filenames and the kernel

[Chris Siebenmann]

 One good reason to keep namei()'s functionality in the kernel is that
old boogy-man, effeciency. Local figures show that the kernel spends
about 25% of its time in namei(), and between 25% and 50% (the
percentage goes down as the load goes up) of the block reads done on
typical systems are done by namei() -- and this doesn't count symlink
reads done during name resolution (appallingly common on our Ultrix
systems).

 A efficient system with namei() in user space might not look very
Unix-like internally; I'd expect it to have good, fast kernel<->user
process and user<->user communication methods (perhaps based on
remapping VM pages) and a very fast context switch.

[A reminder: please avoid casual use of "usa" distribution; otherwise,
 many interested people will never see your article.]
--
	"I shall clasp my hands together and bow to the corners of the world."
			Number Ten Ox, "Bridge of Birds"
cks at white.toronto.edu		   ...!{utgpu,utzoo,watmath}!utcsri!white!cks