Subject: Re: New hash algorithm - FNV - for use in the kernel
To: David Laight <David.Laight@btinternet.com>
From: Jonathan Stone <jonathan@DSG.Stanford.EDU>
Date: 11/28/2001 11:09:46
>No - most caches are (at least) 2-way set associative, so a clash
>between the table and buffer doesn't really matter.
>It is the time taken to load a line into the cache that matters.
Micro-optimizing like this, with the assumption of a small subset of
contemporary CPUs is all very well, if that's the only target. NetBSD
has to run well on more CPUs -- even contemporary CPUs -- than dreamt
of in your philosophy.
Perhaps we should provide two or three hashes, and have
architecture-specific defopt's to select a "good" one for the
kernel target CPU.