[AusNOG] Average Usage Per Service
Tony
td_miles at yahoo.com
Mon May 12 14:18:09 EST 2014
On Mon, 12 May 2014 13:08:51 +1000, Skeeve Stevens
<skeeve+ausnog at eintellegonetworks.com> wrote:
> Yes, I am looking for a figure that providers might use in planning
> capacity. In a network where congestion isn't an issue is what I am
> looking for.
>
> Transit scaling is not actually what I am looking for here, but mainly
> scaling of throughput capability of the network routers/LNS.
>
> Re Transit/Backhaul you need to start with a certain minimum regardless
> of
> contention ratio... you can't have 10 x 100mb users (1Gb) at 50:1
> (residential average) and get away with 20mb. In that scenario you'd
> need
> a baseline of 80-150mb, realising that a couple of users could smash the
> contended parts of your network. It also depends on the types of
> services
> such as Private networks which will affect backhaul (AGVC) but not
> transit.
>
Which is actually a rather large problem for someone starting an ISP from
scratch. There is a lot of flexibility in transit providers in scaling
links up and down or paying for what you need (eg. 95th percentile), that
you could quite easily put in a Gbps connection and just use what you
need, so start at 100M and wind it us as required. The ongoing cost of the
physical connection is likely to be the same for 100M or 1Gbps (usually
the cross-connect costs as you would choose a transit provider in the same
DC as yourself).
Your first 100/40 NBN customer is going to require 100Mbps of bandwidth
all the way through the network (unless you want to sell congestion 100%
of the time to this customer, not just during peaks) and your 2nd 100/40
customer will likely require you increase that to at least 150Mbps (so
they could both get 75Mbps down at the same time, not allowing for any
other users).
> Transit costs are actually cheaper than most AGVC pricing out there at
> the moment but overall they are easy to upgrade.
Yes, with NBN AGVC costs being about TWICE the costs of transit the choke
point is more likely to be at the NBN location, than transit (which is
really rather sad I feel).
>
> So I am wondering based on some mix of speeds, how far will a Cisco 7201
> (1GbTP) get you these days, or a Cisco 1002-F (2.5GbTP) to bootstrap the
> ISP into viability.
>
Given a 7201 will max out at 1Gbps, do you really want to limit the new
fledgling ISP to this sort of speed. You're really putting in something
that is definitely going to need a forklift upgrade in less than 12 months
if the new ISP is in any way successful (ie. more than 100-200 users). I'm
assuming the micro-isp is going to want to be successful and have more
than this number of users ? Any less any the number don't make sense.
If you are planning on having AT LEAST one customer that could
theoretically choose a 100Mbps NBN service, then you need AT LEAST 100Mbps
of AGVC. Given NBN AGVC is about $40/Mbps/mon and then you might have at
least another $20/Mbps/mon for transit and other goodies, the minimum cost
for your FIRST 100Mbps customer is $6000/mon. Given this is a rather
massive cost in itself I find the question of a 7201 or baby ARS rather
pointless as the difference is only about one month at $6000/mon (ie.
delay starting your business by one month, buy a 1002-F instead of a
7201). Even if I'm wildly wrong with my AGVC costs and it's only "the
same" as transit, that pushes your break even of buying the ASR from 1
month to 1.5 months.
With the above costs it's also fairly apparent that "starting a small ISP
from your garage with a couple of cheap routers" is a thing of the past.
There is no "start small and grow" option any more. You need some serious
financial backing and be prepared to bleed cashflow for a number of months
if you're willing to start an NBN ISP (unless you somehow happen to have a
customer base of hundreds from day 1).
This all assumes you are starting an actual physical ISP and not simply a
"reseller" arrangement (where you have no infrastructure of your own, you
just do sales/billing). This is what I think you were asking about though.
In terms of your actual question of average usage per service I don't
really have any idea for residential type services, hopefully you might
get a reply from someone who does have this info.
ABS has stats for Dec 2013:
http://www.abs.gov.au/ausstats/abs@.nsf/mf/8153.0/
Which gives volume of data download by all "fixed line" users for the
three months as 823,421 TB. Dividing by 3 gives 274474 TB/month.
The same stats give the number of fixed subscribers as about 6 million
(mobile users is roughly the same - 6 million). Dividing one into the
other gives about 45GB per user per month.
This is an average, but as long as you have an average user base, then you
should get results that approximate the average usage.
It turns out this works out at about a constant 150kbps per user (or there
abouts). It probably makes sense to multiple this by at least 4 to allow
for peaks of use (ie. 600kbps per user), but that is just a guess on my
part based on historical trends.
This is all rather academic though as it's based on:
a. historical data (mostly DSL/cable users)
b. really large data set (good averaged)
c. some rather large assumptions
As a startup ISP the peaks are more going to be driven initially by the
customers that you attract (based on pricing and target market) than long
term historical data. Once you get a decent base of customers it starts to
average out.
One thing I would point out that makes it still somewhat relevant is the
small number (% wise) of new connections that are NBN fibre now and going
forwards (at least this week). By the end of this year it would seem
likely that we'll just be seeing more 25-50Mbps FTTN/B & HFC connections
than anything really much faster. This would indicate only a small
increase in bandwidth requirements, compares to 100/40 (or faster) fibre
connections.
regards,
Tony.
(apologies for the rambling)
More information about the AusNOG
mailing list