithm Crypto Solved in 2 Hours

Log In/Create an Account
 Top
 6 comments
 Search Discussion
The Fine Print: The following comments are owned by whoever posted them. We are not responsible for them in any way.
(Score: 1, Flamebait) by Anonymous Coward on Saturday May 31 2014, @12:38PM
She said she was discrete but I turn my back for two hours and she spill all my secrets. I knew I couldn't trust that bitch!
(Score: 1, Troll) by Anonymous Coward on Saturday May 31 2014, @01:23PM
That's why you stick it in her pooper then donkey punch her.
(Score: 2, Informative) by Anonymous Coward on Saturday May 31 2014, @08:32PM
That's what you get for being illiterate.
If only she were discreet...
(Score: 2) by Zyx Abacab on Sunday June 01 2014, @03:10AM
From TFA:
So, no, this doesn't mean that current encryption using 128bit keys is broken. With sufficient key sizes, AES, Twofish, and Serpent are still not breakable in any reasonable timeframe.
In fact, the research does not show that discrete logarithms are conceptually broken! The only thing that's been demonstrated is that supersingular binary curves, in particular, are not suitable as an encryption scheme.
(Score: 2, Informative) by No.Limit on Sunday June 01 2014, @08:52AM
AES, twofish & serpent don't even use discrete logarithms. So these ciphers are not affected.
Discrete logarithms can never be 'broken' conceptually. Every crypto that uses discrete logarithms is based on the decisional diffiehellman assumption [wikipedia.org], which basically states that discrete logarithms are hard to compute. The research shows that this assumption doesn't hold specifically for supersingular curves (which are mathematical groups) with 128 bits. So the research shows that this class of groups isn't suitable for crypto based on discrete logs.
Let's elaborate a little on this:
A & B agree on a common number g (an adversary can see it).
Then A generates a private random x and B generates a private random y.
A > B k (=g^x)
B > A m (=g^y)
Then A can compute m^x = (g^y)^x = g^(x*y) and B can compute k^y = (g^x)^y = g^(x*y) so they just shared a private key.
So, about secrecy of this exchange:
An adversary only sees g, k = g^x and m = g^y. If they could find out x or y that is log(g^x) = x (where log is in base g) they find the shared private key.
In this key exchange we've only used exponentiation. Let's say multiplication can be done in constant time, then we can compute exponentiation g^n in asymptotic time complexity O(log(n)) (now with any base) with the fast powering method [wikipedia.org].
What about computing the logarithm?
Well for real numbers we can use that u diffiehellman key exchange), after which the secret shared key can be used as key to any cipher such as AES, twofish, serpent etc.
(Score: 2, Informative) by No.Limit on Sunday June 01 2014, @08:59AM
Ignore the last paragraph, because of the smaller equal a big chunk of text got missing. It should continue like this:
What about computing the logarithm?
Well for real numbers we can use that
and use a binary search (first expanding and then narrowing) to compute it in O(log(n)).
However in cyclic groups this does not hold (in cyclic groups the notion of u smaller v doesn't make much sense).
So we can't use the above trick. Let's say we're unlucky and are only left with a bruteforce trial and error which would take O(n) time.
Now we have O(log(n)) for exponentiation and O(n) for logarithm. We set n' = log(n) and get O(n') for exponentiation and O(e^n') for logarithm.
And that's the relationship we're using to make sure that no adversary has enough computational power to successfully compute the logarithm.
With this we just need groups where the logarithm can't be computed efficiently. Well supersingular curves probably aren't suitable.
Now if P = NP we can show that no such group exists. In fact with P = NP we can break any practical scheme and would be left with OTP as the most efficient crypto. Luckily P = NP is a very hard unsolved mathematical problem and there are still many classes of groups where the decisional diffiehellman assumption is believed to hold true.
Btw the name discrete logarithm comes because these groups aren't continuous like the real numbers. They can be represented by integers.
Discrete logarithms are often used for key exchange (for example in the diffiehellman key exchange [wikipedia.org]), after which the secret shared key can be used as key to any cipher such as AES, twofish, serpent etc.