Tropos: Aggregating Capacity
August 05, 2004
The mesh networking company answers questions regarding how it avoids degradation of throughput on it's metro-scale wireless networks.
Q: Some have said that large-scale outdoor Wi-Fi networks will run into quality-of-service trouble as connectivity breaks down over multiple hops within a mesh. How are you getting around that problem?
A: We say the idea that throughput in a mesh networks [degrades] with the number of hops is just incorrect. It does roll off, but not with the kind of exponential decay that some folks are claiming. In any case, we look at it in a very different way. Our idea is to have a routing algorithm which, rather than trying to minimize the number of hops, is instead aimed at maximizing end-to-end throughput.
Q: So, rather than minimizing hops and risking hitting a bad link, your algorithm may take the long way around, but it will only move through good links?
Q: Even so, there have been claims that big outdoor networks will be fundamentally flawed since they invite too many users to share too little bandwidth.
A: It can be true, but the way you eliminate it is by having an architecture that allows you to add more aggregate capacity as you add more backhaul points. You've got the mesh network and then you have one or more connections from the mesh network to the wired Internet. The way our architecture works is that one connection to the Internet gives you an aggregate of 11 megabits of throughput. Now if you add a second connection to the Internet, some cells cluster around the first connection, while others cluster around the second connection. Now we have gone from an aggregate of 11 megabits of bandwidth to an aggregate of 22 megabits. So the mesh networks can be made to be very scalable if you have the ability to add more capacity to the network as you add more Internet connections.
Q: If you think these likely problems can be solved, what other technical hurdles remain for big outdoor mesh networks?
A: The challenges we have run into are, number one, finding enough suitable locations to bring backhaul to the network.
Q: For example?
A: Let's say I have a network that covers 10 square miles and has 100 nodes in it. Say I need to bring a wired Internet connection to one out of every 10 of those nodes. That means I need to find 10 places within those 10 square miles where I can drop a backhaul connection into that system.
Q: Why is that hard?
A: City fiber rings tend to go where they go, and if you want them to go somewhere else there is trenching and things like that involve. Putting out T1s can be expensive both on a provisioning and a monthly basis.
Q: How have you addressed this?
A: We have ended up using a lot of point to multi-point solutions, which have in a lot of cases has ended up being the least costly way to do it. But it is really something you have to take care of in the planning stage. When you are designing these large networks, you have to really think hard about the backhaul availability that you have and how you going to get that distributed around that network.
Q: Given all these factors, what does it cost to deploy a big outdoor mesh?
A: A comprehensive deployment over 10 square miles will cost about $300,000 with operating expenses -- that can be less than $10 per month per subscriber.
Q: Can a service provider turn a profit here? What economics do you envision driving Wi-Fi mesh deployments in the future?
A: In Chaska, Minn., they have done the entire 16 square miles of their town, and the city will do better than break even selling broadband residential service at $16 a month. We have seen similar experiences around the country. The hotspot model is like the old payphone model: you have to go to the place where the service is available, rather than having that service available wherever you go. People want just as much mobility with their data as they get with their voice. So we believe this is going to be a profitable venture for service providers, given the demand we see among consumers and among governments, coupled with the low cost of deployment and ongoing operation.