Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [day] [month] [year] [list]
Date: Sun, 24 Jun 2012 01:57:38 +0200
From: Frank Dittrich <frank_dittrich@...mail.com>
To: john-dev@...ts.openwall.com
Subject: Some relbench results for linux-x86-sse2i-omp (jumbo5 vs. upcoming
 jumbo6)

Number of benchmarks:		149
Minimum:			0.61545 real, 0.59096 virtual
Maximum:			6.19378 real, 3.11196 virtual
Median:				0.99561 real, 0.99478 virtual
Median absolute deviation:	0.01514 real, 0.01241 virtual
Geometric mean:			1.07296 real, 1.02489 virtual
Geometric standard deviation:	1.30634 real, 1.24840 virtual

Top 10 improvements:

6.19378 real, 3.11196 virtual	Tripcode DES:Raw
2.88779 real, 1.45131 virtual	Invision Power Board 2.x salted MD5:Only
one salt
2.69135 real, 2.69135 virtual	Oracle 11g SHA-1:Many salts
2.57270 real, 2.57270 virtual	Oracle 11g SHA-1:Only one salt
2.16337 real, 2.14185 virtual	SAP CODVN B (BCODE):Only one salt
2.05668 real, 2.03593 virtual	SAP CODVN B (BCODE):Many salts
2.04924 real, 2.06923 virtual	Mac OS X 10.4 - 10.6 salted SHA-1:Many salts
1.93262 real, 1.94195 virtual	Mac OS X 10.4 - 10.6 salted SHA-1:Only one
salt
1.92772 real, 0.97430 virtual	HTTP Digest access authentication MD5:Many
salts
1.90857 real, 0.96882 virtual	HTTP Digest access authentication MD5:Only
one salt

Where real and virtual ratio differs by a factor of 2, the jumbo-5
version didn't support OMP for that format.

And here are the 10 worst regressions:
0.61545 real, 0.61545 virtual	MediaWiki md5($s.'-'.md5($p)):Many salts
0.72660 real, 0.72660 virtual	MS SQL SHA-1:Only one salt
0.81736 real, 0.81736 virtual	dynamic_26: sha1($p) raw-sha1:Raw
0.85106 real, 0.84256 virtual	MySQL:Raw
0.89833 real, 0.89386 virtual	generic crypt(3) DES:Only one salt
0.90378 real, 0.89933 virtual	generic crypt(3) DES:Many salts
0.91972 real, 0.91033 virtual	dynamic_24: sha1($p.$s):Many salts
0.92165 real, 0.93106 virtual	dynamic_9: md5($s.md5($p)):Many salts
0.93204 real, 0.93204 virtual	Generic salted MD4:Many salts
0.93643 real, 0.93643 virtual	NT MD4:Raw


Now a comparison of sse2 without OMP and sse2i with OMP for current git:

$ ./relbench j6-sse2.txt j6-sse2i-omp.txt
More than one benchmark for MySQL:Raw in file 2
More than one benchmark for NT MD4:Raw in file 2
Could not parse: Benchmarking: Raw SHA-1 [128/128 SSE2 intrinsics 4x]...
FAILED (get_hash[0](0))
Could not parse: 1 out of 138 tests have FAILED
Only in file 2: Mozilla SHA-1 3DES:Only one salt
Only in file 2: Mozilla SHA-1 3DES:Many salts

So,there's still the Raw SHA-1 bug, and I forgot to activate the Mozilla
format when compiling sse2 for comparison with jumbo 5.

Number of benchmarks:		199
Minimum:			0.67287 real, 0.37143 virtual
Maximum:			10.91048 real, 7.48647 virtual
Median:				1.01616 real, 0.90334 virtual
Median absolute deviation:	0.08654 real, 0.12691 virtual
Geometric mean:			1.06466 real, 0.81586 virtual
Geometric standard deviation:	1.29945 real, 1.41290 virtual

Here are the top 10 improvements with sse2i-omp over sse2 withput OMP:
10.91048 real, 5.45530 virtual	SIP MD5:Raw
7.48647 real, 7.48647 virtual	MySQL:Raw
1.54819 real, 0.79813 virtual	PKZIP:Many salts
1.44048 real, 0.71176 virtual	OpenBSD Blowfish (x32):Raw
1.41867 real, 0.71702 virtual	HTTP Digest access authentication MD5:Many
salts
1.38333 real, 0.68822 virtual	Lotus Notes/Domino 5:Raw
1.36854 real, 0.68770 virtual	RACF DES:Only one salt
1.36616 real, 0.69331 virtual	SSH RSA/DSA (one 2048-bit RSA and one
1024-bit DSA key):Raw
1.36368 real, 0.67509 virtual	MS Kerberos 5 AS-REQ Pre-Auth MD4 MD5
RC4:Many salts
1.35857 real, 0.67929 virtual	RACF DES:Many salts

And here are the top 10 of sse2 without OMP being faster than sse2i with
OMP:

0.67287 real, 0.67966 virtual	HMAC SHA-1:Only one salt
0.68155 real, 0.68155 virtual	MS SQL SHA-1:Only one salt
0.74027 real, 0.37196 virtual	LM DES:Raw
0.75238 real, 0.37143 virtual	Office 2007/2010 SHA-1/AES:Raw
0.77074 real, 0.76311 virtual	dynamic_26: sha1($p) raw-sha1:Raw
0.81399 real, 0.81399 virtual	Raw SHA-1:Raw
0.82016 real, 0.41208 virtual	Tripcode DES:Raw
0.82105 real, 0.82941 virtual	Raw SHA-1 LinkedIn:Raw
0.82706 real, 0.41767 virtual	BSDI DES (x725):Many salts
0.83461 real, 0.41729 virtual	BSDI DES (x725):Only one salt

Amazing.

Frank

Powered by blists - more mailing lists

Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.