It started as a party trick. I was visiting a network in rural Maharashtra I'd never touched before: my parents' house in a small town in Nashik. I wanted to know what was actually on the other end of the fiber, not "is the wifi working" but what is this thing, really. So I opened a terminal and started running the commands I'd run on any network: curl ipinfo.io, traceroute, whois, dig.

An hour of idle curiosity turned into a two-week thread covering an ISP's corporate structure, India's telecom licensing tiers, the actual, occasionally violent politics of who gets to run a cable down your street, and the physics of how a single strand of glass ends up in forty different living rooms. This post covers the investigation. What I built in response is in part two.

First pass: ASN, routing, CGNAT

First pass, standard toolkit. The public IP resolved to an ASN registered to a company I'll call Regional Netsol here rather than its real name, along with the rest of the identifying details in this post. A whois on that ASN showed it multi-homed through three major Indian carriers simultaneously: Tata Communications, Reliance Infocomm, and Tata Teleservices, each accepting a full route table (accept ANY). I'd assumed a tiny local operator running a single upstream. Three-way multihoming with a full table on each carrier is not that.

A traceroute added a second surprise: the path out to Google's DNS hairpinned through Tata's Bangalore node before doubling back to Mumbai, a thousand-kilometer detour for what should've been a short hop. Not a bug, just a consequence of whichever of those three upstream peers happened to be carrying that particular destination.

Then the CGNAT. The router's own status page reported a private address as its WAN IP, while the outside world saw a completely different, public one. Somewhere between the house and the internet, my parents' connection was sharing a public IP with an unknown number of other Regional Netsol customers. No inbound port forwarding was ever going to work from that router alone; anything I wanted reachable from outside would need an outbound tunnel instead.

The PPPoE log

None of that was the actual finding. The router's own PPPoE negotiation log, the raw handshake between the house and the network, had this line in plain text:

pppoe receive PADO AC-Name(MY CABLE GUY)

That's not visible from any IP-layer tool: traceroute, whois, none of it. PPPoE's discovery handshake happens below IP entirely, at the Ethernet layer, before an address even exists. The only way to see it is to read it straight off the equipment doing the negotiating. It told me that Regional Netsol wasn't actually who was running the wire into the house. A franchise, which I'll just call My Cable Guy, operating out of a small town in Nashik, was the actual last-mile operator. Regional Netsol was the upstream, the backbone, the ASN. My Cable Guy was the one who'd actually shown up with a ladder.

That distinction raises two questions: what does it cost to run an operation like either of these, and why does this two-tier franchise structure exist at all?

What running Regional Netsol actually costs

Regional Netsol turned out to be bigger than my first read of "small regional operator, surprisingly well-connected." It sells retail under a brand I'll call Ghar Broadband, holds a Class A ISP license with UL and NLD authorization plus a VNO license across more than a dozen telecom circles, operates across several states, and has an active infrastructure partnership with a major network vendor: a licensed, mid-tier national operator. The three-carrier multihoming stopped being surprising once I had that: it's what an NLD-licensed operator running backbone infrastructure is supposed to look like. My Cable Guy is a local franchise reselling under the Ghar Broadband umbrella, the standard way a national license reaches a town too small to justify a company-owned PoP.

The distance between "local franchise" and "Regional Netsol, national licensee" is a specific, priced ladder:

• Category A (national): ₹30.5 lakh DoT application fee, ~₹2.1 crore bank guarantee, 20-year validity, 3-5 months to process.
• Category B (state/circle): ~₹2.15 lakh application fee, ~₹11 lakh bank guarantee.
• Category C (district/SSA): entry fee capped around ₹50,000, the tier a franchise-scale operator could plausibly reach on its own.
• UL-VNO: the budget route, mirrors the same A/B/C tiers but only valid for 3 years before renewal, designed for operators on a tight budget.

Or you skip licensing entirely and buy bandwidth wholesale from someone who already holds one, which is presumably close to what My Cable Guy is actually doing under Regional Netsol's license right now. Faster to revenue, but it caps both independence and margin.

A 50-70 Mbps commit, enough for a first 100-150 customers, runs roughly ₹25,000/month. A full 1 Gbps commit from a Tier 1 provider is ₹2-4 lakh/month. Every customer connection an operator sells is a slice of that one bulk pipe, oversubscribed the same way a GPON splitter oversubscribes fiber (more on that below). Wholesale per-Mbps pricing is the number that decides whether a franchise like My Cable Guy is profitable or barely surviving.

Total investment estimates ranged from ₹4-5 lakh for a bare-bones reseller setup (bandwidth contract plus switches, routers, fiber patch cable) up to ₹13-30 lakh for a properly licensed operator with its own network equipment. One concrete data point: a Kerala operator built a GPON network for ₹8 lakh in 2024, grew from 80 to 400 subscribers in ten months, and recovered the investment in 8 months at ₹650 ARPU. That's the economics of running an ISP in a village too small for Jio to bother with.

The last-mile ownership problem

Small local ISPs in India overwhelmingly grew out of the cable TV business, and that lineage explains why dealing with one feels nothing like dealing with a corporate telecom. Worth understanding the industry's history here in general terms, not as a claim about any specific operator, current or past, that my family has actually dealt with.

Cable distribution in India developed as a turf-based, cash business with almost no formal enforcement mechanism, and broadband simply inherited that structure, because the same guys running coax up your street are the ones who ran fiber alongside it. The core problem is structural: an operator's entire business is the physical cable running into your building, and there's no registry anywhere that says "this duct, this pole, this riser belongs to Operator X." Whoever physically laid the wire and successfully defends it keeps the customer. India has an estimated 60,000-70,000 local cable operators who are the literal last mile, and who own the physical coax reaching into homes, sitting underneath every national broadband statistic.

Political patronage got baked in early, especially in dense urban markets. Cable operators in Mumbai have had explicit party affiliations for decades, and disputes over an operator breaking away from an MSO to start a rival network have escalated to actual shootings. Poaching a rival's territory has no clean legal remedy either. You can't sue someone for "stealing my subscriber" when there was never a formal territorial grant to begin with. What happens instead is negotiation, intimidation, or occasionally just cutting the rival's cable. People in the industry refer to "cable mafia" as a standard operational hazard, right up there with rain and trees blocking a line of sight. The ceiling on how seriously this gets defended is dark: a Dhaka cable operator was killed in a dispute explicitly over control of a cable connection business, and a rival's associate was later killed by police in a gunfight while being pursued for it.

Even the national fiber rollout isn't immune. BharatNet, India's government fiber backbone project, got stuck in a turf war between the telecom ministry and the roads ministry over right-of-way fees for laying cable along highways. If two central ministries fight over duct access, it's not hard to imagine how the same fight plays out between two guys competing for the same apartment complex with no formal right-of-way agreement at all.

My Cable Guy is most likely a small reseller under Regional Netsol's license rather than a license holder itself, a normal way to run a business at this scale. It also explains why getting a straight answer out of them about basics (bridge mode vs. double NAT, whether they'd support a customer-supplied ONT) has taken a while: this is a small, relationship-run operation, not a corporate support desk with a ticketing system.

It also explains why an operator like this survives: the big three, Airtel, Jio, and Tata, are increasingly cornering the broadband market in metros, leaving small local operators viable mainly in villages and semi-urban areas the majors don't consider worth the fight. My Cable Guy isn't competing with Jio. It's operating in the gap underneath Jio's interest.

The physical layer: GPON

The last piece was the physical layer. I asked what the ONT, the box where the fiber terminates in the house, actually was, expecting to hear "cheap Chinese media converter." A photo of the unit settled it: it's a Syrotech GPON ONU, identifiable by front-panel LEDs labeled PON and LOS, Passive Optical Network sync status and Loss-of-Signal detection. Both are meaningless labels on a simple point-to-point media converter, but they're standard indicators on a GPON terminal authenticating against an OLT.

That photo ruled out the simpler architecture I'd half assumed: a dedicated fiber strand from My Cable Guy's shack straight to my parents' house. What's actually there is a shared, tree-topology PON, almost certainly GPON, the dominant standard in India's small-ISP space. Three pieces of hardware make up that topology:

• The OLT (Optical Line Terminal) sits at the operator's shack: the active electronics, the ISP-side "PON switch." A modest 4-port unit costs ₹80,000 to ₹2 lakh and can serve up to 256 subscribers.
• Optical splitters are the trick that makes this not point-to-point. A splitter is a completely passive, unpowered piece of glass that divides one incoming fiber's light across multiple outputs: 1:8, 1:16, or 1:32, sometimes cascaded to reach 1:64. These sit in a box on a pole, in a manhole, or in a building's utility closet, and they're the reason one fiber leaving the shack can end up serving thirty or more homes down the line.
• The ONT, the Syrotech box on the wall, terminates the fiber and hands the house Ethernet. ONUs run ₹800-1,500 a unit.

Downstream traffic is literally broadcast: every ONT on a splitter tree receives the same optical signal, all the data for every customer on that port, and each unit only decrypts what's addressed to it (GPON uses AES specifically because of this). Upstream uses TDMA: the OLT assigns each ONT a specific time slot so upstream bursts from different houses don't collide on the shared strand, continuously re-measuring each unit's distance to keep the timing correct.

The capacity number that matters for day-to-day performance: a GPON port typically has a fixed 2.5 Gbps downstream / 1.25 Gbps upstream shared across every home on that splitter tree. If My Cable Guy has thirty houses hanging off one port, each nominally sold "100 Mbps," that 2.5 Gbps pool is being oversubscribed hard, the exact same logic as the bulk wholesale bandwidth oversubscription discussed above, just one layer closer to the house. It's also the most likely explanation for performance that degrades specifically at peak hours even when the ONT itself reports a clean, healthy signal: the contention is happening upstream of the house, on a shared strand nobody in the living room can see.

Open question

By the end of this pass: a licensed national operator's infrastructure, resold through a local franchise running on turf and relationships rather than a corporate SLA, delivered over a shared fiber tree oversubscribed at two different layers. None of it explained the thing that started this in the first place: a stubborn, months-old upload cap that had the house stuck at a ninth of its download speed. Part two covers what I built to actually pin that down.