[Click] SortedTaskSched

[Giovanni Tusa]

Hi Eddie and Nikitas.
In the tests I have done with the old SortedTaskSched element, by monitoring
the task assignment with the ThreadMonitor element, I sometimes observe some
strange behaviors. It seems that the assignment of the tasks among the CPUs,
in order to balance the load, sometimes does not corresponds to the number
of
the calculated CPU cycles.
For example, if I have:

    pd0::PollDevice(eth0);
    pd1::PollDevice(eth1);
    .................................
    td0::ToDevice(eth0);
    td1::ToDevice(eth1);

I can observe a situation like this:
chatter: 58869: pd0, cycles 22874, on 0

chatter: 58869: td0, cycles 8000, on 0

chatter: 58869: pd1, cycles 22561, on 1

chatter: 58869: td1, cycles 7369, on 1

instead of : pd0 and td1 on 0
                 pd1 and td0 on 1

Are there some other mechanisms which can take place during the rebalancing
and
the CPU cycles calculation?
Moreover, which are the main differences in the implementation of the new
BalancedThreadSched element?

Thanks
Giovanni

----- Original Message -----
From: "Eddie Kohler" <kohler at cs.ucla.edu>
To: "Nikitas Liogkas" <nikitas at cs.ucla.edu>
Cc: "Giovanni Tusa" <gtusa at diit.unict.it>; <click at amsterdam.lcs.mit.edu>
Sent: Monday, June 21, 2004 6:56 PM
Subject: Re: [Click] SortedTaskSched


> Nikitas Liogkas wrote:
> > INCREASING just dictates in what order the tasks will be sorted
> > (increasing or decreasing) before the rebalancing takes place.
>
> Yes.  If INCREASING is true, then the load-balancer moves the most
expensive
> tasks first.  If it is false, the load-balancer moves the least expensive
tasks
> first.
>
> Eddie
>
>
>
> >
> > Eddie has recently rewritten that particular element, so it might be
worth
> > it to wait for a few days in order to try out the new implementation.
> >
> > nikitas
> >
> > On Mon, 21 Jun 2004, Giovanni Tusa wrote:
> >
> >
> >>Hi all,
> >>by using the SortedTaskSched element of SMP Click, in order to enable
adaptive CPU scheduling in a Xeon dual processor machine,
> >>I have noticed some change in my router performance.
> >>The question is that, while the meaning of the INTERVAL parameter is
obvious, still I have some doubt about the meaning of the INCREASING
parameter.
> >>
> >>Any advice will be very appreciated!
> >>_______________________________________________
> >>click mailing list
> >>click at amsterdam.lcs.mit.edu
> >>https://amsterdam.lcs.mit.edu/mailman/listinfo/click
> >>
> >
> > _______________________________________________
> > click mailing list
> > click at amsterdam.lcs.mit.edu
> > https://amsterdam.lcs.mit.edu/mailman/listinfo/click
>
>
>