[Cdash] ITK appears in GDCM cdash

[Eric Noulard]

2009/9/3 David Cole <david.cole at kitware.com>:
> Julien's not only worried about the speed of the SQL query. He's also (and
> probably more) worried about the speed of computing the hash on large input
> data (test output ranges from a few bytes up to several megabytes depending
> on the test...)
>
> For every test.xml file submitted that contains hundreds (or even thousands)
> of tests, we have to compute the hash on the test output, look it up to see
> if it's already in the db and then insert it if it's not.
>
> Computing the hash on the test output is the dominant thing to worry about
> at test submission processing time.

Yes now I'm clearly aware of that.
That's why I would suggest to look into non-cryptographic hash functions
which were preciselly designed for speed.

FNV (http://en.wikipedia.org/wiki/Fowler_Noll_Vo_hash) was an example,
MurmurHash (http://en.wikipedia.org/wiki/MurmurHash) seems to be another one.

Note again I'm no expert in this area and I cannot exhibit performance figures
against your current crc32 usage, but from that little search result:

http://groups.google.co.uk/group/sci.crypt/browse_thread/thread/56aaa4c236c7b095?fwc=1

some specifically designed "fast non-cryptographic hash" may well
outperform CRC32.

I would add that in this area the hash algorithm itself may not be the
single key to performance. You may have to use a specifically tuned
implementation
if you want bleeding edge speed.


-- 
Erk
Membre de l'April - « promouvoir et défendre le logiciel libre » -
http://www.april.org