Sander Roobol wrote:
>When it came to my attention that Subversion uses MD5 for it's checksums, I
>suddenly remembered that MD5 was designed as a safer but slower version of MD4.
>This is what RFC 1321 (MD5 Message-Digest Algorithm) says:
>
> "MD5 is slightly slower than MD4, but is more "conservative" in design. (...)
> because MD4 was designed to be exceptionally fast, it is "at the edge" in
> terms of risking successful cryptanalytic attack. MD5 backs off a bit, giving
> up a little in speed for a much greater likelihood of ultimate security."
>
>As the RFC is not very clear about MD5's performance, I searched the web for
>some benchmarks. This one is about a python module that needed MD4 and/or MD5,
>on http://minkirri.apana.org.au/~abo/projects/pysync/swf/librsync.diary
>
> "Quick benchmark of md4 vs md5 using 256K sums of 1K blocks random data. I
> used native md4 and [md5] modules on a Cel-366. Results; md4 12.8secs, md5
> 15.6secs."
>
>I also found an archived thread comparing MD5 to SHA-1:
>From http://www.sandelman.ottawa.on.ca/ipsec/1996/06/msg00003.html
>Someone did some benchmarks and included MD4 too:
>
> "Performance in Megabytes per Second on a 90 MHz Pentium
> MD4 MD5 SHA-1 RIPEMD RIPEMD-128 RIPEMD-160
> 20.9 14.2 6.1 10.3 8.0 5.0"
>
>I ran some benchmarks too, on an AMD XP 1700+ and an Intel Pentium 4 1700MHz,
>using openssl(1). The AMD box runs OpenSSL 0.9.7a on Debian Sid, the Pentium
>runs OpenSSL 0.9.6b on Red Hat 7.2.
>The table below shows the time in milliseconds needed to checksum a 100MB file
>(cached in memory) filled with random garbage.
>I also ran md5sum(1) on that file, and that gives some surprising results.
>Apart from a clear speed difference between MD4 and MD5, there also appears to
>be a quite significant difference between various MD5 implementations.
>
> MD4 MD5 md5sum(1)
> XP 1800+ 550 600 1100
> P4 1700 805 1040 4950
>
>
>These three benchmarks together show that MD4 is approximately 10 to 25% faster
>than MD5. I can't imagine that subversion needs cryptographic safety of it's
>checksums, although I don't exactly know what the checksumming code does. So
>my question is, why is subversion using MD5?
>
>
Heh, another case of misplaced bechmarking. :-)
What you really want to measure is the percentage of time spent in
checksum calculation during checkout/update/commit. I'd guess it's on
the order of several % -- in other words, not worht optimizing until we
have a much faster server, client and libsvn_wc.
--
Brane Čibej <brane_at_xbc.nu> http://www.xbc.nu/brane/
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Received on Fri Apr 25 22:26:53 2003