the isp business model

Capital includes routing equipment, switching equipment and wires, racks.

Operating costs are office space, salaries, bandwidth from the upstream
provider, power, maintainance.

Wiring costs can again be divided into two: the cost of the wire itself,
and the termination cost. It costs exactly as much to lay wire to
provide 256 kbps service, as to lay wire for 2Mbps (or 100 Mbps).

Termination costs, however, vary drastically from the low end to to
higher end. Terminating copper is drastically cheaper than terminating
fibre (you didn't think why so few systems have direct fibre input?).

Again, ISPs can't afford to use cheap equipment (please don't point me
at the cablewallahs who do, they aren't working on the scale of the
ISP). Managable equipment costs significant $$$.

Now, keeping this in mind, realise that the last mile (or last few
metres) cost as much or more to lay than the backbone. Running and
maintaining a single wire over a few hundred km is far cheaper than
running dozens of wires, and terminating them.

With wireless, the cost of the last mile is lower by an order of
magnitude. However, wireless does not scale the way wired networks do.
Wired networks are switched, while wireless is not. This means that the
more users on a single wireless access point, the slower it gets. And if
it boils down to sharing 10 Mbps wireless with 15 people, or a 100 Mbit
switched network shared at the switch with 24 people, I'll go with the
switched network any time.

A single wireless user using P2P can hog all the bandwidth available in
that spectrum. With rural areas, or less densely populated areas,
wireless works out better, because the cost of laying fibre and copper
is so much higher.

Fibre by itself is cheaper than copper per meter (a couple of INR less
per meter). However, fibre termination equipment costs a few thousand
INR more.

Bandwidth costs can be reduced by (1) buying less bandwidth, or (2)
peering between ISPs of equal size.
Peering in India is mired in political problems caused by people drawn
from the BSNL pool of staff.

ISPs try to reduce capital costs by avoiding having their own equipment
in the last mile. This gets pushed off to a cable guy, or with wireless.
The cable guy again oversells bandwidth and puts in cheap, crappy
equipment to save on capex. Alternatively the ISP offers the cable guy a
share in the revenue generated by him, so that the ISP gets a relatively
fixed monthly income and the cable guy bears all the risks (and the low
profit margin). From the numbers I have seen, the cable guy breaks even
with sify if 90% of the ports are lit.

The ISP business is fundamentally like the telecom business, you sink in
a lot of money for the first few years, and then watch as returns keep
flowing in.

If I was to roll out an ISP, I would terminate fibre in the
basement of every building (or the ground floor), and have multiple
loops across buildings. Then run copper from there to the house. This
would cost about 8 to 10K INR/end node, and let me scale upto 100 Mbit/
node with a 1:4 overprovisioning ratio. However, this would have to be
done for a few thousand nodes for it to be cost effective (10 K nodes
approx for those numbers). Adding in routing and server costs would
roughly double that cost. Actual opex for me would be somewhere around
100 INR/node + bandwidth costs.

An E1 runs at about 6 lakhs/port/year or so, IIRC and an E3 for less/mbit.
If I was to offer 1 Mbit/node at an oversell of 1:10, I could sell the E1
to 20 people on the same switch.
This gives me a cost per person of 2600 /node/month, for an initial
investment of 20K/node.

An oversell of 1:40 (or 256 kbps at the same oversell) gives me INR
650/node/mth costing.

If I was to assume a high rate of interest so that 20 would double in 5
years, my costs/month for capex over 5 years would be around 700
INR/mth/node.

So the 1 Mbit/node/mth cost for 20 nodes turns out to be about 3300
INR/mth.

Now, whether I buy an E1 or an E3 depends on the number of concurrent
users and how much data transfer they are doing. By using a low data cap
(a days worth of data transfer at 256 kbps), I can achieve that 1:40
oversell and get far fewer complaints about speeds, since the heavy
users will stay away from my service. On a shared network, the light
users subsidise the heavy users given a fixed rate plan. By removing the
heavy users, I convert that money into profit.

However, that would reduce my opex over the long term (~ 15
years) by a very significant amount. The problem with investing such
volumes of money is that there is no guarantee of any returns. A pure
data service does not generate much interest. Voice and TV will drive
adoption though.

> I am not sure we have members here who are or represent ISPs, but at least DB
> used to work for one and can perhaps enlighten us. If commercial organisations
> do not see value in providing an ethical service, then we need to evolve other
> models. I find it hard to believe that voip/voice is the only missing element
> that can make or break a data networking service.
>
It isn't that voice can make or break the service. In India, the
penetration of data terminals is NOT sufficient to make it profitable
for an ISP which depends on heavily overselling bandwidth to actually
lay the last mile.

Voice, OTOH, offsets everything else because of the demand for the
service. The cost of the voice service basically covers a portion of the
installation and running costs over a few years. It is far easier to
change ISPs than telcos (until number portability happens, at least).
Hence, the telco can assume a relatively better period of customer
loyalty with a wired network.

With higher loyalty (implying more guaranteed returns), and a higher
demand for services, the profitability on voice services is far higher.

VoIP would transfer a large chunk of benefits to the ISPs, encouraging
them to actually invest in the last mile. Loop unbundling would achieve
something similar, where the investor gets returns as long as the end
user is using any of the providers they resell the loop to.

Being able to run television over IP would work too, but again, the
question would be about who pays for the set-top conversion box.

> I hope I am not repeating something everyone knows (or that I may have posted
> already) but there will be a conference in Dharamsala in October this year
> (summit.airjaldi.com) focusing on wireless data naetworking for rural
> development. Issues of FLOSS will be prominent - I do hope some of you take an
> interest and sign up to conduct workshops/lectures on practical matters that
> may help nascent rural network operators understand better how to create viable
> networks. Or even urban service providers, but so far wifi seems to be the sole
> preserve of a few fantastically priced hotels and airports.
>

Again, if I was to recommend an action plan, I would see if my building
society would agree to put in the cost of the last mile (~ 4K/node, and
a copper connection to the backbone at 100Mbit/s on the wire. That's not
100 Mbit/second of provisoned bandwidth, just the theoretical capacity).

The fibre connection would merely save the operators ass in the previous
case by preventing another massive outlay for upgrades when they need to
move to gigabit speeds (or from copper to fibre).

Then the society can buy an E1 from an ISP and get all the members to
pay for it. With 50 flats in the society, this comes to a mere 500
INR/mth/flat at 600000 INR/yr for an E1. However, if only a few people
want Internet access, then this simply will not work out (Loop costs not
included, that should be about 20K/yr or so, last I looked).

VoIP connectivity to the PSTN will simply encourage more people to join
in. As the number of people doing this increases, it will become more
profitable to just ignore the PSTN alltogether.

Devdas Bhagat