Setup a wireless network experiment environment

Hi there

as a student I am planning to setup a wireless network experiment
environment by means to wrapper a set of dual-interface linux boxes,
with a scheduler we tell each pc/eth0 interface to let bind to a
certain nd/eth0/ath0 node at specific moment, and then let all
communications between 'pc's can be controlled to actually route
through binded ath0 interfaces via adhoc network NET2:

_____ _____ _____
| | | | | |
| pc0 | | pc1 | | pc2 |
| | | | | |
eth0 |_____| eth0 |_____| eth0 |_____|
| | |
| | |
NET1 | | |
----+------+-----+------+------+------+------+------------
| | |
| | |
| | |
eth0 ,--+--, eth0 ,--+--, eth0 ,--+--,
| | | | | |
| nd0 | | nd1 | | nd2 |
| | | | | |
ath0 |_____| ath0 |_____| ath0 |_____|
: : :
NET2 : : :
..................:............:..............:... ..........


as an example, at time t0, pc0 is binded to nd0, pc1 is binded to nd2,
so all traffics from pc0 to pc1 will actually be routed as

pc0 --> pc0/eth0 --> nd0/eth0 --> nd0/ath0 --> nd2/ath0 -->
nd2/eth0 --> pc1/eth0 --> pc1

and at time t1 pc1 is scheduled to bind with nd1, then frames from pc0
to pc1 will be routed as

pc0 --> pc0/eth0 --> nd0/eth0 --> nd0/ath0 --> nd1/ath0 -->
nd1/eth0 --> pc1/eth0 --> pc1.

have studied long on some references/howtos, but most documents focus
more on issues for routers and firewalls, I am still wondering, is it
ok to setup such an network, and which keys, ARP tables, proxy ARP,
route tables, IP tunneling will make good points in such a environment?
is it necessary, to program several virtual devices so the packages can
first move and switch smoothly?

Just started to study and look a bit deep into the network, the problem
is we still havent be able to get some linux boxes to setup even a
simple environment for some testing, but already some questions as
starting:

1. for those PC/eth0 interfaces, seems a controled ARP table could be a
key, but said from 2.2.0 above set a proxy ARP entry for the whole
subnet on local interface is no longer possible, so maybe we need to
maintain the local ARP table, set permanent entries and update them all
when binding node need to change; could this be on the right direction?

2. difficulty stays here, for interface nd/eth0 and nd/ath0, thought
setup the box as a bridge could be a good possibility, but ath0 must
also route the package back into NET1 through destination binded node
(ndx/ath0->ndx/eth0->pcx), so maybe we could maintain the a very much
detailed route table, specify gateway for each pc destination based on
current pc<->node binding pairs, and update it whenever a pc changes
its binding node? feel bad with this step, thought proxy ARP setup
might help something, but cannot really figure it out. and, we also
like to keep possibilities to use the ethernet connections between
nd/eth0s for some synchronization issues when necessary, so we need to
let nd/eth0 interfaces also keep active, but with a bridge setup,
mostly both ath0 and eth0 will lost their id, is there a way make it
possible?

3. the last step for all nodes they will forward all packages received
from nd/ath0 interface to nd/eth0, then to right destination on the
same network, seems either by a bridge configuration or setup
interfaces manually are ok, but together with the eth0->ath0
configuration and routing conditions, this has to be considered
together... wondering whether there is a clean solution?

thanks,
Paulour

Paulour wrote:

> as a student I am planning to setup a wireless network experiment
> environment by means to wrapper a set of dual-interface linux boxes,
> with a scheduler we tell each pc/eth0 interface to let bind to a
> certain nd/eth0/ath0 node at specific moment, and then let all
> communications between 'pc's can be controlled to actually route
> through binded ath0 interfaces via adhoc network NET2:
>

You might start with an English composition class, so that we can understand
what you're trying to say.

sorry for the difficulties I made, tried to express well...but only
those who read them can tell the answer

might try it in a simpler way: I am trying to setup a network, which
consisted of 2 group of pcs:

- one group each with only one ethernet interface ('pc's)
- another group each has an ethernet interface and a wireless interface
('node's, embedded Linux)
- all ethernet interfaces are physically connected in same subnet
- wireless interfaces are also connected and composed another adhoc
subnet
- with a scheduler, a 1-to-1 relation between 'pc's and 'node's will be
created and dynamically updated, so assign each 'pc' a related 'node';

then the target: I'd like to find a configuration that can force all
traffics among 'pc's to be routed actually through their relevant
'node's, then through the wireless network, as if these 'pc's are
totally unwired with each other.

with a diagram(preview looks fine, but post still create
problem...please check with "show original" option if it looks still
messed):

_____ _____ _____
| | | | | |
| pc0 | | pc1 | | pc2 |
| | | | | |
eth0 |_____| eth0 |_____| eth0 |_____|
| | |
| | |
NET1 | | |
----+------+-----+------+------+------+------+------------
| | |
| | |
| | |
eth0 ,--+--, eth0 ,--+--, eth0 ,--+--,
| | | | | |
| nd0 | | nd1 | | nd2 |
| | | | | |
ath0 |_____| ath0 |_____| ath0 |_____|
: : :
NET2 : : :
..................:............:..............:... ..........

so as an example, at a specific moment if we mapped pc0 to nd0, pc1 to
nd1, than the traffic between pc0 and pc1 should be routed through nd0
and nd1:

pc0 <--> pc0/eth0 <--> nd0/eth0 <--> nd0/ath0 <--> nd1/ath0 <-->
nd1/eth0 <--> pc1/eth0 <--> pc1.

have read some documents and howtos, but I am a bit lost in this case,
PROXY_ARP, IP tunnelling, manage arp table and route tables, or should
actually write a device driver so try to control every thing at every
interface...? really appreciate if somebody can enlighten me, show me a
bit the big picture and principles I should take....

Regards, Paulour

What exactly is it that you are trying to accomplish with such a setup?
It would help if you can explain what is the purpose.

They way I understand your question so far, if you want to route a
signal through a specific box (as a gateway) just modify the routing
table...

Regards,
Tobias Skytte

the idea is to build a more convenient experiment environment for
mobile network, if setup ok then maybe we could wrapper details into a
controlled virtual wireless device for all 'pc's, so certain wireless
applications and protocols could run on these pcs, bind to the virtual
interface, through changing its associated 'node' we could emulate the
mobility of a wireless network. It could be helpful for
experimentations with large mobile networks.

still wondering, is there still other possibilities except control
detailed ARP table and Route table/gateways to control such a routing
requirement, or somehow I should still simplify the archtecture?


Regards, Paulour


(E-Mail Removed) wrote:
> What exactly is it that you are trying to accomplish with such a setup?
> It would help if you can explain what is the purpose.
>
> They way I understand your question so far, if you want to route a
> signal through a specific box (as a gateway) just modify the routing
> table...
>
> Regards,
> Tobias Skytte