[Battlemesh] WBMv6 test discussion [was: Re: Recon Team - Follow Up]

[Simon Wunderlich]

Hey Frank,

as promised, I'm CCing the battlemesh mailing list here so they
can join the test discussion.

For the battlemesh mailing list: Prof. Frank Fitzek, Martin Hundeboll
and the other people from their research team will host the WBMv6 in
Aalborg University in April 2013. We have started discussing test setups,
but IMHO all battlemeshers, especially the guys doing the tests the
last years should be able to join the discussion. So I'm moving the
discussion to the ml.

On Mon, Oct 22, 2012 at 09:27:50PM +0200, Frank Fitzek wrote:
> Dear all,
> 
> I am happy there is interest in testing at all.
> 
> Benchmarking is the core of engineering. We all like to invent and
> present the solutions found but then we have to compare the
> solutions found. In research I only see research papers that show
> that the own approach is the best. No wonder as they tune the
> environment as long as they find a spot where they are better. And
> of course if you go for testing everybody wants to tune the testbed
> in favor of their own approach.
> 
> As we want to find the best solution for the real world, it would be
> great to understand how does the real world look like? In case of
> wireless mesh I would like to understand what kind of topologies we
> would like to support. Just one source and once destination
> communicating over one relay, or a line of relays, or an arbitrary
> network? Maybe we can get some topologies from open mesh. They have
> to be some data for that. If not we go for a set of networks that
> are likely to exist.

Networks in "reality" often look very messy. Even in hotel networks
where you have some lines of APs per floor, APs from different floors
may (or may not) be able to connect to each other. There are buildings
with a lot of LAN uplinks (they could be used as "gateways") and one 
final gateway to the internet. Or there might be only a single DSL
router in it.

One topology we can find often is "many routers in one spot", or most
of them can connect to each other - with a more or less good quality.

The "hotel" topology with a few good-connected rows, which are not
(or not very good) connected with each other would be possible too.

There are also different types of devices: multi-radio devices and
single-radio devices. I'd suggest to stick with single-radio devices
for now (this is what we tested the last years too).

Some tests have been defined for WBMv5 here:
http://battlemesh.org/BattleMeshV5/Tests

> 
> The next is the traffic itself, unidirectional, bidirectional,
> constant bit rate, variable bit rate, etc

typical traffic patterns might be:
 * "bulk"/"download": unidirectional stream, as fast as it goes
 * "VoIP" - bidirectional stream, 64kbit/s each, small packets
 * mp3 radio stream, youtube video stream? like download above or
   constant rate
 * broadcast traffic from Windows/Avahi/Dropbox/whatever - maybe only
   relevant for Layer2 protocols

> 
> Also the node mobility is important, but this should be the last
> thing we touch.

agreed. :)

> 
> But in all test events or InterOps, you will find very simple setups
> for the start, moving up for more complex ones.
> 
> Out of the blue i would go for two setups
> 
> 1.) A line of nodes (as in a hotel, one bridge to the Internet and
> many relays) with more or less unicast traffic (towards the bridge).

Sounds good, but I think the traffic is going from the bridge to
the individual nodes - most people will probably download, not upload.

> 
> 2.) Network of several nodes 20 or even 50 where each node may
> select another node randomly and send data for a certain time.
> Randomness takes away the advantage for any technique. Be prepared
> to be unprepared.

I don't know how many nodes we can get - Bastian told me he could
provide some routers, although I don't know how many. I think 30-40
would be a good number, if possible.
> 
> If we go for 2.) we would need to ask our technical team to prepare
> battery packs to got outside ... or low power inside but we need a
> hall.

Why can't we do it inside? We did some battery tests the last years too,
but batteries tend to be empty, and then it becomes dark, and then we
can't find out routers anymore. :)

The atrium [1] looks like a good place for that? Can we use this part of the
building, maybe also hallways etc?

Doing it inside might be more convenient to stabilize/fix the setup for the
test.

[1] http://battlemesh.org/BattleMeshV6?action=AttachFile&do=get&target=IMAG1361.jpg

> 
> I think it is good we started this discussion, so hopping for your
> comments and ideas.
> 

I have never prepared the tests, so I'm sure others can comment much more
than I can. :)

Cheers,
	Simon
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 198 bytes
Desc: Digital signature
URL: <http://ml.ninux.org/pipermail/battlemesh/attachments/20121024/5b0c0b88/attachment-0001.sig>