Re: An attempt at optimising the poll code

On Fri, Aug 04, 2000, Henrik Nordstrom wrote:
> I think evenly dividing the filedescriptor sets won't do you much good.
> Instead a suitable priority mechanism should be found telling how often
> an given FD is polled.

That would be the optimal poll() solution but I'm not there yet. THe
idea is based upon some poll ideas people including andreas and
dancer have proposed over the last 12 months. The idea is that
by splitting the fd sets up, you are not going to get as many
non-ready filedescriptors, and so whilst you are handling requests
the OS can populate other buffers.

I will try profiling this code against the current squid code with
a best-effort algorithm to get some hard figures. I've been watching CPU
load on my machine and I've noticed a speedup, but no graphs to prove
it.

> With current code, I think something like this would do good:

[snip]

> Or maybe to have a simple design based on three priorities:

[snip]

Its a nice idea, but I'm going to be concentrating some more on
other areas - I am hoping someone here can improve on the
existing ideas with some code and / or give some benchmarks.

This idea just has been bugging me for a while, so I thought I should
get it over and done with. :)

Adrian
Received on Fri Aug 04 2000 - 12:41:41 MDT