Price should not necessarily be the one deciding factor when choosing a provider (as our private broadband testing over the years for various clients has shown). The way some providers cut the monthly costs and remain profitable is by oversubscribing the services and running very high contention ratios. This is akin to a car salesman selling you a vehicle for AU$13,995 with a speedometer that reads up to 400km/h but the car only has an actual physical top speed of 120km/h. The car salesman, of course, is going to say the speedometer goes up to 400 and hopefully the consumer will automatically believe what they see as a possibility, yet when actually tested, the vehicle will show its true colours. In some cases you get what you pay for, regardless of the marketing hype.
Most people assume that because the technology is "always on" they have a dedicated pipe through to the Internet at the speed they are paying for. This is definitely not the case as you will see from the test results. Luckily there are also some better providers -- one of which in this test ran at average 110 percent of their stated upload speed and 96 percent of their download.
Some providers differentiate their services these days with "business grade" and "consumer grade" services. Usually marketers say business grade is a guaranteed monthly or yearly uptime of around 99.99 percent, whereas the consumer grade is slightly lower at only 99.9 percent. Here again, watch out for tricks such as stating the percentage only for unforeseen outages not "planned outages".
What you really need to look out for in a business-grade service is the committed information rate (CIR) which is a guarantee that the service will not ever fall below a certain speed. This is what really sorts the wheat from the chaff, and when you start asking for CIRs above something like 60 percent of the stated speed, the prices start going up, and up, and up.
So, you ask: "How do I know that I am getting what I am paying for?" Unfortunately the answer is that you really don't. Many service providers love to tell their customers to download and install simple FTP client tests which enable the end-user to run performance tests on the network. This is all well and good but all the client is testing in most cases is the speed of their connection to their provider (effectively on the same network), not the speed of their connection to the Internet (upstream of the provider).
There are a plethora of downloadable client and online bandwidth and speed-testing tools. The Test Lab over the years has tried many of these and found most of them lacking. Unfortunately, there are a great many issues involved such as compression and shaping, not to mention routing, caching, and firewalls, that can all affect results to such an extent that the data recorded really cannot be relied upon and in some cases is totally inaccurate.
Luckily, as you will see from the "How we tested" section, we took it upon ourselves to create, over time, a reliable, accurate, broadband performance test that we can schedule to test provider's connections at regular, or random, intervals. This takes all these factors into consideration and can give the test engineer some really good feedback on issues that may exist within the network paths and pinpoint areas needed for further examination.
All this can be very confusing for consumers, particularly where broadband is concerned, so let's get back to basics and try to explain how the providers cut costs. Imagine that the Internet is a really big pipe, I mean really big -- some of the backbone now runs at multi-Gb speeds.
Those with lots of cash can access the backbone directly. The majority are telecommunications providers, larger Internet service providers, or bandwidth wholesalers/data centres. These are known at Tier 1 providers -- they generally wholesale slower speed services to their downstream clients. Each step away from the backbone drops a tier. The majority of Internet service providers are at around tier two or three.
An upstream provider may purchase a 10Mb/s service and sell six 2Mb/s services, relying on the fact that each of those six customers wouldn't need all the 2Mb/s simultaneously. One of those six customers may then take their 2Mb/s service and sell ten 256Kb/s services. This or course relies on the fact that each of these customers will not simultaneously be using 256Kb/s, if they do, then each will only reach a maximum of just over 200Kb/s (and that's only if there's a a decent load balancer or packet shaper in place to even out the data). What usually results is that user's connection speeds slow right down to a trickly during busier periods.
One cost-cutting measure is achieved when a provider subscribes to a smaller bandwidth necessary to support the amount of customers they have connected to their service. This is similar to the modem ratio in dial-up Internet services. Naturally one could not expect a 1:1 ratio, particularly when dial-up can now be had for AU$9.90 per month. The ISP still needs to pay line rental, plus all the other associated costs in running a business, so if they run 1:1 they will be losing about AU$60 per month per line/user. To break even they need a ratio of 7:1. This is seven users for each line/modem that they have. Most providers with more than 100 lines can easily run a ratio of 10:1 or slightly more. Once the ratio gets to over about 18:1 the service starts to suffer (not so much in terms of speed as dial-up has a relatively low impact in terms of performance) however, end-users start to experience engaged signals during peak Internet periods. Peaks are between 4 and 11pm and, to a lesser extent, 6am and 9am. So the optimum ratio of users to lines for a dial up provider is about 10:1.
Because most forms of broadband are "always on", end users have no issues with engaged signals but providers still need to balance their upstream connectivity with the user's demands. "Always on" Internet access demands still have the same peak times as with dial-up, therefore, between those hours of the day the provider needs to ensure that they are subscribed to adequate upstream bandwidth to ensure each of their users has an acceptable speed. Providers can purchase cabled bandwidth ranging from 2Mbp/s through to 155Mbp/s, mainly delivered via Ethernet, fibre, or Asynchronous Transfer Mode (ATM) circuits. If they are lucky enough to co-locate their data facility in a centre with direct backbone connectivity then they may have access to 1Gbp/s or greater, although they still need to get this to the exchanges, so most fall back to fibre or ATM. ATM in this context is network terminology, not a machine used to withdraw money.
If the ISP oversubscribes the service and runs a high contention ratio of users to bandwidth during the peak daily periods user's Internet performance will begin to suffer. If a provider had, say, a 2Mbp/s connection, they could run eight users on 256Kbp/s with no contention at all. This means that a user who subscribes to 256Kbps gets 256Kbps. The larger your bandwidth the higher the contention ratio that can theoretically be run because not all of the users will want to access the Internet simultaneously. Some smaller providers (it has been rumoured) run contention ratios upwards of 40:1 which can impact on their slower pipes. However a decent average (between 20:1 and 30:1) should keep most services humming along nicely during peak periods.
Wow what an excellent review, keep up the great work. This is what we love to see