[j-sim-users] Re: Sensorsim packet loss

[Conor Beverland]

Hi Nicholas,

Agreed! I'm currently building an in network query processing system for
sensor networks on top of J-Sim (removing the concept of a sink node).
Certainly in the real world I wouldn't want to be using TCP but since I
haven't got any fault tolerance in my code at present, it would be nice
to experiment under "ideal" conditions :-)

Setting the flags as I indicated in my previous mail seem to be good
enough and if I need anything more I think just making a propagation
class that always returns 1 will probably do the trick..

The query system I'm building aims to conserve energy and so I'm very
interested in the stuff you're working on!! The fact that each package
has their own energy model is very annoying. Are you able to get any
useful data from the sensorsim energy model? Unless I'm misunderstanding
something (or, a lot of things perhaps) it seems pretty broken..

How much progress have you made so far? Are you making the changes such
that WirelessPhy can still be used without the sensorsim package? That's
not so important to me but it would be nice to make improvements that
everyone can benefit from.. For instance, (and I've done it already for
WirelessPhy), I think that the energy models definitely need info ports
that can hooked up to the plotter to get some energy data out of the
simulation (which is mostly what I'm interested in).

Anyway, it sounds like I need something similar to what you're working
on and am keen to help if I can.

Cheers,
Conor


On Tue, 2005-05-17 at 18:23 -0400, Nicholas Merizzi wrote:
> Hi Conor
>     It's a little challenging to guarantee 100% delivery rate especially 
> in sensor networks. Since these are energy constrained devices you would 
> not want to use TCP(because of high overhead). Also since J-Sim only 
> currently supports Mac802.11 which is not designed for sensor networks 
> (idle listening, and overhearing problems) you are looking at draining 
> your battery really quickly!
>     An alternative that I am looking into is using a sort of High 
> priority flag on certain packets that need immediate attention.
> 
> If you were to simulate TCP with 802.11 attach a port to a plotter and 
> watch the energy drop really fast! There is also another problem when 
> you combine the wireless package with the sensor package: the energy 
> models. Each package has their distinct energy model. I have re-design 
> the sensor network package so that it correctly fits into the wireless 
> package:
> 
> http://www.cas.mcmaster.ca/~merizzn/
> 
> If anyone has feedback for me on those models or working on something 
> similar let me know!  I am in the process of breaking up the Mac 
> component so that other Mac protocols can be used.
> 
> -Nicholas
> 
> 
> Conor Beverland wrote:
> 
> >Further to my previous question, I've noticed that adding these lines
> >from the wireless tutorial seems to help with the packet loss problems
> >I've been seeing:
> >
> >#enable route_back flag at PktDispatcher
> >! pktdispatcher setRouteBackEnabled true
> ># since 802.11 provides link broken detection, we set the flag in AODV
> >! aodv enable_link_detection
> >
> >However, while enable_link_detection is mentioned in the tutorial, I'm
> >not sure what setRouteBackEnabled actually *does* (haven't had time to
> >delve into the PktDispatcher code and it isn't mentioned in javadoc..)
> >
> >Also, I'd still like to know if it's possible to guarantee packet
> >delivery in the sensor network. Would I have to replace the
> >WirelessAgent class with something like drcl.inet.transport.TCP?
> >
> >Thanks,
> >Conor
> >
> >_______________________________________________
> >j-sim-users mailing list
> >j-sim-users at cs.uiuc.edu
> >http://mail.cs.uiuc.edu/mailman/listinfo/j-sim-users
> >
> >
> >  
> >
>