Monitoring a 1,000-subscriber ISP network

A small-town ISP with about a thousand subscribers has four to eight OLTs, a couple of switches, a few hundred splitters and closures, and a crew of two to five. This is what monitoring has to do for that network, and what it should cost.

The shape of the network

At a thousand subscribers the plant is usually a core router or switch at the head end, an uplink or two, four to eight OLTs (EPON or GPON, often from two vendors bought in different years), a handful of aggregation switches at cabinets, a few dozen OBD splitters feeding a hundred or more FATs, and a drop to every house. Some of it sits in rented rooms on a power feed that fails when the town's does. Nobody sits in front of a screen all day.

What breaks, and how you hear about it today

Without monitoring, the first sign of all five is the phone. The owner learns the difference between them by sending someone.

What monitoring has to do at this size

  1. Confirm before it pages. Every plant drops a ping now and then. A tool that pages on one missed ping trains the crew to ignore it within a week. A real outage should be confirmed in about a minute and a one-off blip should page nobody.
  2. Suppress what is downstream. When an OLT goes dark, its 200 ONUs and the switch behind it are not 201 separate faults. One page, naming the OLT and the count behind it.
  3. Page the person who can fix it, where they already are. In India that is WhatsApp on the field worker's own phone, with an acknowledgement so the owner knows someone is moving, and a way to hand the job to someone else when nobody is.
  4. Say who is affected. The list of subscribers behind the fault, joined to the billing book, so support can answer "is mine down" without a site visit.
  5. Tell them. A notice to the affected subscribers cuts the calls that follow every outage, and it is the difference between an ISP that knows and one that does not.
  6. Say fibre or power. A cabinet power cut needs nobody dispatched; a cut fibre needs a splicer now. The monitoring should make that call for you.

What to read from each device

DeviceReadWhy
Core router, uplinksPing, port state, bandwidthThe slow-for-everyone fault lives here
OLTPing, PON port state, ONU roster and state, per-ONU Rx where the firmware holds it, healthThe roster is how you know who is affected; Rx is how you catch the slow splice
Switches at cabinetsPing, port state, bandwidth, CPUA cabinet without power is a switch that stopped answering
ONUsState and light level, read from the OLTNot pinged one by one
Passive plantNothing to poll; record it on a mapSo a dark PON points at a closure, not a village
NVRs, if you run camerasCamera up or downThe same crew fixes them

What each OLT vendor actually publishes over SNMP varies more than the brochures say; the vendor page states it per family.

What goes in the rack

A probe: one small computer on the network that pings and walks the devices and sends the readings out. A spare Windows machine works; so does a Raspberry Pi 5 class Linux box, which is what we recommend for a dedicated one. It should sit on the same LAN as the OLTs, dial out to the monitoring service, and accept nothing inbound. Nothing is installed on the OLTs or switches beyond enabling SNMP with a read-only community. See outage detection.

What it should cost

Pricing in this market comes in two shapes: per OLT per month, and per subscriber per month. Per OLT is cheaper when your OLTs are full; per subscriber is cheaper when they are not, and it follows the network as it grows or shrinks without a plan change. Ask for the number at your subscriber count, and ask whether the price is published, because a published price is one you can hold the vendor to.

The same network on HansaNet: 1,000 subscribers on 5 OLTs and 6 switches. Subscriber side 1,000 times Rs 3 = Rs 3,000. Device side 11 times Rs 100 = Rs 1,100. The subscription is the larger, Rs 3,000 a month, Rs 3,540 with GST, before any customer notices at Rs 0.90 each. Splitters, closures and cables are free however many you record. The pricing page has the rules and a second example.

What you can leave out at this size

Questions to ask any vendor

  1. How long from the first missed ping to a page, and what stops a single blip paging?
  2. When an OLT goes dark, how many messages does my crew receive?
  3. On my OLT model, where does the per-ONU Rx come from, and can you show me?
  4. Does it join my existing billing panel, or does it want to replace it?
  5. What is the price at my subscriber count, and is it on your website?
  6. What happens to alerts if I miss a payment?
  7. What do I install on my OLTs, and what on my network accepts inbound connections from you?

The answers for HansaNet are on the features pages.

Put a probe on your plant tonight.

One setup .exe on a Windows machine in the rack, or a .deb on Linux. Rs 3 per ONU a month, no plans, no caps.