Do mesh networks scale? This is The Big question. The article linked covers recent articles on a few other websites, including a CTO of a wireless mesh networking company who says they do not scale, and two rebuttals to that view.
Until we have some large scale deployments there are going to be debates. I am still holding the view that there are information-theoretic proofs that mesh networks can scale (but since the math required is way over my head this is still just a belief on my part, despite the "proof" part.)
I could generate examples in which a mesh would be challenged. For example, high traffic between many pairs of nodes which are all distant (in number of node hops).
The Internet, to my understanding, is more of a tree-type infrastructure. There are some mightly fat pipes at the core to handle high traffic between distant connections.
Certainly there are types of mesh networks that do not scale. 802.11 protocols have a mesh mode (seldom utilized) that does not scale (I mean, no one argues that it does scale.)
The Internet is a tree-type network (or at least partly a tree-type network) which definitely seems to work, but also provides the entire system with several points of failure (root DNS servers for sure, but also places like MAE-East and MAE-West here in the U.S.)
David Reed appears to be at the forefront of efforts to convince people that ad hoc mesh networks can scale (even though it sounds counter intuitive.) My lay understanding of the argument is that there is no such thing as interference, only poor listeners, and advances in software radios and digital signal processing take care of the interference straw man. Reed supposedly concludes (I can't even find the canonical paper where this conclusion is made - but it's at least widely alleged that Reed concludes) that available mesh network bandwidth actually increases (!) as the number of nodes increases.
To use your example of a network situation that does not scale, I guess the idea is that increasing the total number of nodes in the mesh increases the number of available routes to carry the high traffic between nodes which are distant. The key is that every node is a sender and receiver (like on the Internet,) and also a router. So every time you add an endpoint to the network (a client) you are also adding infrastructure and thus increasing capacity.
Detractors say that this doesn't really work because beyond a certain size it is difficult for each node to obtain proper routing information, and the network becomes increasingly saturated with traffic whose sole purpose is to get this information (I guess something like every node pinging every other node trying to figure out who it's neighbors are, and what transmission routes are possible.)
The pro side then says that all you need to do is [insert some very difficult topological math hand waving here] and it will work out. (Oh, and build "smart" antennas, and actively modulate power output at all nodes, and invent these super "smart" software radios, etc...)
In any case, I don't see mesh networks as replacing the Internet backbone which is obviously very good at what it does. But mesh networks would be a very desirable addition to our picture of what the Internet actually is. I want local (where this could mean city wide in dense urban areas) ad hoc mesh networks that can but do not have to connect to the standard global (and more tree like) Internet. It's definitely not an Either / Or.
But the fun gadgets of the future want to be part of the mesh!
Okay, if I haven't bored you to death yet, here's a .pdf of a David Reed powerpoint presentation: How wireless networks scale: the illusion of spectrum scarcity.
Okay, if each node is very sophisticated, powerful, yada, yada, then life is good. My thinking of meshes is based on cheap, simple nodes, which will have limited capacity in terms of: number of connections, total thoughput, rf performance, etc.
Right. I know. It's hard for me to tell if the capabilities required are feasible or not. I guess we'll see.
|
Until we have some large scale deployments there are going to be debates. I am still holding the view that there are information-theoretic proofs that mesh networks can scale (but since the math required is way over my head this is still just a belief on my part, despite the "proof" part.)
- jim 7-09-2004 8:05 pm
I could generate examples in which a mesh would be challenged. For example, high traffic between many pairs of nodes which are all distant (in number of node hops).
The Internet, to my understanding, is more of a tree-type infrastructure. There are some mightly fat pipes at the core to handle high traffic between distant connections.
- mark 7-09-2004 8:54 pm
Certainly there are types of mesh networks that do not scale. 802.11 protocols have a mesh mode (seldom utilized) that does not scale (I mean, no one argues that it does scale.)
The Internet is a tree-type network (or at least partly a tree-type network) which definitely seems to work, but also provides the entire system with several points of failure (root DNS servers for sure, but also places like MAE-East and MAE-West here in the U.S.)
David Reed appears to be at the forefront of efforts to convince people that ad hoc mesh networks can scale (even though it sounds counter intuitive.) My lay understanding of the argument is that there is no such thing as interference, only poor listeners, and advances in software radios and digital signal processing take care of the interference straw man. Reed supposedly concludes (I can't even find the canonical paper where this conclusion is made - but it's at least widely alleged that Reed concludes) that available mesh network bandwidth actually increases (!) as the number of nodes increases.
To use your example of a network situation that does not scale, I guess the idea is that increasing the total number of nodes in the mesh increases the number of available routes to carry the high traffic between nodes which are distant. The key is that every node is a sender and receiver (like on the Internet,) and also a router. So every time you add an endpoint to the network (a client) you are also adding infrastructure and thus increasing capacity.
Detractors say that this doesn't really work because beyond a certain size it is difficult for each node to obtain proper routing information, and the network becomes increasingly saturated with traffic whose sole purpose is to get this information (I guess something like every node pinging every other node trying to figure out who it's neighbors are, and what transmission routes are possible.)
The pro side then says that all you need to do is [insert some very difficult topological math hand waving here] and it will work out. (Oh, and build "smart" antennas, and actively modulate power output at all nodes, and invent these super "smart" software radios, etc...)
In any case, I don't see mesh networks as replacing the Internet backbone which is obviously very good at what it does. But mesh networks would be a very desirable addition to our picture of what the Internet actually is. I want local (where this could mean city wide in dense urban areas) ad hoc mesh networks that can but do not have to connect to the standard global (and more tree like) Internet. It's definitely not an Either / Or.
But the fun gadgets of the future want to be part of the mesh!
- jim 7-09-2004 9:28 pm
Okay, if I haven't bored you to death yet, here's a .pdf of a David Reed powerpoint presentation: How wireless networks scale: the illusion of spectrum scarcity.
- jim 7-09-2004 9:31 pm
Okay, if each node is very sophisticated, powerful, yada, yada, then life is good. My thinking of meshes is based on cheap, simple nodes, which will have limited capacity in terms of: number of connections, total thoughput, rf performance, etc.
- mark 7-10-2004 12:36 am
Right. I know. It's hard for me to tell if the capabilities required are feasible or not. I guess we'll see.
- jim 7-10-2004 12:45 am