On Fre 16.10.2009 06:32, Willy Tarreau wrote:
On Fri, Oct 16, 2009 at 12:56:02AM +0200, Aleksandar Lazic wrote:
does anybody have seen or used this 'new' hash?
For me it looks very promising ;-)
I had already downloaded version 2 8 months ago but I was not
interested because it assumes int-aligned input data, which is
obviously wrong for text. On x86 this will not be a problem since the
processor is able to fetch unaligned data, but on most RISC processors,
you'll get an alignment fault when trying to fetch unaligned data.
However, it might be interesting to test it with the consistent hash to
place the servers. And if it's well spread, then we can re-implement it
with bytes on input, so that it works everywhere.
I will take a look in the next year ;-)
You know this issues is a 'timekiller' ;-)