[Ns-developers] GSoC: Proposal of Mobile P2P

[Zhijie Shen]

Dr. Marcello Caleff,

Thank you for your effort to read my draft. You comments are valuable. I
also have something to discuss with you.

On Tue, Mar 30, 2010 at 4:46 AM, Marcello Caleffi <marcello.caleffi at unina.it
> wrote:

> Dear Shen,
> below you find some preliminary comments. Please be very active in the
> discussion, i.e. please defend your chooses against my observations if you
> believe I'm wrong.
>
> > Like the ISO architecture, I abstract the P2P framework as a standalone
> layer, i.e. overlay between the networking and the application layers.
> I doubt that a layered model can be used in Mobile P2P networks, where
> different approaches (traditional, cross layer and integrated) have been
> proposed. However, I agree with you that this can improve the code re-use.
> > Thus, we can decouple the overlay structure and the algorithm with the
> strategy or the template design patterns.
>
> Interesting, but hard to implement.
>
As I mentioned in the draft, the reason why I prefer other simulators to
ns-2 is that it hasn't provided a good p2p framework. I agree with you a
complete and standalone framework is difficult to implement. But it is not
impossible. Please check the OverSim http://www.oversim.org/ and the OPSS
http://minerva.netgroup.uniroma2.it/p2p. They tried to pursue the same
target. See if you have any interest in this. But maybe it is not suitable
for the GSoC project. It requires a long-term effort.


> > Furthermore, a bootstrap module is required because many overlays resort
> to it for starting.
> Please, can you explain which bootstrap services you are referring to?
> > I plan to design an observer to subscribe the nodes in the overlay to
> watch their churn events, messages and the moving information of ad-hoc
> nodes.
> I saw that the tracing module requires a considerable amount of time (three
> weeks) in your plan. Is it strictly necessary? I mean, why you cannot adopts
> standard ns-3 tracing?
> > There is an additional module, i.e. churn generator, which is used to
> generate the node arrival and leave events.
>

> The churn model is useful, but not strictly needed in my opinion. Don't you
> think that the node mobility already provides such kinds of events?
>
I'm not aware any sophisticated modules beyond ./core/log. Can you help me
to name them? Thanks! I have not address this clear. In OMNeT++, there is an
obversor module subsribe the peers and watch their states. My wish is to
provide the similar function to the P2P framework in NS-3. Another issue I
should discuss with you is the mobility. The current module in NS-3 seems to
only support the random and constant models. Is it enough for M-DART? As far
as I know, the clustering model should be another good one.

>
> The general design seems to me very generalist. Please could you give more
> details about the M-DART classes and functionalities?
>
Yes, I've not address M-DART part clearly. I'll try to make up for. As I'm
also working on a paper simultaneously, there should be some delay of my
next draft. But I'll notifiy you as soon as I complete it.

>
> Best,
> Marcello
>
> PS I don't have forwarded your proposal to the mailing list in case you
> prefer to keep it confidentially. However, I'm involving the list in the
> discussion since I really trust that you could benefit from it.
>
>
>
> Il giorno 29/mar/2010, alle ore 14.39, Zhijie Shen ha scritto:
>
> > Dear Dr. Marcello Caleffi, in the attachment is the first draft of my
> proposal. Please help me to figure out the defects and give me some
> suggestions.
> >
> > And Dear Dr. Tom Henderson, if you are willing to give some comments, I
> will appreciate them as well.
> >
> > Thank you all!
> >
> > Regards,
> >
> > --
> > Shen Zhijie
> > School of Computing
> > National University of Singapore
> > <proposal.pdf>
>
>


-- 
Shen Zhijie
School of Computing
National University of Singapore