One of the more frustrating aspects of broadband for consumers is that performance doesn’t always live up to the advertised “up to” bitrates and it has some people up in arms.  Consumers generally see the advertised number and assume that this is the guaranteed minimum speed that their broadband should perform at when the reality is that “up to” doesn’t mean “at least”.  So while the advertising is technically accurate, it may still be perceived as misleading.  And when consumer expectations aren’t met, it leads to customer dissatisfaction and mistrust of the ISP.

The problem for the broadband industry is that no one wants to unilaterally advertise their network’s actual limitations because that would be handing a free marketing advantage to their competitor.  What is needed is an industry wide standard that everyone agrees to such that the playing field remains level.  The question is what would such a standard look like and how should broadband performance be characterized.

The first thing that needs to be understood is that broadband is a pool of bandwidth shared between many people, and it is shared because that’s the only way it can be affordable.  More importantly, consumers get a much faster broadband connection per dollar than a dedicated circuit.  For example, a dedicated T1 circuit operating at 1.554 Mbps costs more than $200 per month which means each Mbps costs more than $128/month.  A dedicated DS3 circuit operating at 45 Mbps costs $2000 per month which means each Mbps costs more than $44/month.  By comparison, a 6 Mbps broadband connection typically costs $35 which is less than $6/Mbps and that’s 20 times cheaper per Mbps.  Even if it only performs at 80% of its advertised speeds, that’s still a great deal and it will be perceived that way so long as the customer is clearly informed of the limitations of their service.

The challenge in explaining the limitations of broadband to consumers is that broadband performance is a very complex issue and figuring out how to do it without overwhelming the customer with too much information isn’t easy.  There are many factors that determine the performance of broadband services and some of the key factors are:

• What is the usage cap (implied or explicit).  While this is seen as a negative aspect of broadband, lower usage caps do lower average network congestion and reduces the cost of broadband services.  Satellite and wireless broadband services depend heavily on low usage caps because of bandwidth scarcity.
• Where does the most sharing occur?  Does it happen on the “last mile(s)” which connects the home to the broadband provider’s equipment or does it happen further upstream?
• How oversubscribed is the system on the most oversubscribed portion of the network?  In other words, how many customers are sharing each Mbps of bandwidth.
• Is the bandwidth shared with other services e.g., IPTV or Video on Demand television services?  From a performance standpoint, this may be just as much of a factor as oversubscription between the neighborhood.  These IP based television services can consume more than half of the total advertised bandwidth though it varies from provider to provider.
• Are the advertised “up to” speeds referring to the sync rate or data throughput?  The sync rate is the raw bits per second a network connects at, but it carries some overhead with it.  The data throughput is the sync rate minus the overhead which is typically 10% to 20%.  It’s sort of like a 10 lb water melon where the melon weights 10 lbs but there’s 1-2 lbs of peel and seed that you can’t eat.
• What percentage of time does the network become congested aside from self induced congestion.  In other words, I accept the fact that I can cause 100% congestion on my own service by maxing out my own connection but what I want to know is how often my fellow broadband subscribers on the same service use up all the shared resources.
• What is the typical packet delay in terms of latency and jitter aside from self induced delay.  In other words, I accept the fact that my own usage of certain applications like P2P can cause massive amounts of jitter, but how much jitter can my fellow broadband subscribers cause for me and how often does this occur.  Note that jitter caused from fellow broadband subscribers is usually not an issue because it mostly confined to each subscriber’s connection in the case of wired broadband, but it can occasionally happen further upstream in the network if there is excessive amounts of congestion.
• Does my broadband provider implement some form of fair bandwidth sharing scheme between the subscribers and if so, how do they implement it?
• Does my broadband provider manage congestion and jitter on the network and if so, how do they implement it?

Answering all of these questions may be difficult and gathering all of the information might not be feasible.  Even if this were possible, it will still be difficult to put all or even part of the information on a simple banner though it might be possible in the fine print.  The problem is that most people may not read the fine print much less understand it.

Computing the average achievable performance of broadband

Ideally, broadband providers would simply advertise the average achievable performance of their broadband service but computing such a number may not be easy.  We could not measure the customers who only want fixed bandwidth for audio or video streaming, gaming, or real-time applications so long as those customers get the bandwidth they’re asking for with the proper latency characteristics.  The question is for customers who are trying to download things as quickly as possible, what is the actual throughput they’re getting compared to the advertised bandwidth throughput?

We can’t simply look at the total throughput of individual subscribers because those customers might be using peer-to-peer (P2P) applications which are relatively immune to the Internet’s congestion control mechanism.  P2P applications have the ability to receive maximum bandwidth even when the network is congested but this comes at the expense of other customers.

It might be possible to monitor and collect data on the throughput of the fastest individual Transmission Control Protocol (TCP) flow going through the network since that would effectively be equivalent to a speed test that people run on websites like speedtest.net.  This would give us a fine-grain performance history of achievable speeds on the network and it would be much more granular than random samples from broadband customers checking speedtest.net.  While these performance numbers would suffer at the hands of P2P customers who are taking a larger share of the network than their neighbors, the results would reflect real world conditions for most people who aren’t using P2P.

If the ISP employed some sort of weighted TCP mechanism that ensured equal distribution of bandwidth between subscribers rather than between TCP flows, monitoring the fastest TCP flow would still be valid.  If however an ISP deploys some sort of bandwidth fair sharing scheme where light usage subscribers are prioritized over heavy usage subscribers, we would no longer be able to use the fastest TCP flow since fair sharing artificially boost the performance of some TCP flows at the expense of other TCP flows.  To accurately gauge the overall average achievable performance of the network, the boost given to light usage subscribers over what they would get under an equal bandwidth scheme would need to be factored out from the performance report.  Low network traffic times would also need to be filtered out since the fastest TCP flows during those times may very well be zero yet the network would obviously be capable of delivering excellent performing since there is little or no congestion.

The other factor that must be accounted for is the bandwidth consumed by IPTV and Video on Demand since some broadband providers use out-of-band capacity to deliver those television services while others don’t.  The average bandwidth reallocated from broadband data capacity to television services must be deducted from the network’s performance history.  There might even be additional factors to consider and there may be better ways to compute the achievable performance averages but that is for the best minds in the industry to figure out.  If it is feasible to gather and process this data, ISPs would be able to advertise a more realistic performance number that consumers can expect.

These aren’t easy questions to answer and the solution will require at least every very major broadband provider in the country to meet and work out some sort of industry wide standard.  Once the standard is agreed upon, broadband providers could come up with some sort of industry certification/logo and all participants will have to abide the same rules.  Without some sort of multilateral agreement, the effort is doomed to fail.  The benefit of managing customer expectation with accurate performance claims is too important for the broadband industry to ignore.  Better transparency and better customer satisfaction goes a long way towards earning the trust of consumers and regulators.

Definitions

Mbps – Megabit per second.  Note that the term “megabit” is used loosely and it actually means a million (1,000,000) bits rather than the strict (but confusing) computer science definition of 1024^2 (1,048,576) bits.

Delay – The time it takes an individual packet to travel through a network. Packet delay should not be confused with file transfer time, which correlates to bandwidth. There are two forms of delay that affect packet transfer times: latency (the time it takes a bit or packet to traverse a noncongested network before arriving at its intended destination) and jitter (a measure of the variation in packet delay).

Latency – A simple measure of delay, which is largely dictated by the speed of light in glass and the physical distance between two devices on the Internet.

Jitter – The measure of the variation in packet delay. High-jitter conditions are essentially micro-congestion storms that last tens or hundreds of milliseconds. High jitter occurs whenever a large number of packets come from a faster network link to a slower network link or where several networks links merge to a single link. When this happens, network devices such as routers and switches get backlogged, and they force packets to wait inside their memory buffers, increasing the time it takes packets to traverse a network.