Jack Unger wrote:
You raise some good points... and here are some more differences
between Matt's fully-meshed WIRED network example and the real-world
conditions under which WIRELESS mesh networks are so often deployed
My example actually used wireless P2P links, although it was meant to
apply just as well to other mediums.
1) REROUTING - Only a node failure or a high peak traffic load would
normally force a routing path change on a fiber/copper network. On a
wireless mesh, routing path changes will also result from interference
caused by other same-network nodes, interference from other networks,
and interference from other wireless non-network sources. Routing path
changes will also be caused by the movement of obstructions and other
rf-reflective objects such as trees and vehicles.
Rerouting occurs regularly with a mesh wired network due to load
balancing and QoS concerns. Although, I would agree with you that in a
more traditional ring topology, rerouting would only occur on a node
failure or overload with a wired network. I would also like to point out
that someone of the issues you raise generally only occur on
street-level networks. Using a wireless mesh network on rooftops or
towers avoids many of the issues you raised. At the same time, many mesh
vendors now use layer2 metrics such as rssi, signal to noise ratio, and
RF frame errors in addition to layer3 metrics to select the best path. I
believe these layer2 metrics are required for a proper street-level network.
2. CAPACITY - Fiber/copper networks typically start out with
high-capacity (compared to wireless) full-duplex links. Wireless mesh
networks start out with low-capacity half-duplex links.
What capacity any network starts with is up to folks deploying it and is
not a function of medium.
3. CONNECTIVITY - Fiber/copper mesh network nodes have two or more
paths to other nodes. "Real-world" wireless mesh networks may contain
nodes that, in some cases (the traditional "mesh" definition not
withstanding) only have a path to one other node. For example,
obstructions may block paths to all but one (or even no) other nodes.
I believe almost any real-world network is going to have paths that
aren't protected. For example, we have a building that served by a
single fiber path because it is not economically to have diversity at
that particular building. Every other building we have on fiber is
connected via diverse paths.
4. ENGINEERING - Fiber/copper mesh networks are typically properly
engineered for traffic-carrying capacity, QoS, latency, etc.
"Real-world" wireless mesh networks are typically deployed in
near-total ignorance of the Layer 1 (wireless layer) conditions.
That's the great attraction (IMHO) of muni-mesh networking today.
These networks are thrown up in the belief that they don't need any
Layer 1 design or engineering expertise and that this will allow for
quick, widespread deployment. Last time I looked however, there was
still "no free lunch". I predict that the muni
mesh networks that are "thrown up" today (Philadelphia will be a prime
example, unless it's re-engineered correctly) will fail and fail
miserably to meet the high expectations that have been raised like
free or low-cost broadband for all. In addition, muni mesh networks
today typically lack adequate traffic engineering and performance
testing under load.
I believe you are entirely correct. However, that doesn't mean that a
WISP can't properly engineer a mesh network. I would suggest the above
is no different than folks stringing a bunch of Ethernet hubs together
and expecting their LAN to work correctly.
I'm not saying that wireless mesh networks should never be used. There
are certain (obstructed, short-link, low capacity) environments where
they will be the best, most economical solution. I'm just saying that
the false claims and marketing hype surrounding MOST (and let me
repeat, MOST) of today's mesh networking claims, particularly mesh
network nodes that contain just a single 2.4 GHz radio are going to
come back to bite both the vendors and the cities that deploy these
networks without sufficient wireless knowledge in the false belief
that wireless mesh networks are just "plug-and-play".
Interestingly, Tropos the current poster child of wireless mesh networks
is a single radio product, but all of their large networks have nodes
backhauled by fixed wireless. In a sense, this means those networks are
not based on a single radio. In fact, some market argue that using
Canopy for backhauling Tropos --as is the case of Philadelphia-- is a
better solution than to use an 802.11-based radio for backhaul like the
multi-radio mesh network vendors do.
WISPA Wireless List: email@example.com