[Click] SMP Click - a doubt

[Giorgio Calarco]

----- Original Message ----- 
From: "Eddie Kohler" <kohler at CS.UCLA.EDU>
To: "Giorgio Calarco" <gcalarco at deis.unibo.it>
Cc: <click at amsterdam.lcs.mit.edu>
Sent: Monday, April 05, 2004 3:03 AM
Subject: Re: [Click] SMP Click - a doubt


> >
> > I have a big big doubt: can the "enable-adaptive" option be used
> > together
> > with
> > the "enable-multithread" one ? Is it effective ? If so,
> > are the min-cpu-share nad max-cpu-share values applied to both
> > processors ?
>
> We have not tested this configuration, but it should work.  (The current
> code does set the CPU share values to all threads.)

ok

>
> > I ask this since if i monitor the CPUs activity using top, the
> > min-cpu value seems not to be strictly applied.
> > When I do not inject any packet into the router, CPU#0
> > is almost idle, while CPU#1 works respecting the min-cpu threshold.
> > On the opposite, when I inject traffic, CPU#0 is kept busy, while
> > CPU#1 is idle most of the time (at least that what seems to happen,
> > consider that top has got 1 sec granularity).
> >
> > Thus, is that what I should expect to happen ?
>
> Two things:
>
> 1. Can you run this test on click-1.4pre1 and tell us what happens?


we are using the e1000 driver, thus I was waiting to move to
click 1.4 since I was reading this email a few days ago from Joao Veiga:

------------------------------
My network cards are Intel Pro 1000 and i am using the driver that comes
with the new Click version 1.4.

The problem is that during the tests i get the error message:

ToDevice eth4 refused a packet!
------------------------------


> 2. Do the min- and max-cpu thresholds work without --enable-multithread?

mmm, maybe I have not undertood anything at all...
The "enable-adaptive" option worked correctly with a monoprocessor system,
but isn't the "enable-multithread" option mandatory for a SMP system ?
Isn't it the option to use to say click that we have got more that one CPU ?


thanks (again)

giorgio


>
> Eddie