Broadband News
News, views and analysis
OFCOM Speed Survey Reaction
29 Jul 2009 | 19.01 Europe/London
Much has been written in the last day or so about OFCOM's survey of practical broadband speeds. A good deal of it has apparently been written by people who haven't read the 113 page research report and most of it by people who just don't understand the issues. It is perhaps indicative of the need to dumb down that OFCOM has to publish a table saying "A is faster than B" rather than leaving the reader to draw their own conclusions from the data presented.
The samknows hardware based testing methodology and results used by OFCOM are widely accepted in the industry, and at least one ISP is using it for their own purposes as well as being monitored by OFCOM's test units. There is no doubt scope to improve the methodology and analysis and samknows.com would be happy to discuss this with interested parties. We are proud that we have been able to produce a set of independent metrics that previously were not available.
Turning to the reactions, the most common cry is "I am paying for 8Meg and not getting it". Well sorry to rain on your parade, but you don't "pay for 8Meg" at all - you pay for a "variable rate service that will connect between 288k and 8128k depending on line conditions, interference and other factors". Obviously the advertising creatives slim down this message to "8 Meg" for popular consumption but the reality is that 8M ADSL is a rate adaptive service. Nobody loses out by getting 4M on the 8M service as there isn't a cheaper 4M service in any case, so quite how these people are losing out is beyond me. So the first lesson to understand is that most ADSL broadband services are variable rate "up to...." on account of the physics. If you want a 0.5, 1 or 2M fixed speed service they are still available from some ISPs (there are over 100 to choose from, remember) or if you're on Virgin cable then you have a fixed speed in any case.
The next lesson to learn is that the quoted speed includes various add-ons or overheads, a bit like a box of chocolates weighs more than the sweets within. A link speed of 8128 kbits/s (ATM data rate) contains 7150 kbits/s of TCP/IP packets which in turn contain about 6800 kbits/s of useful data. For this reason, an "8 Meg" service isn't going to speed test any faster than 7M, ever. Some may call this false advertising, others would say "read the detail". Virgin cable doesn't have the ATM layer so the maximum test speed will be closer to the declared "10 Meg".
Next we come to the least understood matter of contention - the sharing of network capacity between users. To keep the price down to affordable levels the designers assume that not everyone will be using or maxing out their connection at the same time, perfectly reasonable in the same way that not everyone makes phone calls or uses the roads at the same time. A consequence of this is that each user probably has well below 100 kbits/s of capacity available to them on average, for example 3,000 Talk Talk users on a 100 Mbits/s exchange connection. If the average link speed of these users is 6M then the contention ratio is 3000 * 6 / 100 = 180 to 1, well above the "up to 50:1" ratio originally advertised for BT's 0.5M ADSL.
Contention isn't a problem, as long as it is kept low enough to avoid severe congestion. A bit of light congestion results in the slowing down at peak times measured in the OFCOM survey and to be honest the results were better than I would have expected - some broadband services slow down to 1/10 of their daytime speed in the evening due to congestion. My example 3,000 Talk Talk users have 33 kbits/s each (this is what they're paying for, not "8M") and if we assume 4 hours per day of peak time use that's enough to download 1.8 GBytes per month in peak time or running 24/7 it will deliver 10.7 Gbytes per month.
To me the most interesting statistic for comparing ISPs is the ratio of peak time data rate to the highest rate the line can provide. Unfortunately this isn't in the report, but we are given the average compared to the peak which for rural users is 3.6 vs 4.6M and for urban 4.0 vs 4.9M - a drop of 22% and 18% respectively.
A number of commentators have remarked on the sparsity of test units in some parts of the country. OFCOM use a sample set designed to be representative of the population and with 94% of broadband lines covering only 12.7% of the land area it is inevitable that there will be gaps in the sample coverage. If many more rural units were deployed then an equivalent increase in the number of urban ones would be required to maintain the representative sample.
The Communications Consumer Panel have called for ISPs to advertise average speeds rather than up to. This seems a pointless exercise as you will still have half the consumers getting less than the advertised figure, and more to the point the price is not determined by the speed for the majority of ISP packages. It is far better to educate the consumer in the effects of contention and use the monitoring to identify the ISPs that deliver less actual throughput while advertising the same link speed capability. This approach would go some way to OFCOM's goal of stimulating investment in extra backhaul to improve peak time speeds.
If there is a winner from the report it is Virgin Media's cable service, showing a fairly consistent throughput higher than the "up to 8M" ADSL products. VM also use "up to 10M" and I think this should be stopped, as a cable modem connects at the speed it is set to, so it is either 10M or nothing. The fact that the throughput is lower due to packet overheads and contention is not, in my opinion, a matter that should be wrapped up in one all consuming "up to" clause - we should leave that term for the variable rate services. VM have plenty of opportunity to increase their market share from its current ~20% despite covering 50% of the market with an apparently superior speed product.
Finally, it may be interesting to recall the product specification of BT Wholesale's up to 8M MaxDSL service, as this is what many users have and in many areas it is the only choice. It is defined as "a bursty service with peak rates depending on the line speed and at peak times it is expected to slow down to the same speed as a 2M connection in the same conditions". Regardless of what consumers think they bought, that is what their ISP provided, so perhaps it is time to understand this and judge the performance against the specification, rather than what you thought you were getting.
The samknows hardware based testing methodology and results used by OFCOM are widely accepted in the industry, and at least one ISP is using it for their own purposes as well as being monitored by OFCOM's test units. There is no doubt scope to improve the methodology and analysis and samknows.com would be happy to discuss this with interested parties. We are proud that we have been able to produce a set of independent metrics that previously were not available.
Turning to the reactions, the most common cry is "I am paying for 8Meg and not getting it". Well sorry to rain on your parade, but you don't "pay for 8Meg" at all - you pay for a "variable rate service that will connect between 288k and 8128k depending on line conditions, interference and other factors". Obviously the advertising creatives slim down this message to "8 Meg" for popular consumption but the reality is that 8M ADSL is a rate adaptive service. Nobody loses out by getting 4M on the 8M service as there isn't a cheaper 4M service in any case, so quite how these people are losing out is beyond me. So the first lesson to understand is that most ADSL broadband services are variable rate "up to...." on account of the physics. If you want a 0.5, 1 or 2M fixed speed service they are still available from some ISPs (there are over 100 to choose from, remember) or if you're on Virgin cable then you have a fixed speed in any case.
The next lesson to learn is that the quoted speed includes various add-ons or overheads, a bit like a box of chocolates weighs more than the sweets within. A link speed of 8128 kbits/s (ATM data rate) contains 7150 kbits/s of TCP/IP packets which in turn contain about 6800 kbits/s of useful data. For this reason, an "8 Meg" service isn't going to speed test any faster than 7M, ever. Some may call this false advertising, others would say "read the detail". Virgin cable doesn't have the ATM layer so the maximum test speed will be closer to the declared "10 Meg".
Next we come to the least understood matter of contention - the sharing of network capacity between users. To keep the price down to affordable levels the designers assume that not everyone will be using or maxing out their connection at the same time, perfectly reasonable in the same way that not everyone makes phone calls or uses the roads at the same time. A consequence of this is that each user probably has well below 100 kbits/s of capacity available to them on average, for example 3,000 Talk Talk users on a 100 Mbits/s exchange connection. If the average link speed of these users is 6M then the contention ratio is 3000 * 6 / 100 = 180 to 1, well above the "up to 50:1" ratio originally advertised for BT's 0.5M ADSL.
Contention isn't a problem, as long as it is kept low enough to avoid severe congestion. A bit of light congestion results in the slowing down at peak times measured in the OFCOM survey and to be honest the results were better than I would have expected - some broadband services slow down to 1/10 of their daytime speed in the evening due to congestion. My example 3,000 Talk Talk users have 33 kbits/s each (this is what they're paying for, not "8M") and if we assume 4 hours per day of peak time use that's enough to download 1.8 GBytes per month in peak time or running 24/7 it will deliver 10.7 Gbytes per month.
To me the most interesting statistic for comparing ISPs is the ratio of peak time data rate to the highest rate the line can provide. Unfortunately this isn't in the report, but we are given the average compared to the peak which for rural users is 3.6 vs 4.6M and for urban 4.0 vs 4.9M - a drop of 22% and 18% respectively.
A number of commentators have remarked on the sparsity of test units in some parts of the country. OFCOM use a sample set designed to be representative of the population and with 94% of broadband lines covering only 12.7% of the land area it is inevitable that there will be gaps in the sample coverage. If many more rural units were deployed then an equivalent increase in the number of urban ones would be required to maintain the representative sample.
The Communications Consumer Panel have called for ISPs to advertise average speeds rather than up to. This seems a pointless exercise as you will still have half the consumers getting less than the advertised figure, and more to the point the price is not determined by the speed for the majority of ISP packages. It is far better to educate the consumer in the effects of contention and use the monitoring to identify the ISPs that deliver less actual throughput while advertising the same link speed capability. This approach would go some way to OFCOM's goal of stimulating investment in extra backhaul to improve peak time speeds.
If there is a winner from the report it is Virgin Media's cable service, showing a fairly consistent throughput higher than the "up to 8M" ADSL products. VM also use "up to 10M" and I think this should be stopped, as a cable modem connects at the speed it is set to, so it is either 10M or nothing. The fact that the throughput is lower due to packet overheads and contention is not, in my opinion, a matter that should be wrapped up in one all consuming "up to" clause - we should leave that term for the variable rate services. VM have plenty of opportunity to increase their market share from its current ~20% despite covering 50% of the market with an apparently superior speed product.
Finally, it may be interesting to recall the product specification of BT Wholesale's up to 8M MaxDSL service, as this is what many users have and in many areas it is the only choice. It is defined as "a bursty service with peak rates depending on the line speed and at peak times it is expected to slow down to the same speed as a 2M connection in the same conditions". Regardless of what consumers think they bought, that is what their ISP provided, so perhaps it is time to understand this and judge the performance against the specification, rather than what you thought you were getting.