Wednesday, 2 March 2011

OFCOM's speed report

Speed of broadband lines is an interesting issue, and OFCOM have once again released a report that focuses on the larger ISPs.

Statements like "The average broadband access download speed UK users experienced rose 5% to 6.2Mbps, but this is less than half the average headline speed they pay for, communications regulator Ofcom says." just show some serious misunderstandings.

If you are paying for an "up to 8Mb/s" line then you are getting what you are paying for if it is 500K. In fact, you are only not getting what you pay for if it is over 8Mb/s (when you are getting more than you are paying for).

Oddly, I don't recall this being an issue when people purchased 56K modems. Unlike ADSL1 where a significant proportion of people get the full 8128K sync rate, 56K modems rarely, if ever, got 56K.

There are a lot of issues that affect the apparent speed of a connection. It is not simply a matter of "how fast can I download a big file", though that is obviously an important point. As speeds get higher factors which previously did not matter, such as the software in the TCP stacks at each end, the load on the sending server, the latency on the link, and the impact of low levels of packet loss, can have more and more noticable effects. In many cases what an end user sees as "slow" can be a latency issue or a DNS problem and not a line speed issue at all.

So, in light of that, it seems odd that there is so much focus on line sync rate.

For some time the A&A site has not headlined speeds, but listed technologies (ADSL1, ADSL2+, FTTC, FTTP). We then explain in the more detailed pages that, for example, ADSL2+ as a protocol will allow sync rates of not more than 24Mb/s.

It is a shame we can't use common phrases like "up to" and people understand. Perhaps by saying "not more than" people will. Who knows?

To be honest, for the vast majority of people, the difference between 10Mb/s and 40Mb/s is not relevant. Indeed, the 6Mb/s OFCOM quoted is really good as it allows iPlayer and the like (assuming there are not latency and packet loss issues). Yes, faster is generally better, but it gets to a point where the line speed is not the bottleneck.

The problem is people compare headline speeds. We have the crazy situation that one telco can do 100Mb/s fibre links, so another is looking at changing the settings, using gigabit ports, and being able to quote it as 101Mb/s. It makes no odds but people will buy the faster speed. This is why lines are often quoted as line sync rate, e.g. 8.128M for ADSL1 rather than IP throughput (e.g. 7.15M for 8.128M sync). If you have one ISP saying "up to 7.15M" and another saying "up to 8M" then which do you buy? In that case the line is exactly the same, it is just where you measure it.

Anyway, we decided OFCOMs voluntary code of practice was, this time, actually impossible to adhere to. We have not signed up to it. We have our own set of commitments which we think are better.

A&A Speed commitments
OFCOM code of practice

Note for example OFCOM define "the minimum guaranteed access line speed" as "the 10th percentile of the ISP's similar customers". Indeed, lines "at or below" this are considered to have a fault. This is, of course, totally crazy, as 10th percentile is a moving target (if you fix lines below it, or they leave then the target moves up). Also, by talking of "similar customers" you make the range narrow. e.g. if we were to say "ADSL1 lines very close to the exchange" we can expect the 10th percentile to be 8128K sync (as more than 90% of ADSL1 lines very close to the exchange get that top speed). This means that *all* of those customers are at or below the the minimum guaranteed access line speed of 8128K. The fact there is the hard limit on top speed of ADSL1 actually makes the best and fastest lines considered "at fault". Even looking at all customers on one line type, like ADSL1, there is huge incentive to not fix dodgy lines as that keeps the 10th percentile rate lower. An ISP that actually tries to fix slow lines makes their stats look worse and worse as the 10th percentile rises. Someone made these rules with no thought at all.

There is a point about "the spirit" of the code and not "the letter", which is carte blanche to ignore almost all of these annoying details and so make the code a waste of time. The writers of the code clearly want at least 10% of all lines or all ISPs to be considered to be faulty on an ongoing basis. That is clearly the "spirit" of the code.

Our code says we will actually monitor all lines all the time, making that available to customers, and use it to assess the problem, considering the causes of slow speed (like loss and latency) as well, and taking slow speed reports seriously. It even says we will monitor how well carriers are working in their core network and tackle issues we find.

I think our code is much better. It is also a lot easier to follow the "spirit" rather than the "letter" as it is worded in terms of objectives not specific percentiles and rules.


  1. This comment has been removed by the author.

  2. I'm certainly inclined to agree with most of your points. "up to", at least for somebody like me whom is familiar with how ISPs work, isn't really a problem. However I would make some other points.

    A. On we never mentioned dialup speeds and most packages were merely listed as being either "unmetered" or "local call rate". Dialup performance was still a massive issue, actually it was bigger than today's debate but less well known because the internet wasn't as popular. For most people it was simply a case of being either incredibly slow or slightly less incredibly slow. The majority of ISPs also didn't advertise a numerical speed alongside their dialup service. So it's not that odd :) .

    B. It's also not really odd that the current situation has become such an issue. Masses of consumers, and I do mean "!MASSES!" (see.. I used two exclamation marks there hehe), have complained about this issue and to such an extent that the regulator / advertising watchdogs are left with no option but to act. That's their job.

    Likewise consumers haven't complained without just cause, it's very easy to see the problem, even though ideas upon a solution are more likely to vary. Imagine if your packet of crisps said "up to 50 chips" but the bag only contained 5.


  3. Do you have an opinion on the "bufferboat" problem thats being mentioned a lot recently?

  4. I was pondering why OFCOM had a preamble about the "spirit" of the code not the "letter" and then had such prescriptive rules saying what an ISP "must" do.

    I realise what they mean - they mean that even if an ISP complies exactly with what the code says and does all they "must" do, OFCOM will not consider this good enough if a customer complains as the ISP will be expected to go further than what it says they must do in order to comply with the "spirit" of the code.

    I seriously doubt OFCOM expect that *not* complying to at least "the letter" of the code is acceptable, regardless.

  5. Oh, and I am sure !MASSES! of people complain about slow speed. How many of them are in fact complaining about line sync speed though? I bet they complain because "the Internet is slow" not because they were "promised 20M and only have 15M". What will have happened, I expect is that they have less than the 20M (as many will) and have latched on to that as the reason things are slow when in fact it is congestion or a slow server somewhere else on the internet or other such things that mean they are experiencing way lower than their sync speed.

  6. Heh. Somebody doesn't understand the concept of percentiles, from your description I've not read the code).

    I don't care what my ADSL download speed is, as long as it's above ~3Mbps. That's enough for video streaming, is which is about the most demanding thing that I ever do in real time. Beyond that, if it has to sit as a background task it makes very little difference to me whether it takes ten minutes or twenty.

    The bit of dodgy ISP marketing (not from your good selves) that winds me up is their reinterpretations of the meaning of the word "unlimited".......

  7. A customer might also get a high sync rate, but the ISP hasn't made arrangements for transit that can deal with peak loads. In that case, the sync rate would be a bit meaningless.

    I always wonder about this in connection with content delivery networks. With AAISP, the links are good enough that I can download at the BRAS rate from sites in other countries (provided the remote end can keep up, of course). That would suggest that CDNs are a waste of time. However, perhaps people who go with cheaper ISPs have a different experience?

    I would love to know; if people have problems with slow international links, I need to set up my web sites differently!

  8. CDNs are important for large providers to be able to deliver the content. It is not really about the transit as such, though I am sure that is a factor, it is simply not practical for someone like microsoft to serve all s/w patches from their own servers in one place. CDNs provide some economy of scale as used by many content providers. It also helps that they peer with ISPs, e.g. we peer with content providers, reducing some of our operating costs if it was via transit.