The most expensive mistakes in an FTTH rollout don't happen during construction. They happen during design, weeks or months before any fiber goes in the ground, when decisions that feel small at the time become structural constraints you're working around for the next decade. A split ratio that made sense at low subscriber density creates a capacity bottleneck at 60% penetration. An OLT placed for convenience rather than topology adds latency and complicates redundancy. A /24 allocated to a residential block runs out of addresses eighteen months into the rollout.
None of these are exotic failure modes. They're the decisions that get made in a hurry during the planning phase and that show up as real costs, either in network performance, in expensive remediation work, or in limits on how far you can grow the network before hitting a wall you built yourself.
This is a walkthrough of the decisions that matter most, and why they matter, in the sequence you actually face them during design.
Demand Forecasting Before Topology
The first mistake most ISPs make is drawing the network before they've done a real demand analysis. Topology decisions, where to place OLTs, where to run feeder cable, how many distribution points to create, all depend on where subscribers are and where they'll be in three to five years. Getting this wrong in the topology phase means either overbuilding a low-density area you can't recoup costs on, or underbuilding a high-demand area and returning to trench the same streets again.
A usable demand forecast at the planning stage doesn't require a detailed subscriber model. What it requires is honest answers to: what is the current population density of the target area by block, not the whole area averaged, what is the realistic penetration rate at your target price point in this specific market, and where are the anchor tenants or commercial clients who justify backbone investment independent of residential penetration. In Pakistan's residential ISP market, corporate CIR clients often change the economics of an FTTH build significantly, a handful of corporate clients on a feeder route can justify infrastructure that residential penetration alone wouldn't.
From that analysis comes the first major design input: subscriber count by geographic cluster, which drives OLT sizing and placement.
OLT Placement: The Decision That Constrains Everything Downstream
OLT placement is the first big architecture decision, and it's also the one with the highest long-term cost if it's wrong. The OLT is not something you move after the network is built. Its location determines fiber routing, determines where your aggregation layer sits, and determines the latency profile of every subscriber on that OLT's PON ports.
The factors that should drive OLT placement, in order of importance: proximity to the area of highest subscriber density, since fiber distance from OLT to subscriber affects both loss budget and the split ratio you can practically achieve; proximity to your aggregation infrastructure or the ability to build aggregation there cost-effectively; physical facility requirements including power, cooling, and physical security, which rules out some otherwise logical locations; and redundancy considerations, since an OLT at a single physical location with no path diversity to your core is a single point of failure for everyone on it.
The mistake that costs money later is placing the OLT where space is available or where a building was convenient rather than where the topology actually wants it to be. This is especially common in greenfield builds where the ISP doesn't yet have a PoP in the area and uses a partner's facility by default. If that facility is at the edge of the service area rather than near the center of subscriber density, the fiber plant you build around it will be suboptimal for the entire life of the network.
Split Ratios: The Tradeoff Between Density and Flexibility
GPON allows up to 1:128 split in theory, though most deployments run 1:32 or 1:64 for practical reasons. Choosing your split ratio is a tradeoff between current subscriber density, future capacity needs, and the optical budget your fiber plant can support.
A 1:64 split on a 20dB optical budget is aggressive and leaves very little headroom for connector losses and fiber aging. A 1:32 split gives more optical budget headroom and allows for better individual subscriber signal quality, but means you need twice as many OLT ports to serve the same subscriber count. In high-density residential areas with good fiber quality, 1:64 is defensible. In areas where fiber plant quality is uncertain, or where you expect high growth and want the option to add subscribers without optical budget problems, 1:32 is the safer choice.
The decision that costs money later is deploying a flat split ratio across the entire network regardless of density. High-density blocks can support higher split ratios, low-density areas in the same rollout may not. A design that treats all areas identically either wastes OLT ports in high-density zones or creates optical budget problems in low-density ones.
Also worth deciding early: whether to use a single-stage or two-stage split. A single-stage split puts all the splitting at one point close to the OLT, simpler to manage but less flexible for coverage changes. A two-stage split (a lower ratio splitter at a primary distribution point, then a second splitter closer to subscribers) gives you more flexibility to extend coverage into adjacent areas and is generally better for a network you expect to grow, at the cost of more passive infrastructure to track and maintain.
Feeder vs Distribution vs Drop: Getting the Layer Boundaries Right
The three layers of a GPON access network have different design constraints and different consequences for getting the boundaries wrong.
The feeder layer runs from the OLT to primary distribution points. This fiber should be high-count single-mode, routed with redundancy in mind, since a feeder cut affects every subscriber on those distribution points simultaneously. This is where you should spend the most on physical route diversity: dual-route feeders through different physical paths are a meaningful protection against construction accidents, which are the most common cause of major outages in urban FTTH networks.
The distribution layer runs from primary distribution points to secondary distribution points or to street-level cabinets. This is where your splitters typically live in a two-stage design. The key decision here is cabinet placement: cabinets need to be accessible for maintenance, physically secure, and positioned so the drop distances to subscribers stay within your optical budget. In dense urban areas this is relatively straightforward. In areas with difficult physical access or irregular building layouts, cabinet placement needs to be resolved in a physical survey before design is finalised, not assumed from a map.
The drop layer, from cabinet or pole to the subscriber's ONT, is where most of the per-subscriber installation cost lives and where most of the physical variation occurs. The design decisions that matter here are whether drops are aerial or underground, what the maximum drop length is for your optical budget, and how slack loops are handled at the OLT and subscriber ends.
IP Address Planning: The Mistake That's Surprisingly Common
Running out of IP addresses in a live FTTH network is a genuinely painful operational problem. It can't be resolved quickly, it affects new subscriber provisioning, and depending on how your CGNAT or routing is configured, fixing it may require a maintenance window and reconfiguration across multiple systems.
The common mistake is allocating a single /24 to a residential FTTH deployment and assuming it's enough. A /24 gives you 254 usable host addresses. If you're targeting 400 subscribers in that area at 60% penetration, you've already exceeded it before the rollout is half complete.
For subscriber-facing addressing, plan for your target subscriber count at full penetration plus at least 30% headroom for network devices, test addresses, and growth. For CGNAT pools if you're using CGNAT, scale against your concurrent session requirements, not just your subscriber count. For infrastructure addressing (OLTs, aggregation switches, management interfaces), use a completely separate block with no overlap risk to subscriber space.
For new PTA-licensed ISPs, keeping subscriber address space, management space, and CGNAT pools in separate, documented allocations from the beginning makes CTDISR-2025 asset inventory requirements significantly easier to satisfy during audit, since you're not reverse-engineering what address block contains what type of device.
Aggregation Design: Where the Access and Core Meet
The aggregation layer is where GPON subscriber traffic aggregates before hitting your core routing. The key decisions here are: how many subscribers aggregate to a single aggregation switch port, what the redundancy model is for aggregation switches themselves, and how traffic engineering is handled for traffic shaping and QoS before traffic reaches the core.
Aggregation switches are where most of the per-subscriber QoS policy gets applied in a GPON network. Getting subscriber profiles, committed rates, and burst allowances applied consistently at the aggregation layer is significantly easier than trying to do it at the OLT level or in the core. Designing the aggregation layer with the QoS architecture already defined means you're not retrofitting policies into a live network while subscribers are on it.
The redundancy question at aggregation is one most ISPs get right conceptually (redundant uplinks to core, redundant power) but get wrong in the details (both uplinks going through the same physical route, defeating the purpose). Physical path diversity from aggregation to core is worth verifying explicitly during design, not assumed from a diagram.
If you're planning a FTTH rollout and want the architecture reviewed before any physical work starts, that's exactly the kind of engagement we do under Network Design & Optimization. For operators assessing whether their existing FTTH plant is performing as well as it should, or evaluating an acquisition target's infrastructure, ISP Consulting & Advisory covers the diagnostic and strategic assessment work. And for operators building their NOC capability alongside the network, NOC Enablement & Monitoring ensures the monitoring and operations layer is designed for the network you're building rather than bolted on afterward.