Re: Kernel Scheduling

[Seth Arnold <sarnold@wirex.com>]

On Thu, Jun 06, 2002 at 04:09:52AM +0000, Gaurav Sehgal wrote:
>    I am trying to come up with a new kernel scheduler for experminental 
> purposes. The new strategy will schedule a be as follows:

Sounds like a fair scheduler. Google ought to know a bit about these, as
would NEC's citeseer. :)

> If there are N different users in the run-queue, and a user A has P
> processes.  Then how can I calculate what percentage P should get.
> Using that value I will need to change the value of the weight in the
> goodness function of sched.c right?

Heh, if we answer this one, won't that defeat much of the purpose of the
research? :) (Critical thought about common cases and making sure to get
all the corner cases as well, is my guess. :)

A naive first approach would iterate over all tasks, and keep track of
which users have how many processes running, and then make a second pass
setting the relative priorities to match the values you can calculate at
the end.

There might be a better way. :)

-- 
http://sardonix.org/

PGP signature