[Bcc to committers and -stable as relevant there, but
please keep the discussion on -net if any]
I have put up a web page with the latest version of the device
polling patches, both for -stable and -current, at
http://info.iet.unipi.it/~luigi/polling/
so new versions will be available through there, together with
additional documentation, code and references.
Compared to the code recently pulled out from -stable, what you
can find now at the above URL has the following improvements:
* "scheduling" code which lets you control in a reasonably
fine-grained way the sharing of CPU between kernel (polling)
and userland. You can now do something like
sysctl kern.polling.user_frac=40
to reserve 40% of the CPU time to userland processes
(assuming there is enough work to do). Default value is 50.
* a restructured polling loop which prevents livelock and
unfairness even when the standard (no fastforwarding) path
is used.
Together with the kern.polling.enable variable, kern.polling.user_frac
is the only tunable parameter that you need to care about, and the
default value should be a reasonable one.
It would be great if some of you committers would have the time to
have a look at the -current version, or possibly even try it on
Alpha boxes (I think overall there is only one or two lines of MD
code), and send feedback.
thanks
luigi
----------------------------------+-----------------------------------------
Luigi RIZZO, [EMAIL PROTECTED] . ACIRI/ICSI (on leave from Univ. di Pisa)
http://www.iet.unipi.it/~luigi/ . 1947 Center St, Berkeley CA 94704
Phone: (510) 666 2927
----------------------------------+-----------------------------------------
To Unsubscribe: send mail to [EMAIL PROTECTED]
with "unsubscribe freebsd-net" in the body of the message
