2015-06-18

Broadband speed

What do we mean by "broadband speed", and what are the factors that affect it?

I had intended this to be one of my blog posts aimed at my less technical readers (Hi, Pauline), and explain some of the issues in a non technical way. It seems, however, that people like the ASA, OFCOM and Which may do well to read it as well. Again, they miss the point hugely (see recent ispreview article).

Technology

First I'll just summarise some of the different technologies and what they mean. Some technology can be used by many different ISPs via a whole sale service (i.e. exactly the same equipment and wires used, so no difference technically). In some cases multiple ISPs can use the same type of equipment, so the same technology and basically the same speed and service offered but by different actual equipment. In some cases an ISP will have their own dedicated equipment and so offer a very different service to other ISPs even to the same address.

  • FTTC (Fibre to the Cabinet) is one of the most common technologies now. It is provided using a normal copper pair (phone line) that has equipment connected at the main street cabinet to provide broadband over that phone line. The cabinet has glass fibre back to the exchange and on to the ISP and the Internet. It is almost always a BT plc t/a Openreach cabinet and equipment which connects to an ISP or back-haul carrier at the exchange. This same technology can be sold by lots of ISPs and the link speed will be the same regardless. Typically it can be provided with a cap at 80Mb/s down and 20Mb/s up, or a lower cap of 40Mb/s down and 10Mb/s or 2Mb/s up. These are usually tariff options. The speed depends on the line length and quality from the cabinet to the premises. It can reduce a bit over time as more lines on the same cabinet get service. It makes no difference to the actual sync speed which ISP you choose when buying this sort of service.
  • ADSL (Asymmetric Digital Subscriber Line) uses a copper pair (phone line) from the exchange to the premises with equipment in the exchange. Glass fibre is used to connect from the exchange on to the ISP and the Internet. There are typically two variants: ADSL1 and ADSL2+. The former allows up to around 8Mb/s sync speed (around IP 7.15Mb/s data rate), and the latter up to 24Mb/s sync (21Mb/s IP data rate). Again, the sync depends on the line length. Speeds can be as low as 250kb/s on very long lines. Apart from some areas with older ADSL1 only BT kit, most ISPs can offer ADSL2+ either using BT, or TalkTalk or some other back-haul carrier, or even their own kit. In general, it will make no difference to sync speed which ISP you go for. It may matter if one can only offer ADSL1 and another can offer ADSL2+, especially if you are close to the exchange and could get speeds over 8Mb/s.
  • FTTP (Fibre to the Premises) is similar to FTTC but has a fibre from the cabinet to the premises instead of copper wires. There is technology provided by BT plc t/a Openreach with 40/10, 80/20 and even 330Mb/s speeds. Unlike FTTC the speed is the same regardless of distance. There are some ISPs that offer different direct fibre services, and some that even provide gigabit (1000Mb/s) services. This is much rarer.
  • Coax (cable). In lots of places there are alternatives that don't actually use phone lines. Virgin have a network of cable installations originally for cable TV services. These use coax to a cabinet nearby, and may then use more coax to a main cabinet further way. These are then typically connected using glass fibre to a core network and on to the Internet. Like FTTC, this is not a "fibre" service. The speed depends on the length of coax, and can be quite fast (e.g. 100Mb/s). Typically such services are not available to other ISPs, so this is an area where choosing the ISP offering this service may provide a very different service to choosing a different ISP which is inherently using a different technology.
  • Radio. Like WiFi but over a wider area, some ISPs provide radio links. The speed can depend on distance to their nearest mast. Like coax/cable, this is typically a very different offering to other ISPs using different technologies.

Advertised speed

There are different sorts of underlying technology used to provide a broadband (Internet access) service. What we see is companies advertising a possible speed - usually as "up to" some number of megabits per second. This is actually a data throughput, not speed, but that is not really important here.

One of the problems is that some of the technologies used are adaptive to the underlying phone line that is used, and so a long phone line will have lower speed. This is a simple fact of physics, and generally, for a specific technology, the speed you can get will be around the same.

The ASA were unhappy with people advertising ADSL2+ technology as "Up to 24Mb/s". The technology can do that, on short and good quality lines, but on longer lines it is slower. The "up to" is quite correct. Read that as "not more than" and it makes as much sense. However, the ASA felt that this was somehow misleading almost all customers, and they felt that ISPs should only mislead 90% of customers rather than 100%, so they insist that "Up to" speeds are set at a speed that at least 10% can achieve. This is, of course, a totally daft thing to do - if customers do not understand "up to", then all you do is reduce complaints by 10%. What you need to do is make sure customers understand what the statement means.

OFCOM stated that "BT’s ‘up to’ 76 Mbps package: Only 1% of customers received the maximum advertised speed.". For a start, that surprised me - and I suspect they are not actually looking at the line speed but perhaps measuring speeds of transfers in to the Internet (see below). I would expect that 10% of lines can get 76Mbit/s or more on that service. In fact I would say that 90% get what was advertised, which is "not more than 76 Mb/s" and 10% don't (i.e. they get more than 76Mb/s so the advert was misleading to them).

The good news is that regardless of the "up to" rate, ISPs provide means to get a speed estimate for a specific installation, and that is all that matters. It does not matter if you are buying an "up to 40Mb/s" service or an "up to 80Mb/s" service if the service you can get at your address is only 25Mb/s.

If anything, I think adverts should explain the technology and carrier used. This would allow comparison to be simpler. If comparing two ISPs, if both use "BT Wholesale" then the underlying technology that both ISPs can use will be the same. Speeds will be the same. It matters if you order FTTC 40/10 or FTTC 80/20 or ADSL2+, but typically you could order those from any ISP using BT Wholesale backhaul. Similarly, using Talk Talk wholesale backhaul would offer ADSL2+ which will work at basically the same speed as BT Wholesale ADSL2+. For FTTC on TalkTalk the underlying modem and line is identical to BT as it uses Openreach modems in the cabinet. Now, some ISPs are different - some use radio, or coax, and so on. Some even use fibre, and there really should be a ban on claiming that copper coax or copper telephone lines are "fibre" to avoid confusion and aid comparison.

Explaining the underlying connection type would allow people to compare ISPs and the packages more sensibly.

Up To

I have raised this before, but one issue with "up to", and I think a key confusion, is that there are two "up to" speeds involved. One is the fact that the line speed will depend on length of line and quality, and so for a service the speed may be "up to 80Mb/s" meaning that one address may get 80Mb/s and another may get 45Mb/s, etc. It is not a variable, it is pretty much fixed for the address based on its line characteristics. It can change over time (as more lines get broadband in they area).

The other "up to", and perhaps what people are confusing the advertising with, is that a line synced at 80Mb/s is not always transferring data. If you transfer a file, you will get a speed of transfer for that file. If sending several files at once, each file will share that one link. The speed the file transfers will be anything "up to" the line speed you have. When it is the only use of the line, and the other end is not busy and well connected to the ISP, the speed will be the line rate. So a line synced at 45 Mb/s can transfer files "up to 45Mb/s".

The problem I see is that people buying a service that is "up to 80Mb/s" expect that, at some times, when "the Internet is not busy" that they will be able to transfer a file at 80Mb/s. They don't realise that this is not what they are buying. It is an "in some places may be 80Mb/s" service, not a service that "for anywhere that you can get it, can, at times, get 80Mb/s". The "up to" part is ambiguous!

Throughput to the Internet

The other issue is how fast can you download something from the Internet. This is generally what people actually care about, not the sync speed.

There is a huge problem here - the ISP has some control over the service they sell in terms of the technology used to provide it (the "up to 80Mb/s" bit) and the backhaul they install or buy, and the links they have "to the Internet". However most of the services that customers which to access are outside the control of the ISP. Indeed, most are outside of the control of anyone the ISP contracts with.

Now, there are things an ISP can do such as choosing good peering and transit, but ultimately an ISP cannot be responsible for the speed of third parties.

Unfortunately it is very hard to tell if there is a problem with speed to the Internet, whether it is something the ISP does control (back-haul from ISP to premises, core network, choice of links to peering points), or something outside their control (transit, far end congestion, busy servers).

Now, maybe OFCOM could commission some independent speed test equipment that they connect to LINX and LONAP and other peering points, and use that as a reference. The issue is that it is not hard for an ISP to prioritise these speed tests and still have congestion for other connectivity!

What would be good is ISPs having to publish loss and latency from premises in to their core network - that is the area that is most likely to have congestion as it is the most expensive aspect.

Faults

Another issue which OFCOM seem to be trying to address, but in an insane way, is lines that are faulty. The problem here is that a long line running at 500kb/s may be perfect and the best you can get on such a line, and not a fault, but a short line running at 500kb/s may be faulty.

OFCOM tried to define faulty lines by saying that they are the lowest 10th percentile - defining 10% of an ISPs customers as having a fault - which is crazy!

Defining that long slow line as faulty does nothing to actually get it fixed, and allowing a customer to leave with no penalty does not help the ISP or get any investment in infrastructure that could help the customer.

This is especially true when dealing with links such as BT Wholesale. The ISP may be penalised by insisting customers can leave with no penalty even though the ISP may have paid for installation and a router, and hence making a loss. However, BT Wholesale see no issue. Indeed, they see migration fees and maybe even install and cease fees much more often as a result of OFCOM rules. BT Wholesale are not penalised or incentivised to get BT plc t/a Openreach to install an FTTC cabinet instead. OFCOM are penalising the wrong people.

To add to the fun, a line that should get 20Mb/s and is getting 10Mb/s because of a fault is probably not bad enough to meet OFCOMs 10th percentile rule, and so that customer does not have the right to leave with no penalty!

It would make a lot more sense to say that a line that cannot sync at the forecast minimum speed can be ceased with no penalty, and apply that rule at wholesale level as well as retail. That would allow individual lines that do not live up to expectations to be fixed or penalise the people actually providing the faulty service itself (the carrier) rather than the end ISP.

13 comments:

  1. It would appear that 40/2 is gaining popularity as an FTTC speed, with both TalkTalk and PlusNet offering it by default now. Should think that's going to affect download speed significantly at some point, as that's not a lot of upload left after ACK packets have left the building!

    ReplyDelete
    Replies
    1. Just when I thought FTTC made more sense than Virgin because FTTC uplink was a quarter of downlink, they go and do something stupid like 40/2. Are they mad? My dad gets 12/1 on his ADSL 2+ ie. a better ratio, and yet usually when things are slow for my dad it is the 1mbit uplink which is saturated. He regularly complains that sending large emails is slow.

      Delete
    2. We are not planning to do 40/2. It is an odd choice for ISPs. Maybe if the FTTC is slow (e.g. 10M) it may make sense.

      Delete
  2. I did a little more reading about this following your blog post a week or two ago. It looks like the proposed new rules (minimum guaranteed access line speed) are something of an improvement: the 'slowest 10%' criteria is based on 'similar connections', so it wouldn't apply to all ADSL connections as a whole, but only those of a similar distance from the exchange (or in FTTC terms, a similar distance from the cab, one presumes).

    So there would be no need to consider the long slow line getting 500kbps in your example above as 'faulty', provided other lines of similar length were also getting around 500kbps.

    Granted, it still doesn't address the issue that in their view 10% of lines would always be considered 'faulty', which is just plain daft, but at least it *tries* to take into account line length...

    ReplyDelete
    Replies
    1. Indeed. The exact criteria for "similar" was never that clear, but the more you narrow that the worse it gets. It is always at least 10% of lines faulty. Last time it was lines *AT* or below the 10th percentile, so short lines getting max rate (e.g. lots of lines on 40Mb/s capped FTTC) will all get exactly 40Mb/s meaning that 100% are considered faulty...

      Delete
    2. My 40/10 FTTC line is faulty because I only get 40/10! I want a free migrate out to these nice A&A people who claim to fix all faults on migrated lines within a month... :-)

      Delete
    3. I'd say at least 10% of line use aluminium or other poor grades of metal so it may help get cables changed in slow areas. Doubt it though since the reason they aren't changed is the horrendous cost and changing provider wont actually improve the speeds, they could already change at the end of a year anyway.

      Delete
  3. Well I've just quickly perused the new code but nowhere does it mention the phrases "distance from the exchange" or "length of line". So 'similar connections' can mean anything the ISP wants it to mean. Are OFCOM really so dense that they are unable to see that not defining 'similar connections' will inevitably lead to abuse by the more unscrupulous ISPs?

    Furthermore, I assume that minimum access line speed refers to sync speed. If so, it is an even more useless definition. Someone with a congested ISP who gets a sync speed of 12mb but only gets a throughput speed of perhaps 5mb because of under capacity at the exchange, will not be able to break their contract without charges.

    ReplyDelete
  4. I read someone elsewhere claim if an advert says 'download speed' it should mean "download throughput at the application layer". I’m not sure, but the ASA quotes do seem to indicate that the speed should be based on the relevant protocol rather than raw data rate, and based on speed testing representative of downloading rather than sync speeds. Wondered what thoughts were. All of what follows is a copy of what he said…

    The guidance document is here: https://www.cap.org.uk/~/media/Files/CAP/Help%20notes%20new/speed%20claims.ashx

    It is clear to me at least that for a claim of 76 Mb/s download speed, like Plusnet make, the speed should refer to download throughput at the application layer seen by the end-user, not sync speeds. Looking at ASA judgements, the ISPs indeed seem to submit speed tests as evidence not sync speeds, I saw one referencing Ofcom's report and one uSwitch speed test data.

    Quotes from the document
    - Where advertisers make a numerical speed claim that is likely to be understood by consumers as the maximum speed of their service, they should be able to demonstrate that the speed is achievable for at least 10% of the relevant customer base.

    - Specific claims, such as “50Mb Download speed”, should be based on tests of protocols relevant to downloading large files.

    - Speed claims for a service in general should be based on speed testing that is representative of the activities that users generally perform.

    My guess is ISPs think their numbers can comply, and they think they can prove it through calculation of the throughput after overheads, plus real testing on a sample of lines on max sync, which at least 10% of users have.

    ReplyDelete
    Replies
    1. I had not really gone in to protocol overheads, and that is yet another issue. It seems it is, once again, a total failure of ASA and the like to understand what is being sold. Again, connection beyond the link in to the ISP is outside of the control of the ISP, but ignoring that for a moment, the ISP sells, at most, IP connectivity. They don't sell HTTP or FTP protocols. The choice of protocol is down to the user. Some are horridly slow at application, e.g. if you look at actual displayed text from an HTML page to overall IP level data transferred. I mean, if I send a high res image that depicts a single 4 letter work, is that 4 bytes are user level? Some protocols may include compression, so could mean higher application rate than IP rate. Comms lines have always been sold at bit rate. Nobody ever complained at 300 bps modems, or 64k ISDN being misleading. So all an ISP needs do is test a large text file over a gzipped lower level protocol and they can advertise massively higher speeds - yay ASA!

      Delete
    2. I can certainly remember a lot of 28.8kbps modems being sold as 115.2kbps because that was the theoretical top speed after compression (generally limited by the RS232 interface between the computer and modem). Also tape streamer / DAT capacities and throughputs were often quoted after taking compression into account, rather than the raw throughput.

      Delete
  5. I can see where the ASA/BCAP are coming from, as someone else suggested, think of the massive overhead discrepancies between Wifi, 4G, VDSL, ADSL. As I interpret the guidelines, if an ISP advertises "Up to X Mb/s data rate" or "Up to X Mb/s sync speed" then they are fine, as this is a specific claim which can be backed up by reference to the relevant protocol. The problem comes when they start advertising "Up to X Mb/s download speed" and aren't the ASA right, hasn't the ISP itself overstepped the mark and started selling based on maximum performance of download protocols? I know you have a sharp mind for this sort of thing, and wonder what's your interpretation of the guideline? I also wonder if ISPs selling "Up to 76 Mb/s download speed" are using IP throughput as their measure. Forgive me for being dumb, but is IP level data all passed on to end-devices like a tablet or PC? If so, then I can see how they think this counts as a "protocol relevant to downloading".

    ReplyDelete
  6. It's a great shame that the physics of copper don't allow the same kind of progress that the physics of iron on "hard" disks have permitted over the last 20 or so years.
    Recording densities have increased enormously so that mutli terrabyte spinning disks are now small, even tiny, and very inexpensive.
    Copper transmission technology has made a lot of progress but not nearly as much as has been made with "Winchester" disks.
    The only answer proposed at present is to replace copper with glass where possible and financially feasible. This is bad news for anyone not living in demesne populated -and therefore cheap to re-cable - urban areas.

    ReplyDelete

Comments are moderated purely to filter out obvious spam, but it means they may not show immediately.

Missing unix/linux/posix file open option

What I would like is a file open option for "create replacement file". The idea is that this makes a new inode in the same mount p...