Polling For 100 mbps Connections? (Was Re: Freebsd
danial_thom at yahoo.com
Fri Dec 16 07:25:10 PST 2005
--- Cezar Fistik <cezar at arax.md> wrote:
> Hello all,
> > Or maybe FreeBSD just sucks wind? I promise
> > that no machine known to man can pass 150Mb/s
> > be 99% idle. Get a god-damned clue for pete's
> > sake. All polling does is screw up accounting
> > the timings are wrong. At best there's a
> > difference in performance. You still have to
> > process the packets.
> > Lets see, this "Test" would mean that his box
> > could handle 100 x 150Mb/s, or 15Gb/s. Thats
> > quite a little router you have there! lol.
> I have to recognize that I didn't follow the
> thread very carefuly and
> I propbably missed some posts. I just wanted to
> say that when using
> freebsd as a pure router using intel cards,
> polling realy helps. I've
> noticed 10-20% CPU utilization decrease with
> polling enabled. Second, I
> didn't mean 150Megabytes/sec, rather
> Megabits/sec. Third the actual
> CPU load is ranging from 0-4% (according to
> top), but with an average
> of 1%. And finally I just did a test and
> enabled polling on that box.
> The CPU idle state immediately dropped to 88%
> while interrupts
> increased to 10-12%.
> Danial, have a look at this
With all due respect, you haven't a clue.
Do you know math? 99% idle implies 1%
utilization. So 150 Mega Bits per second X 100 is
15gigaBITS per seconds is the capacity of your
system, using match. Which is stupid. Anyone who
thinks that is accurate is just plain stupid.
THE MEASUREMENTS DON'T WORK WITH POLLING ENABLED.
Are you able to understand that?
Do You Yahoo!?
Tired of spam? Yahoo! Mail has the best spam protection around
More information about the freebsd-questions