Conversation started Jan 28, 2012 at 19:20.
Jeff Ferland
Jeff Ferland
Jan 28, 2012 19:20
picocomputing.com
I can't find the summary sheet right now (have a hard copy in front of me), but it has some awesome stats
40 bit wep: cracked by a corei7 in 42 days. cracked by an 18 watt FPGA in 1.72 hours
They sell an array of them that has 48 boards consuming 1500 watts.... can brute-force the entire 40 bit wep keyspace in 2.46 minutes
DES in 20.36 horus
considering I use 1500 watts to heat one room in my apartment during winter, I should replace my heater and make money off it ^.^
Thomas Pornin
Thomas Pornin
@JeffFerland 40-bit WEP cracked by an i7 in 42 days is very unimpressive
Jeff Ferland
Jeff Ferland
@ThomasPornin I know, it's broken and can be done in a minute on my laptop
Exhaustive search for even playing field demonstration of FPGA speed
Thomas Pornin
Thomas Pornin
@JeffFerland I mean, even without precomputations, a single core of an i7 should be able to explore the 40-bit space in less than a week
Jeff Ferland
Jeff Ferland
@ThomasPornin Can you give me a 5-minute benchmark for / reference that? They list 300k keys /sec... if its bullshit, i'd love to be able to point to something to call them on it
Thomas Pornin
Thomas Pornin
@JeffFerland RC4 key schedule is about the speed of encrypting 256 bytes. 300k keys/s means encryption at about 76 MB/s. OpenSSL claims to do RC4 at around 380 MB/s on my machine
Actually key schedule should be even faster than that
(there are less table lookups per round in RC4 key schedule than in encryption)
Jeff Ferland
Jeff Ferland
Jan 28, 2012 19:54
I'll have to put it on my list of things to do for benchmarking when I get home
Thomas Pornin
Thomas Pornin
@JeffFerland Try this: pastebin.com/xTvEWCNJ
it's some code which tries 40-bit RC4 keys, outputting 8 bytes and matching them with a reference output
WEP is a bit different because there is an IV but that should not change speed
on my machine (i7, 2.7 GHz) it goes at 1.34 million keys per second
To bring that figure down to 300k, one would have to generate about 700 to 1000 bytes of key stream per key
Now it is conceivable that in WEP, RC4 is used with such discarding (input RC4 key, produce 1000 bytes which are discarded, then produce some bytes to encrypt the data). Discarding the first bytes of RC4 is usually considered a good idea to avoid some biases on the first bytes. WEP is horrible but maybe they thought of that
Ok, I think I know how they got down to 300k
They attack a full packet, and they have no known plaintext, but a checksum
So they must decrypt a complete packet and recompute the checksum for each key.
This is why they go down to 300k keys/s
Jeff Ferland
Jeff Ferland
Jan 28, 2012 20:10
@ThomasPornin OK, so still good numbers, then?
Thomas Pornin
Thomas Pornin
However, one can usually guess the partial contents of some packets (e.g. ARP requests) so an attacker should be able to obtain some plaintext bytes closer to the beginning of the packet
so, in practice, one should assume that the attacker does 1 million keys/s -- and per core
I have two cores in my CPU, so 2 millions/s, so the total key space in about 6 days
and an average crack in 3 days
(half the key space)
Jeff Ferland
Jeff Ferland
Definitely using this conversation as a resource for, "show me the code, skippy"
 
Conversation ended Jan 28, 2012 at 20:12.