Re: Link Utilisation

[Nguyen Huu Thanh <thanh@informatik.unibw-muenchen.de>]

Hello all,

> Back to link utilisation...
> 
> If you have a duplex link from say node 0 to node 1 the utilisation (as a
> percentage) can be said to be the number of bits on the link at time x
> divided by the total bits the link can contain. (ie 1Mbit/s, 10ms can
> contain 1M*0.01s=10,000bits)
> 

The definition of link utilization here sounds incorrect. As I
understand, link utilization is the bandwidth that a traffic stream
takes from the total link capacity, it has no relation with the link
latency. In my approach I get the number of byte departures from the
output queue in a sample interval, then divide it by this sample
interval, the result is the link utilization. Then I reset the variable
QueueMonitor::bdepartures_ and begin on the next interval.

For more information on queue monitor, refer to ns Notes and
Documentation, Chapter 17 as well as queue-monitor.h,cc; ns-lib.tcl and
ns-link.tcl.

Regards,

Nguyen Huu Thanh

> But to calculate the total bits on the 'wire' is the part I had most
> difficulty with (seems simple now). If the queuemonitor has a settable
> sample period...
> 
> set qmon [$ns monitor-queue $n0 $n1 $tf $sampletime]
> (Is this correct?)
> 
> and If sampletime is set as the 'delay' time for the link (ie the time for a
> bit to travel end to end) the state variable bdepartures_ will then
> represent the total bits at time x when qmon is queried. The utilisation can
> then be calulated via the links characteristics.
> 
> If you are using various delays for various links in a network and wish to
> calculate the utilisation for many links, one possible method is to use a
> sample time of the lowest common denominator, and for any specific link
> record the last set of values to work out the utilisation... (ie link a is
> 10ms, b is 15ms, c is 20ms, sampling time used would be 5ms and you would
> record the last 4 values for c to calculate utilisation...)
> 
> Sorry a little long winded...
> 
> I am in the midst of writing the code...I thought i would ask for any
> questions suggestions before I finish (and to see if it would work...)

-- 
------------------------------------------------
    Nguyen Huu Thanh                      

    Universitaet der Bundeswehr Muenchen  
    Institut fuer informationstechnische Systeme
    Werner-Heisenberg-Weg 39
    85577 Neubiberg                        
    Tel.: +49 89 6004-2279                   
    Email: thanh@informatik.unibw-muenchen.de  
-------------------------------------------------