I am in the process of designing a wireless infrastructure consisting of BH gear to relpace existing ISP leased line infrastructure as well as to deploy AP clusters at each POP/tower for customer access. The design calls for a ring-like infrastructure similar to this:
- NOC -
||
Tower A -> Tower B -> Tower C -> Tower D -> Tower A
I am looking for suggestions for best practices in moving the data around. The system will need to scale very well and data flow control is very important. These requirements lead me towards using a 3 legged router at each POP rather than switching a flat Ethernet using spanning tree or RSTP.
OSPF seems to be what I’m after. Does anyone here do this? If so, what are you using? Cisco? I’m leaning towards using FreeBSD and Zebra. I would certainly like to hear suggestions and real world reports from others in a similar environment.
Thanks!
Landon
The routers and ring structure will add latency and limit throughput to Tower C and its intermediate tower. Do you need the redundancy? Would you consider a star network with Tower A as the hub? Does Tower A have RF LOS to Tower C? When you say the system needs to “scale”, do you mean adding towers or aggregate throughput?
Redundancy is not required but is certainly desired considering the network will carry more than SM customer data. It also will carry some dialup traffic, DSL, and server data from some of the existing POP’s with leased lines.
When I say it needs to scale well, I mean that in terms of both towers and aggregate bandwidth. The ring I have described is basically going to be a backbone. Tower A does not have RF LOS to Tower C. However, most of the areas where we need to add POP’s have RF LOS to one of the towers in the ring.
I should also mention that we will likely multihome with our Internet provider by adding another connection to that provider at either Tower C or a remote location from Tower C. This could reduce the number of hops, decrease overall latency seen by the customer, and again, address our redundancy goals.
You mentioned added latency with the introduction of routers at each POP. Is it not better to segment the network into individual broadcast domains when such a wide scale deployment is necessary?
What affect will this have on timing in the BH’s?
Thanks!
Latency:
Motorola has worked to make the latency on a Canopy link very low; the Advantage models are within an order of magnitude of an Ethernet switch. Routers, on the other hand, are typically 1-2 orders of magnitude higher. You and your customers will both benefit if you keep the number of hops to a minimum; creating a ring increases the hops.
Throughput:
I’d also be concerned with a ring’s demands on throughput. In your example, the A-to-B link would carry the traffic for both B and C. You may, however, have no choice when C cannot link directly to A. Installing a separate landline Internet circuit to C, as you suggest, would eliminate the problem.
Broadcast Domains:
With a large number of IP hosts, you certainly want to control broadcasts over the Canopy links. But this does not mean that the network backbone, built with Canopy, can’t be flat. I’m only questioning the need for routers within a multi-hopped backbone.
Scalability:
A star network, if it’s possible in your situation, would be more scalable than a ring. It’s easier to add towers – if those towers can link directly to the primary tower – and every added backhaul link adds aggregate throughput. The limitation would be the number of independent Canopy links you could install on a single tower. An AP or AP cluster at the primary tower for the backbone would permit a much larger number of secondary towers, but would offer the lowest aggregate throughput.
Redundancy:
The downside to a star, of course, is when a link fails all customers served by that link are down.
Timing:
Synchronization requirements on the backhaul links aren’t affected by your choice of star or ring, switched or routed.