I did quite a lot of research into MeshAP a while ago and reached the conclusion that Mesh in general is distinctly non-trivial.
Yes.
One of the hardest problems to solve is IP allocation and addressing.
Then punt the problem elsewhere. Do all your routing within the mesh at the MAC layer. (I know, bridging is a bad thing but you can call it MAC-layer switching if it makes you feel better.) That allows a single IP address to roam throughout a mesh. Since it will probably be a single administrative domain with a single point of egress, no worries. DHCP assigns you an address within the mesh and you can wander around at will, at least until you reach the edge of the mesh.
There is a scaling issue here. Since a MAC address is an identifier and not a locator, your routing table needs to be as large as the number of devices that are in your network. The upside is that, with the advent of fast processors and cheap memory you can have a forwarding table that supports many thousands of entries. If you manage the diameter of your mesh properly, there will be no problem. OTOH, I hate systems that have arbitrary limits and I have experienced what happens to a bridged network when you exceed the size of the forwarding table.
Want to roam farther afield? Use Mobile-IP and now your long-lived sessions can survive roaming to another mesh. Yes you get a new locator IP but your identifier IP remains the same no matter where you go.
As Broadband usage grows, I'm reaching the conclusion that what we should be aiming for is ubiquitous internet access and graceful
handoff between APs, rather than specifically mesh. If most APs have
a broadband connection then there's few benefits from meshing them
together. One interesting one is aggregating the bandwidth together
for burstable higher speed, but that's also hard to arrange.
But you need to know when you roam out of your IP domain. If you have transparent handoffs at the MAC layer there is no event that tells the IP layer it needs to negotiate a new IP address lease. A 'MAC down' followed by a 'MAC up' event tells IP to negotiate a new lease.
Having said all that Mesh still looks good for remote areas where internet access is expensive. If you're getting one high speed line
in and then sharing it round the community, mesh works pretty well.
But in that scenario, the difference between a $100 hacked WRT and a
$400 dedicated mesh box is less important as you really need the
extra control in the dedicated design.
My hope for the WRT hacking is to come up with something that makes it easy, cheap and safe to share your bandwidth. But even this may become irrelevant as the "Vendor Defaults Community Network" is everywhere. ;-)
Oh, I think it deserves some real analysis. A wireless host should be able to wander at will and retain its long-lived sessions. (This is where the session layer of the ISORM, something that is missing from the Internet 5-layer model, is useful.) Heck, when you reach somewhere with more bandwidth you should be able to just plug into that Gig-E jack over there and keep going. And while you are at it, when you are out of range of your 802.11 connection and all you have is 2.5G PCS service, you should still be able to keep going, albeit with less bandwidth. This is the universal solution. Everything else is a subset of this problem.
-- Brian Lloyd 6501 Red Hook Plaza [EMAIL PROTECTED] Suite 201 http://www.lloyd.com St. Thomas, VI 00802 +1.340.998.9447 (voice) +1.360.838.9669 (fax)
-- general wireless list, a bawug thing <http://www.bawug.org/> [un]subscribe: http://lists.bawug.org/mailman/listinfo/wireless
