nice interface. but.

i would suggest that the results we so far have show that there are a
number of good algorithms with relatively minor performance difference
in simulated inputs; i am not sure the extra interface blubber for two
*separate* hash functions is as strongly justified as one may like. this
is not a caution against any of the choices [both of which are easy to
comprehend, and seem to perform as well or better than FV+n, and have
lower pomposity quotient], but against needless complication. i would
be astonished if anyone could show that for any live production netbsd
system on any architecture, the difference between these two relatively
good functions actually gained us something beyond what would be seen
as noise. 

anyhow, i expect simon to produce more tests soon, and this sure will
be the one kernel group to have explored hash functions to its fullest. :)

oz
---
[ps: thanks for the opportunity to contribute to the discussion. i will
let everyone know when i have a draft of my McKenzie followup paper
available, tentatively titled, "selecting a better hashing algorithm"]
---
ozan s. yigit			staff engineer, sun microsystems/es
http://www.cs.yorku.ca/~oz	ozan.yigit@sun.com || +1 [905] 415 2878
---
narrowness of imagination leads to narrowness of experience. --oz