Laine Stump wrote:
> 
> Speaking of performance *improvements*, what was the conclusion (if
> any) of the thread awhile back concerning a performance *loss* from
> 1.4.x to 1.5? (something about lisp and malloc - I forget the details)

Not that I can remember.  I guess it is time to install 1.4.3 and 1.5
on the exact same hardware, and run lmbench a few times.  I will see
if I can run a couple of tests on a spare machine.

> (I've also heard anecdotes from other people using NetBSD in
> production who experienced similar performance problems using 1.5 in
> place of 1.4.x in a cross compile environment). Are we just gaining
> back part of what we (apparently) lost, or is there a bottomline
> improvement from 1.4.x to -current?

Hmm, I guess add -current to the list to test. :)

> At any rate, based on the reports, I guess it's time to switch my
> sources from the netbsd-1-5 branch back to current ;-) (any gotchas in
> the switch?)

I installed 1.5_ALPHA2 from Hubert's ISO image, and grabbed a -current
kernel from another box. Then did 'cvs co src' from my rsynced
repository.
Then I did a make build (completed in about 52 minutes), make cleandir,
and rebooted.  A subsequent make build took about 52 minutes as well.

I cheated by having a -current kernel already, but you should be able
to build one with no problems on 1.5.  At the moment there don't
appear to be any gotchas for running a make build of -current on 1.5.
(my install was about as clean as possible)

-Andrew