John the Ripper 1.6.38 benchmarks on a Dual 2.7 PowerMac
Written on 7/18/2005
I’m writing a security session for some internal training, thought I would share these John the Ripper benchmarks since I could only find PowerBook benchmarks on the web.
Test platform: Dual 2.7 PowerMac G5 with 1.5GB RAM
Source: John the Ripper 1.6.38 (dev version downloaded 7/18/2005)
Benchmark using make clean macosx-ppc32-altivec-cc
% ./john --test Benchmarking: Traditional DES [128/128 BS AltiVec]… DONE Many salts: 1524K c/s real, 1527K c/s virtual Only one salt: 1328K c/s real, 1331K c/s virtual Benchmarking: BSDI DES (x725) [128/128 BS AltiVec]… DONE Many salts: 51609 c/s real, 51713 c/s virtual Only one salt: 50355 c/s real, 50557 c/s virtual Benchmarking: FreeBSD MD5 [32/32 X2]… DONE Raw: 5135 c/s real, 5145 c/s virtual Benchmarking: OpenBSD Blowfish (x32) [32/32]… DONE Raw: 348 c/s real, 349 c/s virtual Benchmarking: Kerberos AFS DES [24/32 4K]… DONE Short: 231372 c/s real, 232302 c/s virtual Long: 716748 c/s real, 716748 c/s virtual Benchmarking: NT LM DES [128/128 BS AltiVec]… DONE Raw: 8603K c/s real, 8637K c/s virtual
Benchmark using make clean macosx-ppc64-cc
%./john --test Benchmarking: Traditional DES [64/64 BS]… DONE Many salts: 1091K c/s real, 1102K c/s virtual Only one salt: 967155 c/s real, 974954 c/s virtual Benchmarking: BSDI DES (x725) [64/64 BS]… DONE Many salts: 37094 c/s real, 37243 c/s virtual Only one salt: 36416 c/s real, 36488 c/s virtual Benchmarking: FreeBSD MD5 [32/64 X2]… DONE Raw: 4347 c/s real, 4356 c/s virtual Benchmarking: OpenBSD Blowfish (x32) [32/64]… DONE Raw: 273 c/s real, 273 c/s virtual Benchmarking: Kerberos AFS DES [48/64 4K]… DONE Short: 301363 c/s real, 301967 c/s virtual Long: 950323 c/s real, 952227 c/s virtual Benchmarking: NT LM DES [64/64 BS]… DONE Raw: 7680K c/s real, 7696K c/s virtual
As you can see, Altivec makes a pretty big difference in raw speed, although I’m not sure how much the native 64-bit compile would change the raw performance in production. You can compare the 64-bit results with Intel and AMD here., although the benchmark used is an older version. Generally, the dual 2.7 keeps up fine with the single CPUs here.
This seemed a bit odd, since I could see John was running on both CPUs (using MenuMeters). So, I tried launching two instances at the same time. Here are the combined results using 32-bit Altivec for a few of the tests:
Traditional DES [128/128 BS AltiVec] 2907K c/s real 3050 c/s virtual
FreeBSD MD5 [32/32 X2] 9839 c/s real, 10270 c/s virtual
OpenBSD Blowfish (x32) [32/32] 652 c/s/real, 396 c/s virtual
So, pretty efficient on two CPU’s, although you would need two different datasets to work on or two different types of passwords that you were trying to crack on the same data. I’m not sure why John is much more efficient with two processes than with one.
If anyone wants to see any other particular tests with John, just let me know.
Disclaimer: Not an exhaustive test. Actually, I had other stuff loaded (Firefox, etc…) so I could have been able to squeak out a bit more performance.
Todd,
How did you compile john? I noted in my tests that gcc-3.3 gave better results than gcc-4.0 (default gcc for Tiger). I have a compiled version of John at http://www.macunix.net/JTR/john-1.6.38-macosx.zip .
I’d like to get some benchmarks if possible. Thanks.
Erik
Erik,
I’ve been slammed at work but I’ll try out your binary tomorrow!
- Todd
I’m getting 5013k in traditional DES in windows xp 2.8ghz HT p4, 2gig ram, with 25pws in the list
oh, thats v1.6.39
torn,
Those seem like great results. I see in the 1.6.39 changelog:
* Bitslice DES code for x86 with MMX: more than twice faster on Intel MMX.
I wonder if that made the difference for you. I don’t know if that feature was new to .39.
My friends in federal say there are some better ppc optimized versions out there too, but I haven’t looked for them. I’m curious to get my hands on a PowerMac Quad and see how that does.