On Jun 28, 12:47pm, Arne Henrik Juul wrote:
> >   This is getting very, very strange.

  Not so strange :-)

> Does anybody have any bright ideas on this?  I've done some empirical
> testing by adding the following to machdep.c right before the signal
> debug data:
...
> Ignoring the noise it's clear that some groups of 4 values
> give the strange drop in writing speed.
> 
> The mystery deepens, I'd say...

  I tracked down this "mystery" a couple of weeks ago, and determined it's
not a pmax problem, although the write performance slowdown occurs on
the pmax.  See PR 3748 for more information and a suggested workaround
(and maybe fix?) for the "problem".

  The slowdown on the pmax occurs because of the coincidence of the console
font data showing up in memory where it does and the font data containing
a number of words containing 1.  Adding or removing data between bdevsw[]
and the font data will change the slowdown.  It occurs in groups of 4 words
because the size of the data segment generated by gcc is rounded up to 16
bytes.

Michael

-- 
Michael L. Hitch			mhitch@montana.edu
Computer Consultant
Information Technology Center
Montana State University	Bozeman, MT	USA