[bcrl]

80% loading of a link is where queuing theory says latency goes up fast. With 10Gbps links (be they 10Gbps ethernet links or XGSPON), you don't really want to go up above 80% load. 10Gbps PON is becoming mainstream now, while 25Gbps PON is still only in early testing at this point.

[sschueller]

I don't understand why one would run PON when you might as well run the fiber all the way directly to the customer without any splitting and only be limited by the SFP modules attached to the end. Unless the last piece is still copper it makes no sense. Wouldn't a splitter also cost more?

I have 4 fiber strands directly run to my municipal power company and so does everyone else in my city. I can do as much speed as a provider is willing to offer which currently is up to 25gbit. [1]

[1] https://sschueller.github.io/posts/wiring-a-home-with-fiber/

[bcrl]

PON cuts the number of feeder fibres by orders of magnitude. GPON is typically deployed with 1x32 PLC splitters. With XG/XGS-PON deployments with 1x64 and 1x128 splitters are feasible. If you have infinite financing, sure, more fibres are better, but in the real world there are significant cost increases with higher fibre counts that not all providers can absorb. Municipal entities can easily get 30 year amortization, which is definitely not the case for most startup ISPs in unserved markets.

[sschueller]

Thanks for the info. I would think Google would be the one that could make such an expense.