[OpenAFS-devel] how does fileserver read from disk?

Steve Simmons scs@di.org
Thu, 22 Sep 2005 07:36:49 -0400


In any tuning situation, there are two roads to go.  Neither should
be used to exclusion of the other; both are improved by the presence
of the other.

One is to go for the low-hanging fruit, but measure to see if it really
helps and and think moderately both before and after.  One might assume
that the low-hanging fruit has already been grabbed, but if a given
approach hasn't been tried and someones willing to do the experiment,
let 'em go for it.  It often yeilds nice immediate benefits in a short
time frame, but rarely addresses the more complex underlying problems.

The other is to do full analyis.  This is usually good at addressing
the underlying problems, but usually involves a lot of theory and a
lot gaps between theory and practice.  The latter have to be worked
out by experiment and verification, which sometimes looks suspiciously
like the process used in grabbing the low-hanging fruit.

Both methods can and should be used in parallel.  Neither should be
used to the exclusion of the other.  We all have the source.
-- 
Sturgeons Law:      "90% of everything is crap."
Simmons' Corollary: "10% of everything is worth looking at."