Barry Fox

Performance, speed - whatever you want to call it - has always been one of the biggest driving factors of the IT industry. It's also been one of the biggest sticks to beat it with. Nothing could be big or fast enough.

Today's digital cameras have more megapixels than we can afford to store in a memory card. Printer resolution is so fine we need a bank loan for the ink. The PC processor speed race has burned itself out. And now even internet access, which is never fast enough, is on the firing line.

Today's performance bottlenecks are on internet connections, not hardware. Mobile GPRS may deliver around 40Kbps, but only if there are not too many people making phone calls or sending data in the same area at the same time.

3G should deliver 384Kbps, but only when the mobile is stationary, reception is strong and congestion low. Something over 100Kbps is usually good going.

Once you have used ADSL broadband, a 56Kbps modem can induce screen-smashing frustration. Compared to Wi-Fi speeds of many megabits per second, even ADSL broadband is slow. The upstream speed of 288Kbps, often less, can reduce the downstream speed of 576Kbps as data packet acknowledgements are slowed.

Telewest has already started an internet speed war by taunting BT with comparisons between consumer ADSL and cable modems. BT's next move will surely be to offer faster DSL connections. So it's useful to recall how DSL began.

Ten years ago BT's research labs at Martlesham, near Ipswich, were planning a video entertainment by phone service. They rigged up an experimental ADSL system that could send a 1MHz analogue signal down 5km of ordinary twisted pair copper wires.

The analogue signal was a cluster of 256 sub-carriers sharing between one and two megabits per second of digital video.

The system delivered VHS quality pictures and was doomed to failure, not because it didn't work, but because UK law at the time stopped BT from broadcasting.

So BT had to store movies on a server and send them direct to anyone who asked for them, immediately when they asked for them. Compared to renting tapes from a video store, the cost of video on demand was prohibitive.

The failure of video on demand gave us ADSL broadband internet access. The technology can deliver higher speeds than are currently on offer: 3Mbps or 4Mbps over 3km or 4km runs from the nearest exchange and 1Mbps to most homes, except those in deep country. The current 576Mbps rate is a safe option.

If you pay a lot for a business connection (or are lucky enough to work for BT) you can already get 2Mbps. BT's next move will surely be to offer 1Mbps.

We may also see Symmetric, instead of Asymmetric, DSL. For SDSL, the analogue speech channel quaintly known as Pots (plain old telephone speech) is scrapped and digitally compressed speech sent as data at 32Kbps.

The frequency band can then be equally split to give 2Mbps each way at the same time. But if this puts my phone at the mercy of Windows, then thanks but no thanks.

As more connections offer higher speed, the bottlenecks move down the line. There are plans to build a new generation internet, called internet2, with higher quality lines to link universities and companies.

But what we really need is a way of making existing internet connections carry data more efficiently, thus faster.

The breakthrough could come from work being done at the Networking Lab at Caltech, the California Institute of Technology.

Fast TCP (Fast Active queue management Scaleable Transmission Control Protocol) updates the internet TCP that was developed in the 1980s to carry text files and is now creaking under the strain of music and movie downloads.

TCP breaks down large files into small packets of data. The computer sending the data seeks acknowledgements of safe receipt and resends packets that are lost because the routers have not been able to get them down the right lines.

When too many packets are lost, the transmission rate is effectively halved for a while. Making the packets larger can increase speed, but many of the routers round the world would need replacing.

Caltech's Fast TCP uses conventional packet size (so existing routers can still be used) but continually measures the time it takes for the packets to arrive, and the acknowledgements to come back.

This tells the round trip delays on the line, which gives early warning of losses. Intelligent software at the sending end then predicts the highest data rate that the connection can support.

A 10,000km test between California and Geneva clocked an average speed of 925Mbps, compared to 266Mbps from ordinary TCP along the same route.

For most readers, this is academic - getting a broadband connection is what matters. If your area is not cabled, BT is the only answer, except for an expensive two-way satellite link. All BT's exchanges are now digital and upgradeable to broadband, but only around a quarter have been converted.

Take a tip from the 14 year-old who wanted to play broadband games. He knocked on every door in his town, asking if anyone used a PC and would like something faster, and telling them how to register ADSL interest on BT's website. Now, as a result, they can all have ADSL.

Like this story? Spread the news by clicking below:

 del.icio.us     Digg this     reddit!