< Back

Broadband Speed - Why it Doesn't Matter (so much)!

Thu, 26 Jun 2014

networking

There’s continually much press about the speed of UK broadband. Just last night the BBC television programme Watchdog had a segment on the misselling of broadband speeds – claims by an ISP that users would achieve 50Mbps and beyond speeds, but achieving “only” 30Mbps in practise. Users were complaining they still had buffering on their streaming and that the speed increase hadn’t made a difference to their service quality.

Read most articles entitled “which broadband should I choose” and they calculate value of the package based on the cost and bandwidth (largely synonymous with the term “speed”). Television adverts from ISPs tell us that packages are now “higher speeds” for “faster downloads”.

The question is: do most households really need that 50Mbps advertised speed? Will it really make BBC iPlayer or Netflix any better quality? What does the quoted “speed” even mean?

Network Bandwidth

Firstly, let’s define what is meant by bandwidth: this is the available capacity of a circuit to carry data every second. For example, a circuit with 10Mbps bandwidth can carry 10 million bits of data every second, or in more conventional units, 1.25MB/s (that’s megabytes per second, since every byte contains 8 bits).

So if you have a file that is 100MB in size to download over 10Mbps bandwidth, the best possible download time is 100MB / 1.25MB/s = 80 seconds.

That is the best possible result for a line which is tested to pass traffic at 10Mbps – but there are plenty of other factors which can slow this down. Unfortunately, the (maximum) available bandwidth is the only metric ever quoted by broadband ISPs.

Contention, Congestion and Packet Loss

A quick back of envelope calculation tells us that an ISP selling 50Mbps broadband with 100000 customers (and that’s small for most of the large players!) requires 5000Gbps (or 5Tbps) of overall connectivity to service all those customers at their contracted speed at the same time.

That’s a ridiculous figure, especially when we consider that on the largest UK peering exchange (LINX), British Telecom (BT), Virgin Media and Sky each have between 250Gbps and 300Gbps of available capacity at the time of writing. This is just one connection between those networks and others on the Internet (such as the BBC, Netflix, Amazon, Google etc.), but it shows that if those ISPs were selling 50Mbps circuits with guaranteed bandwidth, they could only serve up to 6000 homes via LINX – and at a monthly cost on LINX in excess of £300k, this is clearly not profitable at prevailing broadband prices!

On top of the costs to reach other networks, there’s the costs of connectivity to the local exchange and finally to individual homes to consider. Those are over large distances and so are an even larger cost factor to consider, making broadband even less profitable to deliver perfectly. We conclude that:

  • There must be a large aspect of contention across the network: hundreds of thousands of users all online at the same time are sharing the available capacity of the connections from their local exchanges to the ISP’s core network, and from their ISP to everyone else on the Internet.
  • At peak times, this contention can lead to congestion as everyone is fighting for a share of the ISP bandwidth which isn’t actually available. This can lead to the “slow speeds” experienced during peak periods (typically evenings and weekends), and causes some ISPs to implement “fair usage throttling” during those periods to reduce the impact of the heaviest users on everyone else.
  • That congestion leads to packet loss – meaning that data which should be sent or received reliably in fact doesn’t make it to its proper destination. For example, if you’re watching BBC iPlayer and experience packet loss, some of the programme will not download correctly and the video may stall.
  • Of course, also if your broadband ISP has a fault or outage, then you’ll get no service at all – that’s a 100% packet loss scenario, and then you’ll have zero Internet access.

An ISP can have the largest bandwidth, but if little of your data can pass due to congestion, it’s useless!

Packet loss: less reliable data transfer and slower observed “speeds”

Latency

This one is often neglected entirely, but noticeably affects “speeds”. Latency is the time for data to travel between two points on the Internet. It might also be called a delay. As data takes more time to travel longer distances, it follows that downloads from the USA to UK will take more time than a download between two places in the UK. For example, compare Facebook in North Carolina with Google UK over an ADSL circuit:

Latencies: difference over ADSL between Facebook (UK-USA) and Google UK (UK-UK)

Latency is typically measured in milliseconds, with a typical transfer between data centres in the UK being 1-3 milliseconds, and between data centres in the UK and New York being 70-90ms. A typical ADSL connection adds another 15-25ms of additional latency between the home and the ISP’s core network.

Regardless of the available bandwidth, this affects the speed experienced because the higher the latency, the more delay there is in retrieving that data. The Internet requires that data is split into small chunks (called packets) for transmission, with all the packets needing to be re-combined to give the original data when it arrives at the destination. Many protocols (those based on TCP to be technical) which reassemble the data require that the packets are sent roughly sequentially, and that they don’t get too far behind if they get “separated” from each other en route. This involves a process of sending a request and awaiting a response, before sending another request (the size is known as the TCP window). The latency causes a delay between all these request/response acknowledgement processes, which considerably slows down the transmission of the data.

A high latency connection - such as over an older ADSL line compared with a fibre-to-the-cabinet (FTTC) or even better fibre-to-the-premises (FTTP) - will appear to be slower even at the same bandwidth.

Jitter

The jitter on a circuit is the typical variation from the expected latency. A network with constant latency has zero jitter, and a network where the latency can vary wildly has a large jitter. This is technically called packet delay variation (PDV).

When downloading files, jitter is unlikely to be noticeable (other than the increased latency causing more delays – as discussed above). However, for realtime services like voice-over-IP (VoIP) and streaming (BBC, Netflix, Spotify), jitter can cause havoc: a VoIP conversation will degrade quickly with speech being lost or broken, and a streamed programme which is downloading fine at first may suddenly start buffering intermittently, and of course always at the most inconvenient of times!

So What Does Speed Really Mean?

The perceived speed of a broadband circuit is the real time it takes to transfer specific data to/from a home and somewhere specific on the Internet. As explained above, this involves not only bandwidth but also packet loss and latency.

A typical ADSL bandwidth of 5Mbps is actually ample to view one BBC iPlayer HD programme or stream music or watch YouTube videos, and concurrently surf the Web and check e-mails. As an example, BBC iPlayer HD streaming requires up to 3Mbps of capacity. For a small household therefore, a 5-6Mbps ADSL package is going to be ample. A larger household with a few concurrently active Internet users may want to take a 10Mbps or higher service for the added available capacity.

What usually leads to the ever-annoying “buffering” effect is actually congestion, packet loss and jitter (collectively, quality of service) and has nothing to do with speed. Therefore spending more money on a higher speed broadband package isn’t the magic bullet to improve your experience unless your ISP also improves its quality of traffic forwarding.

The annoying “buffering” sprite going round, and round, and round…

Fibre, and most commonly fibre-to-the-cabinet (FTTC), services should help to decrease latency and noise on the line (which causes jitter), so should give a better quality.

We all know the saying: it’s quality vs. quantity!

How Does IP Transit Differ?

In data centres, we consume and supply a service called IP transit. This is bandwidth on a carrier-grade level with guaranteed throughputs, very low occurrence of outages (100% uptime targets) and stringent service level agreements (SLAs) surrounding packet losses, latency and jitter.

These are very specific guarantees which ensure that critical traffic reaches its destination every time, and the most efficient manners. For example, one of our upstream carriers NTT publishes service levels of:

Intra-EuropeTransatlanticEurope-Japan
Latency35ms90ms300ms
Packet loss<0.1% monthly<0.3% monthly<0.3% monthly
Average jitter<0.5ms<0.5ms<0.5ms

An IP transit service may appear more expensive than home broadband connectivity, but in reality it’s backed by reserved capacity, SLAs and importantly assurance that all efforts are made to move the traffic reliably and efficiently. This is true quality, and in fact quantity too – we can supply IP transit up to 10Gbps within our data centres.

Our Advice?

The next time you consider a new broadband package, perhaps it’s worth asking not “how fast is it?” but instead:

Ask “what will the average packet loss and jitter be?” and “What frequency and duration of outages can I expect?”

No doubt that’ll confuse the salesperson!