[Cu-wireless] Re: Are we making things too hard on ourselves?

[Stephane Alnet]

Hi,

> As a side note to the network junkies, my guess is that we'll
> probably end up running something related to OSPF since it is a
> link state protocol (instead of a distance vector protocol with
> their associated problems).  I have no idea of the consensus
> of the group.

Well, that's the main problem: there's no consensus, IMHO for two reasons:
(a) different people have different ideas of what the objectives are (Sascha only 
wants to connect 4 or 5 points, others just want to be connected, others want the 
bigger picture (a-la "Ubiquitous Internet", for example) but do not necessarily know 
how to get there, etc.);
(b) a discussion on how the "real thing" would work started a few times but didn't 
conclude because of (a). (Maybe also because I'm a pain in the b..t and keep asking 
questions.)

We should have a discussion on what the "real thing" (i.e. "Community network", 
i.e. "more than 20 nodes") should look like. Trying to argue on the price of stuff (over 
the long term) before then is IMHO useless. ("Purchase what you can afford" seems 
the best advice right now. If somebody wants to get the 3Com box let's do it -- that 
will build knowledge on more products; if people want to invest in the geek-cool 
aspect of building a linux-based wireless router from OEM stuff, let's do it too -- still 
builds knowledge.)

> So the real problem with trying to use just a wireless access
> point is that you are most likely _not_ getting a true router.

Agree. It's always the same thing: depends on your requirements. For the size of 
network most people have in mind and what we can expect to be able to do in the 
next 10 months, a dumb box would be sufficient... ;D

If we want to grow further, we need to (a) define what kind of architecture (or set 
thereof) we want to go with, (b) investigate what protocols and product requirements 
this architecture implies, and _then_ (c) select a product or set of products that 
match said requirements. :)

Note that even defining [(a) Architecture] poses some problems: do you want a layer 
2 only approach (doesn't scale past a few dozens nodes), a layer 2 and 3 approach 
(in which case you need a routing protocol, L3 addresses allocation procedures, 
define how you route among multiple providers, and you may need to provide some 
VPN-like environment (think IMC-network)), or a more complex situation (L2-L4? 
L2-L6?)

We already had some discussion on this but we haven't really looked into it yet (of 
course, I've my idea like everybody else :D ). If there's enough interest we should set 
up an "Architecture" group (or whatever you guys want to call a bunch of network 
geeks looking at design, huh) and move on. :)

My 2 eurocents,
Stephane ("It's that french guy again again")