Fiber Network Architecture for ISP Scale-Up: Rings, Hubs and PON Migration

Fiber Network Architecture for ISP Scale-Up: Set the Design Basis

Fiber network architecture for ISP scale-up is the arrangement of transport and access plant that lets an operating ISP expand without rebuilding its first service area. XGS-PON provides a standardized 9.95328 Gbit/s line rate in each direction, but growth still depends on feeder capacity and the routes connecting aggregation sites.

A faster access card cannot create a spare strand through a full feeder cable. That is the mistake we want the architecture review to catch before an expansion becomes a procurement exercise. We recommend starting with the constraint that is hardest to change in the ground, then working back toward the equipment that can be replaced inside a rack.

The rate comes from ITU-T Recommendation G.9807.1, 10-Gigabit-capable symmetric passive optical network (XGS-PON), the February 2023 edition approved in February 2023, with Amendment 1 approved in May 2025. It is a shared optical line rate, not a promised customer throughput. We keep that distinction on the design basis so commercial service targets do not become unsupported equipment assumptions.

Keep the operating network in the drawing

The existing build is a boundary condition, not a blank canvas. We start by reconciling its recorded connections with the expansion boundary, especially where an apparently unused fiber terminates before the proposed new serving area. A cable count without its splice continuity can make the first construction phase look easy while concealing the connection that will stop commissioning.

Our preferred starting point is a centralized splitter location where demand is still uneven and technicians need to reassign ports without opening several field enclosures. That preference has a cost. It can require more distribution fiber and concentrate maintenance activity at a cabinet that becomes awkward to expand; we would reject it where that physical constraint outweighs the flexibility.

Set operating conditions before comparing drawings. We want the owner to state what must remain available during a transport failure and which service areas can tolerate a planned interruption. Those are different requirements. A residential expansion that can move during a maintenance window should not silently inherit the protection promise written for a dedicated business circuit.

Growth also needs a boundary in space. We distinguish committed extensions from possible later territories, and we reserve routes according to the difficulty of reaching them again. Our design should show where the next feeder can enter a hub without disturbing working circuits. If the only path requires rebuilding the entrance facility, the future capacity shown on the schedule is incomplete.

Capacity is not a single percentage. Busy-hour uplink utilization answers a different question from available splitter outputs, and neither establishes whether the cable entering the next neighborhood has usable spare fibers. Identify each layer's constraint separately. This prevents an unused chassis slot from being offered as evidence that the outside plant can absorb another phase.

Separate a passive fiber distribution hub from an active aggregation site in the site schedule. The passive hub does not gain forwarding capacity when an ISP changes its routing design. An active site needs a power and thermal plan for its eventual equipment population, including operation during utility loss, even when the opening build installs only part of that population.

We use the choices below as an expansion review, not as a universal network diagram. Each row names a decision that can make later phases easier or harder. Carry the selected option into the HLD with its reason, and record the condition that would justify changing it; otherwise the next designer inherits a symbol without the decision behind it.

Core and Aggregation Rings Must Survive the Same Failure

Core rings versus aggregation rings

A core ring connects major network sites; aggregation rings gather access traffic into that core. We keep their failure domains visible rather than drawing a single circle around every site. A break in an aggregation segment should have a defined effect on the core, and the remaining links must carry the rerouted load without relying on normal-operation headroom that has already been sold.

Physical diversity is the first test. Two logical directions that occupy the same duct or cross the same vulnerable structure can fail together, regardless of how clean the ring looks in the routing diagram. We trace the shared section on the civil plan and identify its consequence. Protection is only as independent as the paths it actually uses.

For an expanding ISP, a smaller aggregation ring can localize maintenance and make capacity planning easier to explain. Its interconnection points need deliberate protection. We compare that arrangement with dual-homing aggregation nodes to separate core sites, using the available routes and operating requirements; the shape that looks more symmetrical on a slide does not win by default.

Failover loading deserves its own calculation. We model the traffic that lands on the surviving direction after the specified failure, including any traffic already using that path, then compare it with usable link capacity. There is no defensible universal subscriber threshold for closing a ring. The trigger is the service commitment and the failure load the current topology cannot support.

A protected transport ring does not automatically protect the PON feeder beyond the OLT. We mark that boundary explicitly in the HLD and test the intended failure behavior during acceptance. Where a circuit requires protection through the access layer, its design needs that protection specified separately; an upstream ring cannot repair a severed passive branch by changing a route.

Hub Sizing and Centralized Versus Distributed Splitting

Size the working hub, not just its port label

A hub is full when it cannot accommodate the next work safely, even if its nominal termination count has not been reached. We check the installed configuration against the intended expansion configuration, including splitter mounting and fiber routing. The fiber distribution hub sizing guide addresses the cabinet detail behind that check. Reserve space must remain usable after patching, not merely visible on an elevation.

We prefer adding a serving hub when extending the original one creates long distribution paths or an unworkable cable entrance. Expanding the original hub is the better choice when its feeder reserve and working space remain suitable and the additional footprint would create avoidable maintenance. The decision belongs in the route model, where actual serving boundaries can change the result.

Centralized splitting

Centralized splitting places the split at a common field location, making it easier to assign available distribution fibers to the required PON ports. For an ISP filling gaps inside an established footprint, we value that flexibility because demand may not follow the original phase boundaries. The tradeoff is a larger concentration of fibers and terminations at the central location.

That benefit requires accurate assignment records. A technician cannot safely reassign a connection from a drawing that no longer matches the cabinet. We therefore design the hub identifier and port naming with the operating team before release. Our FTTH HLD and PON engineering service connects those decisions to the detailed design so the field labels carry the same meaning as the architecture.

Distributed splitting

Distributed splitting moves part of the split farther into the distribution network. It can reduce the number of fibers required on some distribution segments and reduce the size of individual access points. The branch arrangement becomes more fixed. We choose it where the serving pattern is stable enough that the saved cable capacity does not create repeated reassignment work later.

Do not judge a cascade by its final split ratio alone. A single-stage 1:32 split and a 1:4 followed by 1:8 both create 32 nominal outputs by arithmetic, but their actual component losses and connection counts need separate budgets. These are layout examples, not recommended subscriber counts. We also check how a technician isolates a fault beyond the first splitter.

ITU-T Recommendation L.250, Topologies for optical access network, the January 2024 edition approved in January 2024, distinguishes optical access arrangements including point-to-point and branching topologies. It supplies a vocabulary for the design, not a reason to impose the same split placement everywhere. Our recommendation changes when the route geometry changes, even within the same operating footprint.

Spare Fiber Has to Reach the Expansion

A spare percentage attached to an entire network hides the exact place the reserve may be missing. We allocate usable fibers by cable segment, with particular attention to crossings that would be difficult to reopen. The reserve on the accessible side of a crossing cannot substitute for reserve through it. This is why an average utilization figure is inadequate for an expansion decision.

Keep restoration reserve separate from fibers already assigned to future service. Otherwise the expansion consumes the recovery option while the inventory still describes those strands as spare. We require an identifiable end-to-end path before counting a reserve as available for an extension. An unspliced fiber ending in an intermediate closure is a different resource from a tested path between sites.

Reserve also has a documentation cost. Our fiber as-built GIS documentation standards explain why connectivity needs to survive the handoff from construction records into operations. We want reservations tied to an actual route and released when their purpose disappears. Otherwise a future designer must choose between trusting an obsolete reservation and risking a live circuit.

The physical plan should show how new cable reaches a reserved termination without displacing existing slack storage. We review closure access against the planned splice work, not just the drawing symbol. A larger cable may solve the strand shortage while creating a closure problem downstream; that trade belongs in engineering before the material order, when the alternatives are still practical.

GPON to XGS-PON: Reuse the Plant Only After Checking It

Separate overlay versus shared-plant coexistence

A separate overlay assigns the new service its own optical path. Shared-plant coexistence keeps compatible GPON and XGS-PON systems on the same distribution network using the appropriate wavelength arrangement. We favor reuse where the installed plant and selected equipment support it, but an architecture note saying upgrade-ready does not establish compatibility or create optical margin.

ITU-T Recommendation G.984.2, Gigabit-capable passive optical networks (G-PON): Physical media dependent (PMD) layer specification, the August 2019 edition approved in August 2019, specifies GPON physical-layer requirements. The familiar 2.48832 Gbit/s downstream and 1.24416 Gbit/s upstream combination differs from the symmetric rate in G.9807.1. Existing GPON terminals do not become XGS-PON terminals when the upstream equipment changes.

We recalculate the optical path for the intended system, including any coexistence component inserted during migration. Use the actual splitter loss specification and measured plant condition, with the required operating margin. Check both directions and the applicable optical interfaces. A passing GPON acceptance result is useful evidence about the existing plant, but it is not an XGS-PON acceptance result.

Migration can expose an aggregation limit before it exposes an access limit. We compare the proposed access upgrade with the OLT uplink and transport path under the failure case already defined in the HLD. Adding faster PON ports while leaving a congested surviving ring direction untouched moves the bottleneck; it does not fulfill the service objective that justified the upgrade.

For the cutover, we need an explicit mapping from the existing connection to the intended terminal and service profile. The GPON design guide on split ratios and OLT placement provides the access-planning context. We also keep a workable rollback path until acceptance confirms the new service, because the ability to reverse a change should be established before the old assignment is removed.

Choose Fiber Network Architecture for ISP Scale-Up by Constraint

ISPs extending a working footprint: retain the existing hub where its usable reserve reaches the new boundary, and use centralized splitting when port reassignment is the controlling need. Do not add a new active site just to tidy the map. Our priority is to preserve working service while making the next extension possible without reopening the same feeder.

ISPs adding a separate serving area: give the new area a hub boundary that fits its routes, and select distributed splitting where the branch arrangement will remain stable. Protect its aggregation connection according to the service requirement. The new territory should not depend on undocumented spare capacity in the oldest part of the network.

ISPs changing the service mix: resolve migration compatibility and surviving-path capacity before selling the upgrade. We recommend a separate optical path when shared-plant coexistence cannot meet the required loss budget or operational constraints. Faster optics are the access decision; the transport decision must still hold when the normal path is unavailable.

Draftech connects feeder continuity to hub assignments through engineering and full turnkey delivery services, with engineering performed 100% in-house and construction delivered by Draftech-managed subcontract crews. Active in 24 states. Available across all 50 U.S. states. For an expansion with uncertain capacity at the old-to-new boundary, send the route context to info@draftech.com so we can discuss the engineering scope.

If the next phase needs a route engineered before the architecture can be finalized, consider Draftech free route design: Draftech engineers the first 20,000 linear feet of a qualifying route at no cost, from feasibility and field survey through permit approval. The owner reviews each request before Draftech commits the package.