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

Timers, time resolution and traffic control

Hello all,
I would like to ask some questions concerning the accuracy of the
bandwidth estimation both at the estimator and at the tbf qdisc.
As I understand, the PSCHED_CPU option is the best option for the
PSCHED_CLOCK_SOURCE constant for Pentiums with rtdsc (is rtdsc a typing
error? I believe that it must be rdtsc, ReaD Time Stamp Counter). At the
include/net/pkt_sched.h file it is mentioned that not  all Pentiums have
the rdtsc instruction, so which are those? (possibly the non Intel
586?). Also, should the APM (power management) be turned off? So the
PSCHED_CPU option provides time measurement with great accuracy, (CPU
cycle time?).
Another issue is the minimum timer expiration time. I believe that this
is 10 msec for all Pentiums. I would like to find out if there is an
option to reduce this time. I have seen that there is a usec resolution
timer project at http://hegel.ittc.ukans.edu/projects/utime/index.html.
Has this been tested with the diffserv patch, in order to program a
timer of less than 10msec? (At sch_tbf.c it is written that the maximum
bit rate that can been shaped is only 150kbits/sec) Are there any other
alternatives to address this problem?

Thanks you all.

Panagiotis Stathopoulos