overhead of namei

[Snoopy]

In article <1245 at muffin.cme.nbs.gov> libes at cme.nbs.gov (Don Libes) writes:
|In article <201 at sopwith.UUCP> snoopy at sopwith.UUCP (Snoopy) writes:
|>"With the addition of the two caches, the percentage of system time
|>devoted to name translation dropped from 25 percent to less than 10
|>percent."
|>
|>	_The Design and Implementation of the 4.3BSD UNIX Operating System_
|>		by Leffler, McKusick, Karels, and Quarterman
|>
|>It appears that namei() soaks up more cycles than one might expect.
|
|You misunderstood the quote.  It doesn't say anything about percentage
|of *total* system time.

>From the beginning of section 7.8:

"Profiling studies show that nearly one-quarter of the time in the kernel
is spent in the pathname-translation routine, namei(), translating
pathnames to inodes."

Then they tell what namei() does, and how they added caches to make it
faster, and finally give the results quoted above (25% -> 10%).

If I misunderstood, I still do.  If they aren't refering to the total
system time, what *are* they refering to?


    _____     						  .-----.
   /_____\    Snoopy					./  RIP	 \.
  /_______\   qiclab!sopwith!snoopy			|  	  |
    |___|     parsely!sopwith!snoopy			| tekecs  |
    |___|     sun!nosun!illian!sopwith!snoopy		|_________|

		"I *was* the next man!"  -Indy