In addition to knowing what parameters to change one of the biggest problems
with tuning the cache parameters is actually being able to characterize
the usage patterns of the machine in question (and then knowing what to change)
e.g. Tuning for a CAD workstation is not the same as a machine in a general
purpose Unix lab, which is not the same as an AFS admin's workstation.
A thought I've had is about the feasability of a program which might take
the output of something like a kdump and make recommendations for changes on
that specific machine. At last year's DECORUM I mentioned this to several
people who thought it was too hard and complex....But since this tuning
stuffs come up again I thought I'd at least bring the thought up again.
This might also be doable on fileservers??
What do others think?? Or has someone already done something along this line??
Rich
Rich Sudlow
Office of Information Technology
University of Notre Dame
