[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: TC calling sequence timing



Hi all,

> Several things you can do:
> -Change the HZ value if you have >= Pentium pro. (careful: change from
> 100->1024)
> -Select CPU as the clock source in pkt_sched.h

My tests are already performed based on the above configuration


> - Linux doesnt do the EOI functionality; but it should be easy to add;

Really? Well probably this is the reason for this behaviour; I have
misunderstood that after EOI net_bh is marked active and so run queues is
performed.

>
> Well, it is not easy to believe that it is faster.
> It only adds more work, right?
>
> Also, when EOI occurs, qdisc will not give you something to send
> because of shaping constraints and main work will be made from timer
> or on push of the next packet in any case. Hence, shaping will not
> be more accurate even a bit.

Yes work will be done from the timer, but it will programmed to do the
transmittion at the proper timer and not a more or less random interval (at my
example timer 15ms, random interval aproximatelly 150ms), so shaping will be
in fact more accurate, but any way it sure adds more work for this very
special case.

>
>
> I am not sure, really... Well, it is even not 2.5, just add it now
> and check. 8)
>
> Alexey

Thank you both!
Panagiotis
--
-----------------------------------------------------
Panagiotis G.Stathopoulos
National Technical University of Athens
Telecommunications Laboratory
E-mail: pstath@telecom.ntua.gr
Tel: +30 1 772 1479     Fax: +30 1 772 2534
------------------------------------------------------