Ironically, I remember reading some American cable dimensioning data that said *precisely* that: 3 users on the top tier.
Be interesting to see that scaling information. I sincerely hope there weren't 2000 other modems in the service group though! I don't think that would work, let alone work well.
I've found my copies of the 2 documents I read this in; let's hope that they say what I think they say...
Both came from Arris, and seem to describe the same network sizing information. They seem to date to around 2014.
1. "NIELSEN'S LAW VS. NIELSEN TV VIEWERSHIP FOR NETWORK CAPACITY PLANNING"
Rule of Thumb Approach for Network Sizing
To date some MSOs have sized their network on a method of multiplying the billboard speed by either doubling (2X) or tripling (3X) the billboard speed to determine the amount of DOCSIS capacity per service group, this is sort of a Rule of Thumb method for DOCSIS Network Sizing.
The document goes on to explain that, as service groups get smaller, that "rule of thumb" might end up allocating too much bandwidth.
This document doesn't give a basis for *why* 2x or 3x is chosen, but does explain a better method for the future.
The next document describes the same future, but has more detail for the past...
2. "IS NIELSEN READY TO RETIRE? LATEST DEVELOPMENTS IN BANDWIDTH CAPACITY PLANNING"
Many MSOs transitioned to the (Consumption+Tmax)-oriented Period roughly around the 2009 time-frame in conjunction with the MSO transition to DOCSIS 3.0 channel-bonding. This is where we are today. In this period, The SG Aggregate Average Bandwidth roughly equals the rate of the highest service tier identified by the DOCSIS value: Tmax.
This has allowed traffic engineers to use a very simple rule to estimate the size their High Speed Data (HSD) networks:
Required Service Group Bandwidth Capacity = 2xTmax (1)
The Rule of Thumb can often be used as a quick guideline for estimated capacity needs. However, MSOs have developed tools and optimizations based on volumes of empirical data and traffic analysis that allow them to engineer networks for high quality of experience as key variables change.
This Rule of Thumb has been morphing over time. It was originally 3X and has been shrinking with time (and smaller Service Groups, SG).
I read the first part of this quote to say that Tmax is the speed of the top tier they sell (the "billboard speed" of the previous document). And that, at the current time, it just so happens that average aggregate bandwidth of an SG
is about the same value as Tmax. The calculation for the required capacity then comes from Tmax once (as a reflection of the average bandwidth) and from Tmax a second time (as a reflection of one top-tier user running a speedtest).
If the old rule-of-thumb used a 3x factor, it hasn't changed to 2x because usage has lowered. The reduction is because the size of the SG has reduced.
The document then goes on to develop the rule-of-thumb to reflect this by directly including the size of the SG:
Required Service Group Bandwidth Capacity = S*Tavg + Tmax (2)
where S is the number of subscribers within the Service Group, Tavg is Per-Subscriber Average Busy-Hour Bandwidth, and Tmax is the Maximum Sustained Traffic Rate (Billboard Bandwidth) offering to the subscribers.
The document goes into more details of the earlier an later periods, when other calculations were used for network sizing.
Seems a tad arbitrary, takes no account of the size of the service group or the usage patterns.
You are right. It looks like it depended on appropriate scaling of service groups - which very much depends on the investments being made by individual cable companies.
The papers undoubtedly attempt to point at the fact that this was an old "rule-of-thumb" that, while it worked for maybe 5 years, was no longer appropriate.
It isn't at all obvious how well this relates to VM in the UK...
At the time I read it, VM were selling a top tier of 152Mbps with 8 bonded channels (440Mbps); with my limited knowledge, I found the rule applied well.
But I now understand that the service group (of the time) was likely to have 12 channels for the whole service group, from which each modem would bond a subset of 8. The service group bandwidth would then be 660Mbs.
If we believe that VM themselves worked in the (Consumption+Tmax)-oriented Period
, then they would be leaving 500Mbps for average consumption across the service group.
If average busy-hour consumption at the time was 1Mbps per subscriber, then that would need VM to have been working with service groups of around 500 homes.
Does any of that sound appropriate?