Re: [ns] link layer feedback and AODV

[Felix Ko <felix.ko@mchp.siemens.de>]

Hi Attila!

I have observed the same effect in my simulations:
In scenarios with high traffic (50 nodes, 40 sources,
4 CBR-Pkts/sec per source -- this the scenario simulated
in the paper written by Das, Perkins, Royer "Performance Comparison of
two
on-Demand ...."), but *no mobility*, the LL permanently reports
LL-errors because it couldn't obtain the medium due to high traffic.
As you stated, this unneseccarily triggers route discoveries by AODV
although no node is moving!

The reason is, that the LL-callback of AODV doesn't distinguish between
the various types of LL-error (Link broken, no medium or no ACK
received).

The reason could be, that AODV should discover a new route, which leads
*around* the congested area of the network...
Nevertheless the route discovery worsens the congested situation,
especially when it is totally unnecessary, because (like in your and my
situation)
no node is actually moving!

Bye

Felix

PS: The most up-to-date version of AODV on Mahesh Marina's homepage
handles LL-errors in exactly the same way



Attila T�r�k wrote:

> Hi!
>
>  I tried to simulate TCP over ad hoc networks, using AODV for routing.
>  I used a simple line topology with a single TCP flow (packetSize_
>  =1460, window_ =20, and MAC 802.11 RTS/CTS).
>  Enabling the local repair function I noticed the following problem:
>    In many cases the MAC layer reports link_layer_failure (cannot
>    obtain media for TCP packets) and triggers the AODV for a new route
>    lookup. Sometimes this problem occures even after the establishment
>    of a new route, thus a RREQ will be initiated again!
>
>    It is normal to have so many MAC layer retransmissions?
>    Any similar experiences with real WaveLAN testbeds?
>
>
>    Thanks in advance: Ati