[LINK] A non-sensationalist look at Australian internet speeds

Paul Brooks pbrooks-link at layer10.com.au
Tue Mar 29 13:38:29 AEDT 2016


What I'm getting at is that, over the course of a month or 6 months, the average
broadband link utilisation is less than 1%. Sure there are instantaneous peaks when
somebody is actually trying to do something, but most of the time the link is idle.
If the average link was (say) 10 Mbps, then most of the time it might peak to 2, 4 or
even 10 for a few milliseconds, seconds or even minutes, then falls to zero or
background noise most of the time.
The average over a month is almost never 2 Mbps, these days its still less than
300kbps for most people.
Even watching video is not a steady stream - if you actually watch the traffic of a
nominal 2 - 3 Mbps video stream, it consists of a sequence of bursts up to full
line-rate for a fraction of a second (shorter or longer depending on your actual
bandwidth line-rate) punctuated by 3 - 5 seconds of idle - when anything else can jump
in and have no effect whatsoever on the stream, or the anything else.

There are plenty of time windows for devices in the household to do much much more
without impacting on anything else. And the vast majority of applications are not
perturbed significantly even if there are multiple things trying to happen at the same
time, which cause the link to drive to full capacity for a few seconds - increase the
number of phones/tablets/laptops accessing email from 1 to 40, and they'll never be
aware of simultaneous access congestion, because at all happens in the background and
the only visible effect is that a message might take 2.6 seconds to appear on the
screen instead of 2.3 seconds (for example) if it was the only device in the house.
Total aggregate data transferred through this example link would jump from 1x to 40x
(roughly), without any discernable requirement for extra bandwidth.

And all this breaks down on low speed links that some are lumbered with, around the 1
Mbps level, when many applications can drive it to 100% utilisation for many minutes
or appreciable fractions of an hour - THEN you'll notice the congestion effect of
trying to do more things at the same time, and look for extra bandwidth before you can
play a bigger part in the data volume statistics - but these don't form a significant
fraction of the ABS stats.


Note I'm definitely NOT saying there isn't a case for increasing most people's
bandwidth - all I'm saying is that data volume can rise greatly without bandwidth
increasing significantly. Its definitely not a linear relationship - its logarithmic,
and can be modelled using queueing theory and Ehrlang equations. Confusing the two
things, or assuming that you have to double one to double the other, is a common fallacy.



On 29/03/2016 9:14 AM, Karl Auer wrote:
> On Mon, 2016-03-28 at 23:36 +1100, Paul Brooks wrote:
>> Except that's not what the ABS stats measure or show  at all. 
>> The ABS measures data volume transferred not link capacity or 
>> bandwidth - these two aspects are only loosely related with each 
>> other. Data volume can increase by many times without link bandwidth
>> changing at all.
> I'm not sure what you are getting at.
>
> As long as the volumes being transferred do not exceed the available
> bandwidth, you are right. But as soon as the aggregate volume being
> transferred approaches or exceeds the available bandwidth, the data
> volume is effectively capped. Available bandwidth puts an upper bound
> on data volumes. The relationship is only "loose" as long as the data
> volume is not soaking up all the available bandwidth.
>
> So if we say (simplistic example here) that the average link is 10Mb/s
> and the average data volume is 2Mb/s, we don't (on average) have a
> problem. Our data volume can double twice before we have a problem.
>
> But it CAN'T double again, because the capacity is not there to let it.
> We will be constrained by the available bandwidth.
>
> Talking about averages is also tricky. There are plenty of people who
> are already being constrained by their available bandwidth, and saying
> that "on average" we are all doing fine really doesn't help them. The
> same is true for those with data quotas, who are suffering an artificia
> l constraint on their data volumes. These people are presumably not
> reflected in the stats because their data volumes are being capped at
> less that what they actually need.
>
> Regards, K.
>




More information about the Link mailing list