[altq 65] bandwidth oscillation

[Geoffrey CRISTALLO <geoffrey.cristallo@alcatel.be>]

Hi,

I want to test ALTQ on the following topology, but before to start the
tests, I'd like to be sure that the TCP behaviour TCP that I've noticed
is correct...

Maybe this is not the right mailing, but I don't know right one.


The topology is the following. There is no ALTQ running, only FIFO
queuing.

	____     ____     ____
       |    |___|    |___|    |
       | V  |   |  S |   |  W |
        ----     ----     ----
		   |
	         __|__    
                |     |  
               /|  P  |\ 
              /  -----  \
             /           \
	    /             \
          _/__             \____
         |    |            |    |
         | Q  |            |  R |
          ----              ----

I begin with 4 TCP flows toward the same destination:
* V - R
* W - R
* Q - R
* P - R

The bandwith is equally divided among these 4 flows (they all receive a
little bit more than 20 % and the bandwidth for each flow is stable)

But if I add a fifth flow (from S to R), the bandwidth of the individual
flows start to oscillate. Although the total bandwith remains smooth.  

I've tried several combinations of source-destination pair, but the
oscillations always appear when I add the fifth flow.

I was wondering if these oscillations were typical TCP behaviour. Can
somebody justify them ?

Thanks in advance

Geoffrey

       


Geoffrey Cristallo               
 
mailto:geoffrey.cristallo@alcatel.be