Original URL: http://www.theregister.co.uk/2007/10/27/data_processing/

21CN: It's not the data saviour

If there's more, we'll take more

By Guy Kewney

Posted in Broadband, 27th October 2007 08:02 GMT

Column It does seem to be a rule of data processing that we use all we can get of it. As BT starts to test its 21st century networking (21CN) with partners like Entanet, it's a safe bet that this rule will mean the dream of universal high speed data will turn sour as people try to squeeze more data through a limited pipe.

The world of broadband is, largely, a mess right now because of this rule. When we first heard of ADSL, most of us looked at our 54 kilobit modems and just swooned at the thought of all that bandwidth - 512 kilobits.

And today, we have eight megabit download speeds and "free broadband for life" - and yet, somehow, the service seems to stutter and hiccup, and drop. And then there are the "unlimited" downloads which are unlimited until you reach the maximum.

Just for fun, I set up a TalkTalk broadband installation and ran a ping test. Over an hour or so, the number of pings lost averaged about 20 per cent or more. I tried talking over Skype with it, and pretty much gave up. It works fine for browsing the web, and that's pretty much it.

So when 21CN starts its trials in Birmingham this November, there will no doubt be those who say "Wow, 24 megabits. That won't get congested." And they will be wrong.

The problem with ordinary broadband is that a lot of ISPs assumed they could work out their data budgets by calculating how much stuff people wanted to download. Much to their astonishment, it turns out people want to upload too.

And the main problem with people uploading isn't the obvious one (that the upload capacity is inadequate) but that the data they upload is being downloaded too. It multiplies all traffic to the point where congestion becomes a real problem.

As one small ISP recently complained to me: "To stay in business, we've had to cut prices. But the prices we pay to the network operator are fixed. So today, we cannot supply a 100 per cent 24/seven service. We can't even supply 100 per cent times Contention Ratio 7x24 to all users. We have to rely on the fact that most users don't use anything like that level of their available bandwidth most of the time."

But as peer to peer services become more popular, the amount of bandwidth used goes up. And for anybody using an internet phone, like Skype, the first effect of that is that there are blips and glitches and interruptions in the speech.

Skype strenuously denies this, but talking to customers, I find a general acceptance of reality: the once-excellent audio quality is now only good for a short time - after five minutes of a conversation it tends to degrade.

Other VoIP providers have had to do deals with internet providers to get them to improve from "best endeavours" to "advanced services" to prioritise streaming traffic. And some ISPs simply don't allow new P2P connections after the number reaches a set level. You try to set up a BitTorrent and it fails to connect.

The good news about 21CN is that, as Samknows summarises, there are more options for quality of service:

The QoS (Quality of Service) options are where the WBC products really stand out against [the ADSL option of] IPStream. QoS allows you to prioritise certain types of internet traffic; so you could prioritise a VoIP phone call (which is sensitive to latency) above a large peer-to-peer download (which is far less sensitive).

The new services are "best effort", "assured rate" and "real time", with the last being the highest quality with the least latency or interruption.

Will that solve the problem? Initially, I'll bet it does. But, at first, there will be no need for real time and its associated higher costs, because with a higher-speed pipe a simple best effort service will work fine. So people who are offering real time based services will be undercut by those who use best effort and get away with it - at first.

And the problem of upload remains. In a sensible world, the need for fast uploads could have been predicted. Telecoms has always been half upload and half download - I dial your number and we spend our time sharing the load in each direction. The reason it was decided to make DSL asymmetric was partly technology - it goes MUCH faster down if you restrict the upload - but also expectation, because many of the people buying DSLAM equipment wanted to compete in the cable TV market.

In short, they wanted to be broadcasters.

The mobile world is suffering from the same delusion right now, with operators clinging desperately to the idea that they can sell video clips to 3G phone users. In reality, they could generate far more traffic just by making it easier for users to upload more. The reason they don't go that way is that they simply don't have the spare capacity. Oh, OK, they have the spare capacity today, but if the video traffic started to become popular (that is to say, if more than five per cent of us started uploading video clips and sharing them over the air) the network would collapse.

It's a rule: code expands to fill available memory; data expands to fill available disk, and transmissions increase until congestion becomes a problem. Providing unlimited "dark fibre" and completely free switchgear at zero power drain would be the only way to avoid that basic truth. ®