New ISP Core Network Installation Guidance

ISP FTTH core network Proposal:

1. Summary

  • What we build: A redundant, standards-based core with five parts:
    • a border router
    • two subscriber gateways (BNG/NAT)
    • a dual aggregation switch pair
    • GPON OLTs with passive fiber distribution
    • two servers for RADIUS, DNS and monitoring
    A battery-backed −48 V power plant sits underneath all of it.
  • Design principle: Buy small, proven, current-generation “bricks” and add one per growth step, instead of one large chassis now. A subscriber gateway or OLT chassis is added roughly every 1,250–2,500 customers.
  • Budget (planning figures, not quotes): Phase 1 is about (57 Lakh) including 15% contingency, for 2,500 customers on hardware that can grow. Phase 2 is about (24 Lah) to reach 5,000. ONUs and drop cables are per-customer costs.
  • Not included: The ISP license, upstream bandwidth, outside-plant fiber, and building works.

2. Sizing assumptions (please validate against your real traffic)

  • Customers: 2,500 active now, 5,000 design target, and scale-out beyond that.
  • Busy-hour traffic: I assume 2.5 Mbps average per customer. That is about 6 Gbps today and about 12.5 Gbps at 5,000. With 30% headroom, plan for about 16 Gbps. Measure your current network and adjust.
  • Customer sessions: PPPoE authenticated by RADIUS. Plan speeds are enforced in the OLT, not on the routers. A forum user reports about 2,000 PPPoE clients at roughly 35–40% CPU on a CCR2116 after moving NAT elsewhere and bandwidth management into the OLT.
  • NAT: Customers sit behind private addresses with NAT on the gateways. We add IPv6 as soon as the upstream provides it, which takes load off NAT.

Why these platforms

  • MikroTik for routing: The CCR2216 lists at US$2,795 and the CCR2116 at US$995, against several times that for carrier-class routers. Both use the same 16-core CPU. Operators report real trouble when PPPoE and NAT share one box at around 3,000 customers. So, we scale out with several CCR2116s instead of one big router. 
  • ZTE for the OLTs: A modular, DC-native GPON platform with a large ONU ecosystem.
ZTE ZXA10 C300:

ZTE ZXA10 C300


ZTE ZXA10 C300


  • No end-of-support gear: The previous quote’s MX5 reached end of support on 30 June 2026.

Here is how everything connects:

ISP Core Network Installation Guidance Diagram

Dell PowerEdge R660 server

I’d use two identical Dell PowerEdge R660 servers (1U, or a newer-generation equivalent like the R670 if your supplier stocks it). I changed only the Servers x2 box in the diagram. Everything else is as in the author’s original.

Dell PowerEdge R660 server

Why two identical servers

Each server runs the full service set, so either one can fail without customers losing logins or DNS:

  • Server A: primary RADIUS, primary DNS resolver, and the NMS (LibreNMS).
  • Server B: secondary RADIUS, secondary DNS resolver, and a standby or backup copy of the monitoring data.

Configure the BNGs to try both RADIUS servers, and hand out both DNS IPs to customers.

Suggested spec per server

  • CPU: one 8–16 core Xeon Silver or Gold. RADIUS and DNS for 5,000 subscribers are light, so a second CPU isn’t needed.
  • RAM: 64GB, which leaves room for LibreNMS and a DNS cache.
  • Storage: 2x enterprise SSD (about 960GB) in RAID1. Avoid HDDs for the RADIUS accounting and NMS databases.
  • Network: a dual-port 10G SFP+ NIC. The diagram shows a 10G link from each aggregation switch, so use one port to each CRS326 (bonded or active/backup).
  • Power: dual redundant PSUs. Dell offers a -48V DC PSU option on this generation, which would match your battery-backed plant. Confirm it with your supplier, or run AC PSUs from an inverter or UPS instead.
  • Management: iDRAC Enterprise for remote access, since you won’t always be on site.

Software

  • Use Proxmox VE or plain Linux with FreeRADIUS plus SQL, Unbound or BIND, and LibreNMS.
  • Pick one NMS rather than the quote’s Nagios, Cacti and LibreNMS together. LibreNMS covers most of what you need.

Alternatives

HPE ProLiant DL360 Gen11 or Lenovo ThinkSystem SR630 V3 are equivalent. Buy whichever has the best local warranty and spares in Nepal, and use the same model for both servers. Refurbished R650/R640-class machines are a cheaper option. They are far newer than the quote’s R620, but check how long the supplier will honor the warranty.

ODF (Optical Distribution Frame) + Splitter

ODF (Optical Distribution Frame) + Splitter

An ODF and splitters make up the passive fiber layer between the OLT and your customers. They need no power, which is why the diagram calls it “passive fiber plant”. I explain what each part does and how to size it from the numbers.

ODF (Optical Distribution Frame)

The ODF is a patch panel for fiber, mounted in a rack or on a wall in the OLT room. Each PON port on the OLT is connected with a short patch cord to an adapter on the ODF. Behind the adapter, a pigtail is spliced to the outgoing fiber cable that leaves the building. It is where you can patch, test and re-route fibers without touching the cables outside.

Splitters

A passive PLC splitter divides one PON fiber into many (1:8, 1:16, 1:32 or 1:64), so one OLT port serves many customers. Each customer’s ONU gets its own drop fiber from the last splitter. Splitters come in several forms:

  • Rack or LGX modules for centralized splitting in the OLT room.
  • Cassettes or boxes installed in street cabinets (often called FDTs).
  • Terminal boxes on poles or buildings (often called FATs).

Use PLC splitters, not the cheaper FBT type.

Where to put the splitters

This decision drives the quantities:

  • Centralized: a 1:32 splitter at the ODF. Every output needs its own feeder fiber and ODF port, so 64 PON ports means about 2,048 feeder fibers. That is heavy and expensive.
  • Distributed (the usual way): one feeder fiber per PON port goes out to a street cabinet. A first-level splitter (such as 1:4) sits there, and second-level splitters (such as 1:8) sit closer to customers. This needs only about one feeder fiber per PON port.

Sizing for your design

For the distributed approach, with the article’s 64 PON ports in phase 1 and 128 later:

ItemPhase 1 (64 PON ports)Phase 2 (128 PON ports)
ODF adapters, OLT side (SC)64128
ODF feeder-side fibers (plus about 20% spare)about 77about 154
First-level splitters (1:4 or 1:8)64128
Second-level splitters (1:8) at 1:32 overall256512

For a 1:64 overall ratio, using 1:8 then 1:8, you need 64 first-level and 512 second-level splitters in phase 1. Adjust the numbers to your area. Real counts depend on your cable routes and where customers are.

Check the split ratio first

At 1:32, 64 PON ports can serve at most 2,048 customers, which is fewer than your 2,500. Even at 70% fill, 2,500 customers need about 110 ports at 1:32. At 1:64 the same 2,500 customers need about 56 ports at 70% fill. Make sure the “64 PON ports” assumes the right split ratio. Also confirm the optical budget works at 1:64, including fiber length, splices and connectors. Typical splitter losses are about 17 dB at 1:32 and 20 dB at 1:64, and the OLT’s SFP class (C+ or C++) sets how much you can afford.

Other points

  • Connectors: Keep the connector type consistent end to end. Check whether your OLT SFPs and ONUs use SC/UPC or SC/APC, because mixing them damages connectors and adds loss. Some ZTE ONUs use SC/UPC, but verify the models you buy.
  • Racks: ODFs commonly come in 12, 24, 48 or 96-port sizes. For 64 OLT-side ports, two 48-port panels or three 24-port panels would do.

Post a Comment

Previous Post Next Post