At 9/14/2010 02:59 AM, Jeromie Reeves wrote:
>...
> > First-generation "mesh networks" with one radio were awful. I didn't
> > think that was what Greg had in mind. Second-generation mesh
> > networks separated the backhaul ("meshing") from the user access, and
> > worked better. I actually like to use mesh in its original sense, to
> > refer to a network with an arbitrary topology allowing multiple links
> > per node. It contrasts with star and ring. A mesh is thus a
> > redundant network. The Internet is a mesh. SPF is a meshing
> > protocol, as is distance vector.
>
>Wifi is not well suited to the lower layer software. It works, but the
>radios just were not designed for it. I have been doing multi radio
>'mesh' since I started using wifi based gear over canopy.
I agree about WiFi. But I don't think of Ethernet radios as
WiFi. Sure, they use Atheros chips that were designed for WiFi, and
the 11n modulation techniques, but WiFi to me means "LAN", with all
of that ugly contention and hidden transmitter stuff. I just want
cheap radios, with a WAN-oriented contention protocol like Airmax or
an Nstreme. Alas, no standards or interoperabiilty there... at least not yet.
So if I treat the radios as PtP links in the mesh topology and let
the software find a path across multiple hops, it should work. Bits
is bits, after all. I am not doing anything with "broadcast topology
subnetworks", to use the good old ISO term that covered orange-hose
Ethernet, RIP. Even the shared sectors will be functionally PtP,
just muxed onto one radio.
> >
> >>Multiradio relays generally do not benefit from most of the
> >>mesh software.
> >
> > Depends on the software, but you may be right with some of the
> > freeware mesh software designed for home routers running DD-WRT et al.
> >
> >>OSPF or rSTP works very well and are generally stable
> >>and are generally supported and have many tools.
> >
> > OSPF operates in IP. I'm trying to do everything at a lower layer
> > than that.
>
>What layer are you doing your routing at?
The mesh will run at "layer 2", meaning a layer below IP. Here's a
place where the wireless vendors are a few years behind the fiber
vendors. In the fiber world, there's a lot of SONET around, great
old TDM technology, but for non-TDM applications the hot thing is
Carrier Ethernet, as defined by the Metro Ethernet Forum. This
basically shares one thing with the real Ethernet, the name! Carrier
Ethernet is essentially Frame Relay at Ethernet speeds, using
Ethernet frameing. It uses the VLAN tags, not MACs, and while there
is a LAN Emulation PtMP mode, and it can use RSTP, it is switching,
not bridging. This isn't rocket science and I'd love to see it
cleanly implemented in RouterOS (Level 4 and up). MAC bridging has
no place outside of the pure LAN.
> >RSTP works but is still fundamentally stupid; spanning
> > tree just avoids loops, but doesn't optimize.
>
>I did not say it did. There are places to use it, like when you have a
>end point that can see 2 AP's and failover as needed. If those AP's
>are only connected to a switch rstp can be used to disable a port (ill
>need to find the script, it was not pretty nor was used long)
Yes, for a very simple case, it's fine. The network I'm designing
needs a few dozen nodes; the paths are "opportunistic" (i.e., too
many hills and trees).
> >When the "mesh" gets
> > complex, you want real routing capabilities. I just have reasons to
> > avoid doing this in IP.
>
>Why do you want to avoid the IP layer?
Not all traffic needs to be IP. I don't particularly like IP. It
was great fun in the 1980s but I don't use MS-DOS any more
either. Routing is impacted by IP's flaws, so I'd rather minimize IP
routing and treat it as an application-support layer, as payload. A
replacement for IP is one of my ongoing projects. (And I just say no
to IPv6 -- tastes worse, more filling!)
>The radios while can pass what
>ever traffic, are inherently IP devices
Not really. They are layer 2 frame relaying devices. They can have
IP routing turned on in software, but it's optional.
>and switch slowly enough that
>speed does not seam like the driving factor.
>
> > But "bridging" is just a bad LAN hack too;
> > it doesn't scale to the radio world. I remember when we introduced
> > Ethernet bridges at DEC, somewhere around 1985. It sounded like a
> > good idea at the time. But a few years later, the head of network
> > archicture quipped, "Networks are a drug. Bridged networks are a
> > dangerous drug." Of course that was shortly before some
> > Microsoft-worshipping IT types started pushing big bridged multi-site
> > LAN Manager networks...
>
>At what lower layer? Unless you are changing out the protocol on the
>wireless layer its still wifi
The physical layer shares modulation with WiFi but those Atheros
chips are really just relaying frames; as noted above WiFi is the
local area application, while UBNT et al stretch them into being
wide-area Ethernet relays.
>and still has the limits of wifi.
>NStream might work better for this.
Yes. I may end up with a mix of Nstreme and Airmax (different links,
not talking to each other!), which is sort of ugly but the nice
Rocket radios don't do Nstreme and there's a limit to how many mPCI
cards I can fit into a Routerboard... I wish they still made the RB
600a! I don't trust the RB 800; it has a fan (moving part, a real
maintenance problem when outdoors up a tower or pole!).
>Bridging is a tool, like all
>tools when used poorly, is called a poor tool.
Agreed. It's suitable to small-scale applications.
> >
> >>I keep everything off
> >>the radio I possibly can and use them as bridges with a router behind
> >>them. With Mt having $40 units that is even easier and faster then
> >>ever.
> >
> > I agree that this is a good idea. I am likely to do some radios as
> > cards plugged into the RB, like the SR71-15s which look really
> > good. But the access points and some links will be separate, on
> > Ethernet. Besides, the Ethernet RocketM sectors look better than
> > anything MT has. The specs are pretty close but the scuttlebutt on
> > the boards is that the UBNT cards outperform the R52HNs.
>
>I have completely dropped all wireless that is not Ubnt based (well
>other then the gear I have not had time to swap and the indoor ap's).
There's one reason why I'm thinking of using R52HNs. The UBNT radios
simply will not make use of the 5825-5850 spectrum. I guess it's not
available in Latvia or something... So a few MT PtP links can go up
there, and thus leave the rest of the band free for UBNT
radios. Yeah, some RF Armor and premium pigtails may be necessary.
>A site can
>now have up to 6 AP's between 2.4 and 5ghz and link to a number of
>other sites. OSPF works well here. Self interference is still the
>biggest issue. I have a site 15 miles out, with a ap pointed 70* off.
>Using the Ubnt specan, I can see a few db noise rise (goes away when I
>have that ap powered down). All of my R52's (not Ns) died or went into
>a super spew mode with lots of noise.
Are you using any RF Armor or other shields? What sector antennas
seem to be the least leaky? Thanks....
--
Fred Goldstein k1io fgoldstein "at" ionary.com
ionary Consulting http://www.ionary.com/
+1 617 795 2701
--------------------------------------------------------------------------------
WISPA Wants You! Join today!
http://signup.wispa.org/
--------------------------------------------------------------------------------
WISPA Wireless List: [email protected]
Subscribe/Unsubscribe:
http://lists.wispa.org/mailman/listinfo/wireless
Archives: http://lists.wispa.org/pipermail/wireless/
