GPON has been the technology of choice for well over a decade. The GPON technology delivers a download path of 2.4-gigabits bandwidth to each neighborhood PON. Most of my clients have deployed GPON in groups of up to 32 customers in a neighborhood PON. In practical deployment, most of them pack a few less than 32 onto the typical GPON card.
I’m curious about how ISPs will deploy XGS-PON. From a pure math perspective, an XGS-PON network delivers four times as much bandwidth to each neighborhood than GPON. An ISP could maintain the same level of service to customers as GPON by packing 128 customers onto each GPON card. But network engineering is never that nicely linear, and there are a number of factors to consider when designing a new network.
All ISPs rely on oversubscription when deciding the amount of bandwidth needed for a given portion of a network. Oversubscription is shorthand for taking advantage of the phenomenon that customers in a given neighborhood rarely all use the bandwidth they’ve been assigned, and never all use it at the same time. Oversubscription allows an ISP to feel safe in selling gigabit broadband to 32 customers in a GPON network and knowing that collectively they will not ask to use more than 2.4 gigabits at the same time. For a more detailed description of oversubscription, see this earlier blog. There are ISPs today that put 64 customers or more on a GPON card – the current capacity is up to 128 customers. ISPs understand that putting too many customers on a PON card will start to emulate the poor behavior we see in cable company networks that sometimes bog down at busy times.
Most GPON networks today are not overstressed. Most of my clients tell me that they can comfortably fit 32 customers onto a GPON card and only rarely see a neighborhood maxed out in bandwidth. But ISPs do sometimes see a PON that gets overstretched if there are more than a few heavy users in the same PON. The easiest solution to that issue today is to reduce the number of customers in a busy PON – such as splitting into two 16-customer PONs. This isn’t an expensive issue because over-busy PONs are still a rarity.
ISPs understand that year after year that customers are using more bandwidth and engaging in more data-intensive tasks. Certainly, a PON with half a dozen people now working from home is a lot busier than it was before the pandemic. It might be years before a lot of neighborhood PONs get overstressed, but eventually, the growth in bandwidth demand will catch up to the GPON capacity. As a reminder, the PON engineering decision is based on the amount of demand at the busiest times of the day. That busy hour level of traffic is not growing as quickly as the overall level of bandwidth used by homes – which more than doubled in just the last three years.
There are other considerations in designing XGS-PON. Today, the worst that can happen with a PON failure is for 32 customers to lose bandwidth if a PON card fails. It feels riskier from a business perspective to have 128 customers sharing a PON card – that’s a much more significant network outage.
There is no magic metric for an ISP to use. You can’t fully trust vendors because they are going to sell more PON cards if an ISP were to be extremely conservative and put only 32 customers on a 10-gigabit PON. The ISP owners might not feel comfortable leaping to 128 or more customers on a PON. There are worse decisions to have to make because almost any configuration of PON oversubscription will work on a 10-gigabit network. The right solution will balance the need to make sure that customers get the bandwidth they request without being so conservative that the PON cards are massively underutilized. Over time, ISPs will develop internal metrics that work with their service philosophy and the demands of their customer base.