I have for many years measured Internet access service performance by measuring the transfer rate for a single HTTP download which is scheduled regularly.
Since moving to iiNet / NBN about 6 months ago, I have had difficulty reconciling apparent workstation performance with the measured download speed.
During investigation of variation in download speed measurements, it became apparent that in the absences of other network traffic, the first download was faster than further downloads within a short space of time.
The above chart shows the download speed on 10 downloads, each download ran for 5s and there was a 2s pause after completion before the next download was started. The file downloaded is one nominated by iiNet for download speed tests and is internal to their network.
The first transfer in this cluster ran at 5.8Mb/s (about half of the ‘headline’ access service rate). The following nine were between 0.9 and 3.1Mb/s, averaging 1.9Mb/s.
This behaviour is reliably reproduced on subsequent tests. It may be a result of the MPLS network, it may be some clever management of what you see vs what you get.
To better capture network performance, I have changed by measurement strategy to perform a 2s download, then wait 2s after it is completed, then perform a measured 5s download to calculate a more realistic download speed.
Above is a plot of a day’s tests using the adjusted algorithm. The median speed is 3.9Mb/s, less than a third of the ‘headline’ rate.
Little wonder that we need intervention as announced by the competition regulator, the industry cannot regulate itself effectively.
On 5/10/2016 we cut over to a new Internet access service, switching from Telstra 8Mb/s ADSL to iiNet NBN 12Mb/s.
Over some years, I have run an automated file transfer to measure the speed of our access service. The tests are done between 6:00 and 22:00, I am interested in performance during the times I want to use the service, and less interested in times when I would usually be sleeping.
Above is a graph of yesterday’s performance. Note that the service does not delivery anything like the 12Mb/s description of the service, and it collapses in the evenings when IP television demands exceed the network’s capacity.
Above is a chart of the tests over the previous week… the collapse in the evening is common. Continue reading Review of 5 months of iiNet Internet access
Our iiNet / FTTN NBN 12Mb/s service has been connected for just over a month now, and the first full month of download test results (ISP performance graphs) bears examination.
Above is a plot of the measured transfer rate of the test file from iiNet’s site. Continue reading First month of iiNet FTTN NBN Internet access
Modern data networks route or switch relatively small ‘packets’ of data across shared links that along with the switching nodes, form the wider data network.
One of the roles of the switches is to receive packets on one link, and send them onwards on the most appropriate link. Since links may be at different speeds, and many links may source packets to be sent on any link, there exists a mechanism in the switch to store packets pending transmission, in the simplest case it is a first come first served link queue.
The function of the queue then is to hold packets until they can be sent on the link, and to offer them in first come first served order. That raises two important questions:
- how long will packets be delayed;
- how many slots does the queue need.
Queuing theory gives us a method of estimating these quantities.
Lets make some assumptions about the traffic:
- service requests arrive randomly in time; and
- service time is exponentially distributed with an average time of 1.
Above is a plot of normalised average response time (service time + queue wait) vs resource (link) utilisation (pu means per unit). It can be seen that when the link utilisation is 0.5pu (50% busy), that response time is 2pu (ie twice the average service time), twice that needed to send an average packet at very low utilisation. Response time rapidly degrades:
- at 70% link busy, response time is 3.3 times packet transmission time; and
- at 90% link busy, response time is 10 times packet transmission time.
To ensure packets are not discarded, the queue need sufficient slots to hold packets even in most peak bursts. Whilst at 70% link utilisation, the average queue size is 2.3 slots (3.3-1), a larger queue size accommodates bursts better. Discarded packets can severely affect performance, not only are they likely to be resent after some delay and network overhead, they can break a higher level protocol unit in simple systems and waste the link capacity and other links used to send the rest of that protocol unit. Continue reading Queueing in data networks
Our iiNet / FTTN NBN 12Mb/s service has been connected for just over a week now, and the first full week of download test results (ISP performance graphs) bears examination.
Above is a plot of the measured transfer rate of the test file from iiNet’s site. Continue reading First week of iiNet FTTN NBN Internet access
The NBN is coming….
Our NBN street cabinet, or Node in FTTN. The pic is not crooked, the cabinet is out of level.
We received advice that NBN access is now available in our street… about 10 months after the street cabinet was installed.
This is the best hope we have had of escaping Telstra’s appalling service that we have endured for 6 years.
Above is a plot of the median download speed of the Telstra 8Mb/s ADSL service.
Above, the distribution for last week show that although HTTP speeds would appear to be capped at 2Mb/s on an 8Mb/s access pipe, the inevitable slide to poorer performance is shown where around 10% of the measurements are below 0.25Mb/s. Continue reading iinet – pre-connect experience
I mentioned that NBN is coming….
This NBN cabinet has been in place for about eight months, no electronics in it… but an important showpiece in the government’s desire to show that Malcolm Turnbull’s copper based NBN was on track. Empty cabinets probably persuaded some naive voters at the federal election earlier this month.
The deception worked, Malcolm was returned, albeit with barely the majority to allow majority coalition government.
NBN is a wholly government owned ‘enterprise’ operating on a ‘commercial’ basis. Who else could afford to spend capital on infrastructure that delivers no service.
Continue reading Another chapter in the NBN debacle – Jul 2016
The NBN was the Australian government’s response to a dominant carrier that would not engage in government’s attempts to guide industry to develop an equitable access network beyond that delivered off the telephony copper that existed in the street (the 2007 RFP).
That ‘initiative’ sees us return the good old bad old days when government owned most of the street infrastructure used to deliver telecommunications, and of course it has been in their interest as a provider to weaken consumer controls that otherwise gave some measure of protection of service levels.
The above graph shows the performance of my ADSL Internet access which reliably achieved greater than 7Mb/s when I moved here in 2009 but fell to around 0.5Mb/s. A recent upgrade has seen around 2Mb/s fairly reliably, though not enough for reliable IPTV, a stunning improvement on the previous 3+ years. Continue reading NBN is coming… it seems
I have been asked whether the Field strength / receive power converter can be used to solve a Eb/N0 (Eb/N0) design problem.
Eb/N0 is a method often used for specifying the relationship of signal and noise that will give adequate bit error rate in a data demodulator.
Whilst the calculator was not specifically designed for that purpose, and you cannot directly enter the desired Eb/N0, with the help of a hand calculator for simple calculations, a solution can be found. Continue reading An example of Eb/N0 design with the Field strength / receive power converter
I had cause to document Internet access performance recently, an interesting exercise.
When I moved into this house about five years ago, we enjoyed an excellent broadband service delivered on an ADSL1 link of about 800m to the RIM/DSLAM. The downstream rate was capped at 8Mb/s and most of the time, most of that speed was available to the end user.
For years I have run a test file transfer every half hour to document access performance, and when asked about recent performance that was an ideal source. I needed to go back to the previous week as last week started with an outage, the RIM batteries seem to have gone flat and Telstra had not been proactive in responding to the condition that left it running on batteries. Continue reading When governments tinker in telecommunications carriage…