Il ven, 2003-06-06 alle 18:54, JJ Furman ha scritto:
> Hello Mario,
> 
> I have been planning to do something almost exactly like this.  Using IPv6
> is smart because so much work has already been done for you.  

>   I've been working on a protocol that is a mix of greedy
> geographic routing, link state and distance vector.  


>   GPSR is a good starting point but probably not enough for the
> kinds of topologies to be expected.
> 

I agree with you and I will post more documentation about my algorithm
in following days.

> What function are you planning to use to map lat/long to IPv6 address?  At
> least two reasonable schemes have been put forth.  I am currently designing
> around the proposal by Hain:
> http://www.globecom.net/ietf/draft/draft-hain-ipv6-pi-addr-02.html.  It
> makes a lot of sense because it leaves the bottom 64 bits free for the MAC
> address making DHCP easier to implement. 
> http://www.itee.uq.edu.au/~leonard/essay/geographic.html  

I have yet to decide, but 95% the first one.

> 
> I read through your website.  How do you plan to write this in Java? 

Sorry but... I have already written most of the program, look at CVS.


>  I'm a
> big Java fan and long time programmer but I think it will be infeasible in
> the long run because Java runs in user space, thus take a huge performance
> hit.  It might be possible to write all of the interrouter communication in
> java but eventually the routing itself will probably have to be done in the
> kernel. 

I am a fan of microkernels.
Java is slow but now there is GCJ compiler that compiles directly to x86
code so I pratically use java as a better c++. 
Decisions are taken by java part, route add and delete is a matter of
using kernel netlinks sockets. For a test I will fake them with
system("iproute..."); calls, but I need someone that helps me with this.
I suppose again that working in user space is not slow: look at zebra
software, it does all calculations in user space, after that it calls
kernel to set routes. The zebra hurd version define it self as a kernel
part.


>  I was considering using the click router from MIT:
> http://www.pdos.lcs.mit.edu/click/   It's very modular and has built in
> functions for traffic shaping, packet modification, IPv6, ethernet, etc and
> its easy to write new modules. What do you think?
> 

I have seen it and printed some documentation and studied it for a week.
I liked the idea that with a switch you can recompile your code as a
kernel module.
But, I have then discarded it because it does not use at all kernel
network part. It replaces all things, traffing shaping, etc. etc.
So if they do not make a version for future kernels we are out of luck.



> I've gotten quite a bit of resistance from people asking me why they would
> want an alternative network.  Since I can't come up with a way to convice
> them, I have stopped trying.  But even as a research only project I think
> this is fascinating.  It puts a whole new dimension to what is possible with
> these little commodity wifi cards.
> 

Yes, and I am researching it for my master thesis ;-)

I think that, because wi-fi will get cheaper and cheaper in the future,
more people will join: we are just at the beginning.
And wireless will be useful in a lot of places (here in Italy adsl and
other fast internet connections are not available in lot of places)

> We can probably take this conversation offline but I wanted to echo my
> interest to the group at large.  Does anyone else share this vision of an
> alternative network?  Think it is useless or totally infeasible?
> 
I am interested too!
-- 
Mario Giammarco
Via Calamandrei,5-48022 Lugo (RA) ITALY
tel: +39054522965 cell: +39(0)3284146845
[EMAIL PROTECTED]
[EMAIL PROTECTED]
[EMAIL PROTECTED]

--
general wireless list, a bawug thing <http://www.bawug.org/>
[un]subscribe: http://lists.bawug.org/mailman/listinfo/wireless

Reply via email to