My 3 year obsession with Broadband speeds
Mike Lyons Executive IT Architect, IBM
Since it first started to be rolled out the Australian National Broadband network have been controversial. Between the cost, technology compromises and the question of why does a government need to build such infrastructure in the first place there has been no end of public comment about the NBN.
While there a many emotional commentaries available on-line about the quality of the service I won’t be drawn on how I ‘feel’ about the NBN. I choose to do what all good engineers do which is to obsess over the objective data I can collect!
Telecommuting through the years
Since I have been a full time work from home telecommuter since the turn of the century internet speeds have had a huge impact on my work-life. Yes this means when I started on this journey I was connecting to my employer’s Remote Access infrastructure was via 56k modem. Just keep that in mind when you feel the need to gripe about working from home.
Over the last 20 years I have upgraded through just about every generation of xDSL as it was available, all over the same copper wires. So when the NBN finally came to my corner of regional Victoria it was Fibre to the Node (FTTN).
Not surprisingly my early experience was a little disappointing (or character forming depending on your point of view). So early in 2018 I took delivery of a Telstra NBN modem and set about customising it for my home network such adjusting the DHCP setting and so on. The fancy web console allowed me to notice that while I had signed up for the standard 25Mb/s service the VDSL was actually sync’ing at more like 60Mb/s. So I signed up for the 50Mb/s service and was pleased to find this worked pretty well for the first week. Then the drop-outs started!
From this point in time I became a regular on the NBN service desk line and also have a small pile of replacement modems that have been shipped to me to resolve on-going issues all of which seemed to centre around (you guessed it) the quality of my copper connection. So by this time I had started collecting as much objective data as I could.
Give me data
The first thing to sort out is what I might expect from the actual copper service I have. Finding the brand new Alcatel VDSL node on the street wasn’t hard and pacing the distance to my house I estimated around 300 metres of copper connected my model to the NBN node. On paper this isn’t too bad and reading around I should be able to get close to 100Mb/s down-load and if the signals on the wire were that good, why was it dropping out so much?
Another thing that I noticed immediately was the Radio Frequency (RF) interference that the modem was causing. This appears as a set of persistent and irritating signals all across the HF radio spectrum and suggests that a lot of signal energy leaks out of the modem and the copper wires. Sadly this is not a big surprise as the copper wires where only every really intended for traditional phone services and the fact that VDSL works as well as it does is a technical marvel.
OK, so what does my internet speed actually look like over the last 3 years? The red line represents the sync speed negotiated by the modem. So this data shows approximately 80,000 rate changes over the 3 years of data.
So the first conclusion I drew for the graph is that my internet speeds are distinctly seasonal! In colder months it objectively performs better. The data here has some big gaps in it where it performed very well (ie the full 50Mb/s) for months at a time and I didn’t bother to collect the modem logs.
By now you are probably wondering how I got the data in question. Will the Telstra NBN modems are a linux machine and you can navigate to the diagnostic page and save the system log to a local file. Sadly I have not as yet been able to forward the syslog data to an external collector so nearly every day I have been logging on in the morning and grabbing the current log as a text file. The modem seems to be able to store about 4.5MB of log data and this typically covers between 2 and 3 days of log data.
Each speed change is logged with a single line of text and a date-stamp so it isn’t hard to build a time based graph that shows my down-load speed varies from 50Mb/s to as little as 5Mb/s. What is really striking however is that it has often stayed at 50Mb/s for months are a time during Winter. Conversely Summer brings lower speeds and more instability.
So what’s going on? I needed more data!
Having a detailed look at the options on the modem there is a great panel under the advanced tab of the modem’s home page show here.
This reveals a great deal of information that is worth discussing. Firstly the Maximum line rate is a lot higher than the selected line rate, but the important things are the noise margins, the line attenuation and the Output power. For a communications engineer these parameters are the very foundation of our telecommunications infrastructure.
The noise margin shown here (around 12dB) is typical for my service, but it wasn’t always that way. When I first got the NBN this number was around 6dB which was pretty standard however while trouble shooting my early performance issues the NBN help desk applied a “stability profile” which increased the margin to 12dB or so.
While I was able to view this using a browser I didn’t have the patience to write a script to extract this data automatically so I went back the daily log data I had collected to see if I could find these parameters recorded there. Now one of the advantages of the Telstra NBN service is they do regular software updates on the modem and over the 3 years I have to say the stability has improved as a result of these updates however this doesn’t change the nature of the copper wires over which the VDSL signals run.
One particular software update in November 2019 was significant enough for the software version reported on the home screen to change from v17.2 to v18.1 and this also changed the format of the log data so that I was easily able to grab signal data each time the modem reconnected the VDSL signal.
So a ‘grep user.notice xdsl’ yielded a line of data like this….
There were 799 such events recorded in the 2020 log data which should be enough to provide a statistical view of the behaviour of the VDSL service. So I built a table of all the data and used a graphing tool to create a bunch of scatter plots to find any correlations between the parameters.
I tend to find this approach a good way discovering the things that matter in any trouble shooting so any scatter plots that show strong correlations (ie reasonably obvious lines) tend to show useful attributes and lines with bends in them are even more useful (something for another blog). Of all the graphs I was able to produce this one leaped out at me.
This graph that maps the signal power down from the node and up from modem verses the reported line attenuation showed a few really important facts about my VDSL service.
1. The line attenuation reported be the modem varies by 10dB, this is a huge variation.
2. The VDSL node responds to this line attenuation by turning up the power, but stops at 9dBm so it is only able to respond to the signal loss up to about 25dB. So the loss on the line is more than the VDSL node can routinely manage.
3. There are more reconnection events recorded when the attenuation is higher.
One word of caution with the data reported in the logfiles is that it doesn’t seem to line-up very well with the real time status screen. For example the power up and down fields appear to be reversed on the status screen and the log data never records a value for some of the fields. Most importantly the Attenuation data reported on the status screen doesn’t match up at all with the log data.
Anyway I already suspected that the VDSL performance was seasonal so I plotted the attenuation data over time and this is what I got.
Attenuation over time
This attenuation over time graph clearly shows that the line loss is much lower during the winter months and this fits with my anecdotal observations of better performance.
So this leads to my hypothesis that as the ground above the copper gets damper it provides a better signal shield in some way that reduces the loss. Sadly this is not something that I can control, but at least provides an indication of what seems to be the major factor in my broadband speed. The data also implies that if the VDSL node could deliver more power it might well be able to overcome the significant attenuation during the hotter months.
Final Note
For others who might want to try a similar analysis I would suggest finding a way to interrogate the status screens via HTML as there is real time data for Cyclic Redundancy and Forward Error Correction which I have not attempted to analyse here.
Best of Luck!
