>>>>> On Wed, 8 Jun 2005 11:25:13 -0400,
	Thor Lancelot Simon <tls@rek.tjls.com> said:

> If SGI's claims can be believed, BUFQ_PRIOCSCAN should do better both
> on clients and servers than either the traditional sorting (which is
> pretty much CSCAN) or pure read priority.

Well, at the NetBSD BOF of Japan NetBSD users' group in last year,
I've heard from Yamamoto-san (who actually implemented PRIOCSCAN as
you know) that PRIOCSCAN is still not good enough as expected, because
currently UVM doesn't distinguish busy-but-valid pages (for writing)
from busy-and-invalid pages (for reading).

This issue was already reported by him in the year before last:
http://mail-index.netbsd.org/tech-perform/2003/12/17/0003.html
http://mail-index.netbsd.org/tech-perform/2003/12/17/0006.html
http://mail-index.netbsd.org/tech-perform/2003/12/17/0013.html
--
soda