A GPON OLT 1 port unit is not a 64-subscriber device: current single-port boxes register up to 128 ONUs on that one PON port, so the real planning question has moved from "how many subscribers fit" to "do the optical budget and the uplink survive all 128".
Rayin (Shenzhen Rayin Technology) is a manufacturer of GPON OLTs and industrial Ethernet switches.
One PON port registers up to 128 ONUs on current equipment. ITU-T G.984.1 supports split ratios up to 1:128 with 60 km logical reach and a 20 km maximum differential distance; the 128 figure on a datasheet is a vendor limit that has converged across the market, not what the protocol itself allows.
Once registration stops being the constraint, two others take over: optics and bandwidth. A 1:128 splitter costs about 24–25 dB of industry-typical insertion loss, so it normally needs a Class C+ (32 dB) or higher OLT module — the ONU side stays Class B+ at 28 dB with roughly −27 dBm sensitivity.
The shared pipe is 2.488 Gbps down and 1.244 Gbps up. Split 32 ways that is about 78 Mbps down and 39 Mbps up per subscriber at full concurrency; 64 ways gives 39 and 19 Mbps; all 128 ways gives about 19 Mbps down and 10 Mbps up, before framing overhead is removed.
One PON port is one PON tree. The port drives a single fiber into a passive splitter cascade, and every ONU hanging off that cascade shares the same downstream broadcast and the same pool of upstream timeslots. No configuration option moves that boundary — adding subscribers means adding ONUs to the same tree, until either the loss budget or throughput, not port count, says stop.
That shared-medium behaviour is why the port needs real intelligence. Per ITU-T G.984.3, the OLT measures the round-trip delay to each ONU during ranging and assigns an equalization delay (EqD) so that bursts from a subscriber 200 m away and one 18 km away still land inside their own grant window. It grants upstream transmission windows every 125 µs frame, maps subscriber traffic into GEM ports and T-CONTs per G.984.4, encrypts the downstream with AES-128, and manages the ONUs over OMCI. A mini OLT runs exactly that stack; it just runs one instance of it.
What the PON port is not is an uplink. Uplink interfaces — copper RJ45 or SFP — carry aggregated traffic from the whole tree toward the router, the BRAS or the aggregation switch. Count them separately, because on small units they are usually the first thing to saturate.
The old planning habit — cap a port at 64 subscribers — comes from a Class B+ optical budget, not from the OLT hardware. Three separate ceilings decide the answer, and only the first one is about registration.
| Ceiling | Typical value today | What changes it | What it does not fix |
|---|---|---|---|
| Registered ONUs per PON port | 128 on current small-form-factor units | Nothing — this is a vendor and software limit, not a law of physics | Contention between subscribers, or a saturated uplink |
| Optical loss budget | Class B+ 28 dB vs Class C+ 32 dB on the OLT side; ONU stays Class B+ | A higher-class OLT PON module, shorter paths, fewer connectors and splices | A single-port unit serving two separate distribution areas |
| Usable per-subscriber bandwidth | About 19 Mbps down at full 1:128 concurrency | Statistical multiplexing: real traffic is bursty, not simultaneous | Peak-evening scenarios where everybody really is active |
Read that first row carefully, because it is where most specifications get misquoted. The ONU addressing space in ITU-T G.984 is eight bits wide, so the protocol has room for far more than 128 identities. What stops you is the vendor limit baked into the PON MAC and the software license of the box in front of you — and across current single-port and compact chassis units, that number has settled at 128 ONUs per PON port. Confirm the figure for your exact unit and software release, because that market-wide convergence is not a guarantee on an older model.
That means the honest planning order has changed. You no longer size around an imaginary 64-subscriber wall; you size around how much loss the tree can absorb and how much bandwidth you want to hand each subscriber.
Do the division once and the rest of the design review gets easy.
| Split ratio | Subscribers | Downstream per subscriber, all active | Upstream per subscriber, all active | Where it is used |
|---|---|---|---|---|
| 1:8 | 8 | 311 Mbps | 156 Mbps | Business parks, high-tier packages |
| 1:16 | 16 | 156 Mbps | 78 Mbps | Dense apartment buildings |
| 1:32 | 32 | 78 Mbps | 39 Mbps | Default FTTH planning target |
| 1:64 | 64 | 39 Mbps | 19 Mbps | Mid tiers, well-inside-the-budget designs |
| 1:128 | 128 | 19 Mbps | 10 Mbps | Entry tiers, low take-up areas, budget-constrained rollouts |
Those figures are the worst case: 2.488 Gbps and 1.244 Gbps divided by the number of subscribers, with every subscriber pulling at the same instant. Real traffic is bursty, so a 128-subscriber tree where 15% of users are active at any moment serves each active session far more than 19 Mbps. But the worst case is the number you can defend in an SLA, and it is the number that decides whether the design survives Christmas evening.
Note: The 2.488 Gbps figure is a gross line rate. GEM framing, forward error correction and the 125 µs frame structure consume part of it, so plan on roughly 2.3 Gbps of usable downstream payload and a little over 1.1 Gbps upstream. Design against the usable number, not the headline one — at 1:128 that already trims the ~19 Mbps headline to nearer 18 Mbps down.
Upstream is where the maths bites first, and the deeper you split, the earlier it bites. Dynamic bandwidth assignment per G.984.4 distributes the 1.244 Gbps using T-CONT types 1 through 5 — fixed, assured, non-assured and best-effort — so a subscriber on a best-effort T-CONT has no guaranteed rate at all when the tree is busy. At 1:128 the whole upstream pipe is worth about 10 Mbps per subscriber at full concurrency, which leaves very little for any product with an upstream commitment (cloud backup, surveillance upload, site-to-site replication). If those services are in your package sheet, size the tree at 1:32 or 1:64 and stop worrying about the splitter cost.
The optical side follows the same pattern. Moving from 1:64 to 1:128 adds roughly 3–4 dB of industry-typical splitter loss — around 21 dB for a 1:64 device versus roughly 24–25 dB for a 1:128 device — which is exactly why a 1:128 design usually steps up to a Class C+ module or better rather than staying on Class B+. Whether you can actually run the full ratio end to end is answered by the link budget, not by the port count; the full walk-through is in Can You Run a 1:128 Splitter at Full Ratio?
The last constraint is the uplink, and on compact units it arrives well before either the PON rate or the registered ONU count does. If the uplink is a single gigabit port, 1 Gbps is the ceiling for the entire tree regardless of how deep you split — a 128-subscriber tree with even a 10 Mbps busy-hour package would already be at the cap with everyone active. Check the uplink capacity of the specific unit before you check anything else.
Four scenarios come up repeatedly, and each one has a specific thing to verify before the purchase order goes out.
| Scenario | Why one port is enough | Verify before buying |
|---|---|---|
| First area, small ISP | 128 homes passed at 30–50% first-year take-up means 38–64 active subscribers, comfortably inside one tree | That the second area has a migration path: a second PON port, not a second chassis |
| Campus or plant fiber extension | PON replaces long copper runs and intermediate cabinet power; one tree covers one building or one production line | The far end needs an ONU plus a switch. PON carries the fiber, Ethernet still lands the devices |
| Temporary or emergency link | Events, construction sites and disaster recovery need one link for weeks, not years | Fast provisioning: ONU registration by serial number, plus a pre-staged service template |
| Lab, staging or training bench | One port exercises the full stack — ranging, DBA, OMCI, encryption — with two ONUs on a bench splitter | That the unit runs a complete protocol stack and exposes the same management interfaces as production gear |
Temporary and emergency links are the case people forget until they need one. A construction site that needs connectivity for six months, a festival that needs a point of presence for a week, or a recovery site standing in for a failed link all share the same profile: one location, a handful of users, no second phase. One PON port with two ONUs and a pre-staged service template can be commissioned in under an hour, and the whole kit fits in a flight case. The catch is that a temporary link usually has no redundancy and no spare, so keep a second unit on the shelf rather than betting on a single box.
The campus case deserves a note, because it is the one most often mis-sold. A factory that wants to reach a remote building 3 km away does not need a PON tree per building; it needs fiber where copper cannot go. One PON port, one splitter, one ONU at the far end feeding an industrial switch — that is a fiber extension dressed as an access network, and it is a perfectly legitimate use of a one-port box. Buying a four-port chassis for it is spending money on ports that will stay dark.
The lab case is underrated. A single PON port with two ONUs on a 1:2 splitter lets a NOC team rehearse ONU registration, template changes, firmware upgrades and fault injection without touching production. It is also the cheapest way to find out what your service profiles actually do before a truck roll makes it expensive, and because the lab unit registers up to the same 128 ONUs as production gear, you can load the table before you load the truck.
The failure list is short and unambiguous.
More than one tree. A port registers up to 128 ONUs, and that is plenty of subscribers — but they all have to live on the same splitter cascade and inside the same loss budget. A second distribution area 4 km in the other direction is a second tree, and a second tree means a second port.
Protection requirements. ITU-T G.984.1 defines four protection types. Type B — dual PON ports on the OLT with a redundant feeder — and Type C, which duplicates the entire path, both require at least two PON ports. A single-port unit cannot offer either.
Dual uplink protection. If the service commitment depends on surviving an uplink failure, you need two uplinks with link aggregation or a failover path, which means a larger unit.
Redundant power. Small units ship with a single PSU. A central office or a cabinet that feeds paying subscribers on an SLA normally wants dual feeds.
Managed-environment expectations. Redundant fans, hot-swap modules, rack-grade thermal design and out-of-band management belong to chassis platforms, not to an edge unit designed for a wall cabinet.
Warning: The most common mistake in this category is not buying too small; it is buying one port for a site that will clearly need a second tree within twelve months. Port count is the one thing you cannot add later without replacing hardware.
Five checks, in this order.
Start from homes passed, not subscribers. A 128-home area at 40% take-up is roughly 51 subscribers, and a 1:64 splitter leaves room for take-up to double without touching the ODN.
Install the splitter for the final ratio, activate subscribers later. A 1:128 splitter serving 40 subscribers costs about the same as a 1:64 device, provided you have already budgeted the extra 3–4 dB of loss and upgraded the OLT module to suit.
Reserve fiber. Laying one spare tube during the initial build costs a fraction of a second visit, and it is what makes a second tree a splicing job instead of a civil works job.
Size the uplink before the PON. A 1 Gbps uplink caps a tree long before the 2.488 Gbps PON rate does, and with 128 subscribers behind one port it is almost always the first thing to upgrade.
Write down the growth trigger. "When the second area is approved, we buy a second PON port" is a plan. "We will figure it out later" is how single-port units get replaced at month eleven.
On cost, the short version: a two-port unit is not simply double the price of a one-port unit, because the chassis, power supply, management software and PON MAC silicon are fixed costs that do not scale with port count. The full argument is in the linked reading below, and it is worth understanding before you treat port count as a linear price multiplier.
When you evaluate a Rayin unit for a first area, the capacity conversation starts in a different place than it did five years ago. Ask for two numbers — maximum registered ONUs per PON port for this software release, and the optical budget of the installed PON module — and then let the loss budget and the uplink decide your split ratio. Rayin builds mini OLTs for edge-FTTH and fiber-extension roles, alongside the industrial Ethernet switches that sit on the far end of a PON link; port-count availability changes as the line evolves, so confirm current single-port and two-port options rather than assuming a datasheet you read last quarter still applies.
Less than most people expect, provided the templates exist first.
Power the unit, set the management IP, and give the PON port an optical budget you have already calculated. Register ONUs by serial number or password per G.984.3, bind them to a service template carrying the VLAN, the DBA profile and the T-CONT mapping, then verify each ONU receives between roughly −8 dBm and −27 dBm — the overload point and the sensitivity of a Class B+ receiver are the two edges of that window. With a 1:128 tree those readings matter more, not less: the difference between a comfortable arrival level and one that cannot absorb a single bad splice is only a few dB.
Run a speed test on one ONU with the other ONUs idle, then again with the tree loaded, and compare. If the second number collapses, the problem is upstream contention or the uplink, not the ONU. The step that separates a smooth rollout from a month of truck rolls is the template: registering twenty ONUs by hand works, registering two hundred does not.
How many subscribers can one PON port really support?
Current compact units register up to 128 ONUs on a single PON port, and ITU-T G.984.1 supports split ratios up to 1:128. The practical ceiling is usually economics rather than hardware, because 128 subscribers sharing 2.488 Gbps downstream get roughly 19 Mbps each at full concurrency, so many operators still plan at 1:64.
Does running 1:128 mean I need a stronger optical module?
Usually yes. A 1:128 splitter adds about 24–25 dB of industry-typical insertion loss, roughly 3–4 dB more than 1:64, which leaves little headroom inside a 28 dB Class B+ budget once fiber, splices and an ageing margin are added. Most designs move the OLT side to Class C+ at 32 dB per ITU-T G.984.2 and leave the ONU unchanged.
Is a one-port box always cheaper per subscriber than a two-port unit?
Not always. Chassis, power supply, management software and PON silicon are fixed costs that do not double when a second port is added, so the per-port price of a two-port unit is usually lower than the price of a one-port unit. Compare per-port economics, not sticker price.
Can I add a second PON tree later without replacing the OLT?
No. One PON port terminates one PON tree, and the ranging, timeslot grants and encryption state are all maintained per port. If a second distribution area is credible within a year, buy the second port now — the ODN is the expensive part, and it will outlive the box.
Do I still need a switch if I deploy a single-port OLT?
Yes. PON replaces distribution fiber, not the local Ethernet edge. The ONU terminates the PON and hands off copper Ethernet, and anything with more than one device behind it — an apartment, a desk cluster, a machine cell — needs a switch after the ONU.
A single PON port solves a specific problem well: one tree holding up to 128 registered ONUs, planned at 1:64 for headroom and pushed to 1:128 when the loss budget and the uplink have both been checked, in places where the alternative is copper that cannot reach or a chassis that will run dark ports for years. It stops being the right answer the moment redundancy, protection or a second distribution area enters the requirement. Check the registration ceiling first, size the optics second, measure the uplink third, and the port count becomes obvious.
Service Templates: The Shortcut to Repeatable ONU Provisioning
First-Time OLT Setup: Power On, Register an ONU, Configure Service
About the author: Sara — Sara is a Customer Manager at Rayin with over 10 years of experience in the communications field. In her free time, she enjoys badminton and swimming.
About Rayin: Shenzhen Rayin Technology Co., Ltd. — Company Profile
