Slashdot Mirror
Amazon EC2 Enables Cheap Brute-Force Attacks
snydeq writes "German white-hat hacker Thomas Roth claims he can crack WPA-PSK-protected networks in six minutes using Amazon EC2 compute power — an attack that would cost him $1.68. The key? Amazon's new cluster GPU instances. 'GPUs are (depending on the algorithm and the implementation) some hundred times faster compared to standard quad-core CPUs when it comes to brute forcing SHA-1 and MD,' Roth explained. GPU-assisted servers were previously available only in supercomputers and not to the public at large, according to Roth; that's changed with EC2. Among the questions Roth's research raises is, what role should Amazon and other public-cloud service providers play in preventing customers from using their services to commit crimes?"
The basic story is slightly hysterical. Firstly, WPA2 does use a multiple-iteration key derivation function. Secondly, even with the claimed performance, he can only "brute force" five or six characters, depending on the character set in use. It's enough performance to deal with dictionary words, because, indeed, it's a dictionary attack. But even at 400K password derivations per second (ie 400M SHA-1 hashes per second), eight random characters drawn from the 96 character printable ASCII repertoire are going to take 571 years to perform a brute force attack on, or an average time to success of 285 years. Don't like the odds? My home network uses 12 characters drawn from a 64 character set (ie base 64 encoding), which needs 374 million years (average 167 million) at that performance. Do I give a shit if that number gets reduced by a few orders of magnitude? Not really: I can always move to 15 characters...
The assertion that high end Tesla cards (often $2k) are required for this crack is nonsense. In terms of integer, single precision floating point and memory bandwidth, a GTX 580 is actually FASTER than the most expensive Tesla card. Tesla cards have better QA for 24/7 usage, 4x faster double precision floating point, and 3 or 6 GB of memory, plus some other occasionally useful features. But anyone with an NVIDIA SLI gaming rig built in the last 2 years could easily have done what this guy did in less than 20 minutes.