4 ms·

> curious why RSTP is so deserving of a call out I consider RSTP a core foundational technology for FTTH, not some condiment sprinkled on the network afterward

> curious why RSTP is so deserving of a call out I consider RSTP a core foundational technology for FTTH, not some condiment sprinkled on the network afterwards. Plus it is the Ethernet technology I find most typically missing from FTTH network design. Let me explain by way of an example: imagine a typical small town, population less than 50,000, with 2.60 people on average per household => less than 19,231 households. Traditional FTTH networks were modeled much like a POTS network in a star pattern with a dedicated optical fiber to each household from a central office. That then evolved (or should I say got bastardized) to a HFC (hybrid fiber coax) model by the MSO's as they already had a coax local loop going to each house. Clearly any kind of centralized distributed FTTH is doomed to financial ruin. As an alternative, image this: same community. You get the local school district as your first customer (or other geographically dispersed participant), where you deploy a ring of fiber circumnavigating the community, interconnecting every school. As above you over build their fiber, keeping some for yourself. Of course, you will have a peering point (or two) to interconnect 10 gigabit Ethernet with a tier one provider. You now have a ring of optical fibers less than a few miles from every home in town (typically less than a mile). Built on top of RSTP, the ring of fiber gives every node on the network route-diversified redundancy and protection. But not every residential customer is going to sign up with you, so the last thing you want to do is build out to every home. Instead you build out from your ring to those households that do sign up, where every home has a route-diversified optical connection extending the RSTP ring to the customer premise. This topology negates the need for a massive star network emanating from a central office. Literally with a handful of fibers you can provide for an entire community at a standard of engineering and quality of service the MSOs cannot even fathom. Massive reduction in physical plant requirement, massive reduction in capital per customer.

ftth_finland 4y ago
Spanning tree is an exceptionally bad protocol choice for a ring network. It’s not like you want to create a giant layer 2 network anyway. I suggest you look into routed networks and redundant GPON topologies.
- windexh8er 4y ago
  This. I'm not sure why L2 is even being discussed at such a high level. I've worked for Tier 2 carriers and have deployed fiber across multiple states (transport) and have experience dropping out into FTTH solutions. The last thing you want is an unscalable L2 dependent solution that's a royal PITA to debug when things go wrong (and, they will). IMO one of the core components to think about when doing a buildout is your total solution. I've worked with Infinera in the past for both long haul and PON solutions and the important thing is getting the most return on investment out of your deployed glass. DWDM / band splitting is where you really need to engineer up front.
derefr 4y ago
Can you dumb this down for someone who understands Internet networking (incl. internet backbone routing), and vaguely understands shared-medium networking like wi-fi and cellular networks, but who knows next to nothing about coax, and especially nothing about fiber? What does a fiber loop (bus?) let you do hardware-wise, that a star topology wouldn't? Either way, you're running a piece of fiber to the home from somewhere, and connecting it to your network somehow; and I would assume that it's that connection that's inherently expensive, because of the equipment involved. In star topologies, that's a switch, or a hierarchy of switches. For a shared medium, that's a hub—or, I guess, some kind of direct splice, if that's somehow cheaper in TCO. Is the goal of using a loop, to avoid the cost of having neighbourhood-distribution-box optical network switches, instead directly splicing customers onto the ring, making the customer's fiber "part of" the shared medium of the loop, and then allowing customer equipment to directly push packets / transmission cells onto the loop? If so, why does it sound like you're implying that in a traditional FTTH star topology, there wouldn't be per-neighbourhood optical network switches, and instead only one wide-branching layer where all customers' lines terminate directly at one central location? That seems kind of ridiculous. Are optical switches really that expensive, that those are the only two options? Especially when neighbourhood-distribution-box optical switches would be relatively-low fan-in + relatively-low bandwidth, and therefore could be built to use simple low-frequency TDM circuit-switching protocols like coax's ATM? Also, is client-accessible fiber head-end equipment really trustworthy to do TDM (FDM?) onto a shared fiber medium, without ruining it for everyone else? We don't trust client equipment to do that on enterprise Ethernet networks (if we can at-all help it); we use switches instead of hubs, and we even put VLANs in place to further isolate flows in the merged path for QoS purposes. Is the difference just a matter of a fiber provider only certifying specific (presumably robust) equipment; and of that equipment not being coupled to power delivery issues of an arbitrary host device? Is that enough to get you peace-of-mind that your shared-medium network won't be being blasted with noise from new failing head-end equipment every few days?
- geocar 4y ago
  > What does a fiber loop (bus?) let you do hardware-wise, that a star topology wouldn't? Each site has two links, left and right, that have engineered different paths. If your link is to home base, chances are your second link to home base is going to take the same path.

ftth_finland 4y ago

Spanning tree is an exceptionally bad protocol choice for a ring network. It’s not like you want to create a giant layer 2 network anyway. I suggest you look into routed networks and redundant GPON topologies.

windexh8er 4y ago
This. I'm not sure why L2 is even being discussed at such a high level. I've worked for Tier 2 carriers and have deployed fiber across multiple states (transport) and have experience dropping out into FTTH solutions. The last thing you want is an unscalable L2 dependent solution that's a royal PITA to debug when things go wrong (and, they will). IMO one of the core components to think about when doing a buildout is your total solution. I've worked with Infinera in the past for both long haul and PON solutions and the important thing is getting the most return on investment out of your deployed glass. DWDM / band splitting is where you really need to engineer up front.

derefr 4y ago

Can you dumb this down for someone who understands Internet networking (incl. internet backbone routing), and vaguely understands shared-medium networking like wi-fi and cellular networks, but who knows next to nothing about coax, and especially nothing about fiber? What does a fiber loop (bus?) let you do hardware-wise, that a star topology wouldn't? Either way, you're running a piece of fiber to the home from somewhere, and connecting it to your network somehow; and I would assume that it's that connection that's inherently expensive, because of the equipment involved. In star topologies, that's a switch, or a hierarchy of switches. For a shared medium, that's a hub—or, I guess, some kind of direct splice, if that's somehow cheaper in TCO. Is the goal of using a loop, to avoid the cost of having neighbourhood-distribution-box optical network switches, instead directly splicing customers onto the ring, making the customer's fiber "part of" the shared medium of the loop, and then allowing customer equipment to directly push packets / transmission cells onto the loop? If so, why does it sound like you're implying that in a traditional FTTH star topology, there wouldn't be per-neighbourhood optical network switches, and instead only one wide-branching layer where all customers' lines terminate directly at one central location? That seems kind of ridiculous. Are optical switches really that expensive, that those are the only two options? Especially when neighbourhood-distribution-box optical switches would be relatively-low fan-in + relatively-low bandwidth, and therefore could be built to use simple low-frequency TDM circuit-switching protocols like coax's ATM? Also, is client-accessible fiber head-end equipment really trustworthy to do TDM (FDM?) onto a shared fiber medium, without ruining it for everyone else? We don't trust client equipment to do that on enterprise Ethernet networks (if we can at-all help it); we use switches instead of hubs, and we even put VLANs in place to further isolate flows in the merged path for QoS purposes. Is the difference just a matter of a fiber provider only certifying specific (presumably robust) equipment; and of that equipment not being coupled to power delivery issues of an arbitrary host device? Is that enough to get you peace-of-mind that your shared-medium network won't be being blasted with noise from new failing head-end equipment every few days?

geocar 4y ago
> What does a fiber loop (bus?) let you do hardware-wise, that a star topology wouldn't? Each site has two links, left and right, that have engineered different paths. If your link is to home base, chances are your second link to home base is going to take the same path.