[Ecls-list] string hashing

[Juan Jose Garcia-Ripoll]

2007/1/13, Dustin Long <dlong at stevens.edu>:
> > and I see that ECL was relatively slow in some areas of interest, namely
> > string hashing.  I am wondering to what extent hash table performance
> > has improved since then?  Has there been any rewrite of this functionality?
>
> Looking at the code in hash.d, the hashing looks really complicated,
> with many operations per byte. I didn't do any analysis or testing
> myself, but maybe if someone would like to improve it, I've heard that
> Jenkin's One-at-a-Time works really well:

The code in ECL changed some time ago from a simple rotate & xor
hashing to CRC32 and later on to the SBCL algorithm for hashing. The
latter was due to some emails in the mailing list pointing out the
slowness and also the large amount of collisions with _both_
algorithms

I am open for improvements here, but this is definitely a field where
other people can contribute/test without much hassle :-)

Juanjo

--
Dpto. de Fisica Teorica I, Fac. de CC Fisicas, Universidad Complutense,
Ciudad Universitaria s/n Madrid 2804 (Spain)
http://teorica.fis.ucm.es/~jjgarcia/