[LINK] Advertised NBN plan speeds impossible to achieve: ACCC

Glen Turner gdt at gdt.id.au
Sat Nov 9 10:49:40 AEDT 2019


Antony Barry wrote:
> https://thenewdaily.com.au/life/tech/2019/11/07/nbn-speeds-accc-measuring-broadband/

If seems to me that the ACCC is trying to redefine a term widely
understood in the industry. It wants the speed of the service to
reflect the usable bandwidth of the service, not the minimum bit-clock
of the parts which could be used to build that service.

For example, 100Mbps ethernet does not transfer 100 megabits of user-
accessible data every second. Ethernet's clocking preamble, protocol
header, CRC, and inter-frame gap all consume bandwidth.

The not user-data components are:
 - ethernet clock establishment preamble: 8 bytes
 - ethernet protocol header: 14 bytes
 - ethernet CRC: 4 bytes
 - ethernet inter-frame gap: time equivalent to 12 bytes
so overhead = 38 bytes.

 efficiency = (data)/(overhead + data)
 equivalent_bandwidth = clock_rate * efficiency

So we end up with an equivalent bandwidth of 2.5Mbps for a 1 byte
payload and an equivalent bandwidth of 97.5Mbps for a 1500 byte
payload. ACCC would of course require the lower of those in the service
description, which it itself a weird result.

NBN RSPs provide an *Internet* service not an ethernet service, so
there's a little more overhead on top of basic ethernet:

 - NBN overhead, two 802.11ad headers: 8 bytes
 - ISP PPPoE header: 2 bytes
 - IP packet overhead, IPv6 is larger so: 36 bytes

This brings per-packet overhead to 84 bytes. That is, throughput of
worst-case 1 byte packets is equivalent to 1.1Mbps across a 100Mbps
clocked transmission.

Of course a more realistic measure would be to use a typical mix of
traffic sizes. But there's no such single distribution: the same
service could be used for terminal access, for data transfer, for
streaming, for gaming. So that same 100Mbps service would need to be
marketed as "8Mbps gaming" or "94Mbps data transfer".

I hope this makes it plain to readers why the industry uses the bit-
clock to describe parts, and thus describes services using the minimum
bit-clock of the parts which could be used to build that service. I
hope it's plain why the ACCC diverging from this practice will give an
undesirable outcome.


I do understand the ACCC's desire for a one-number metric to measure
ISPs services. But I can say after spending a working life achieving
high performance networks that a single metric is not useful,  and that
desirable metrics are based upon categories of end-to-end performance,
not link performance.

Best wishes, glen






More information about the Link mailing list