Monday, 10 August 2009

Basic truths about broadband speeds (reaction to Kevin Walsh on GigaOm)

On GigaOm Kevin Walsh has written down some basic truths about broadband, as he sees them. He sees a big problem with politicians wanting the wrong things when it comes to guarantees on bandwidth speeds. As he says:
Many believe that broadband service providers selling, say, a 5Mbps service should be required to set aside the same amount of capacity in order to fulfill that implicit service-level agreement (SLA). In other words, if you pay for 5Mbps, it’s there when you need it. But the reality is that networks, just like hotels and airplanes, are almost always oversubscribed — the owners of these assets sell more capacity than they have available.

Unfortunately this article messes things up just as badly as the politicians criticized. I reacted in the comments and post the comment here too. There are the important elements that need considering.

  1. The speed of the line from the end-user to the local aggregation point (switch, cable head-end). DSL and wireless technologies are particularly crappy, as the speed the end-user can have is a function of the distance the end-user is from the DSLAM or the antenna. Unfortunately there are still telco’s that sell up to 8mbit or up to 20mbit subscriptions that can only attain 4-8mbit/s because of distance limitations. The really nasty one’s try to upsell the customer to a top tier 20mbit/s line where only 4mbit is achievable and so a 4mbit/s subscription would have sufficed
  2. The speeds that can attained between 6pm and 10pm and the speeds that can be attained between 2am and 6am. These can be limited by oversubscription on:
  • a. the local segment (cable and wireless) and
  • b. on the ISP’s WAN
  • c. from the ISP to the rest of the world

Oversubscription on the local segment is a fact of life on cable and wireless networks. It is a shared medium. This should be clear to end-users. It is not bad, it is a fact of life. It’s effect is that between 6pm and 10pm the speeds can be erratic. between 2am and 6am the listed speeds can quite often be attained.

Oversubscription on the WAN is part of the problem you’re describing above. Oversubscription on the WAN is not an economic fact, it is a result of crappy network planning. Oversubscription on the WAN can be completely unnoticed by the end-user if the network operator builds enough bandwidth into it’s WAN. If the network owner and the ISP are the same entity, this shouldn’t be a problem. With traffic growing 50% per year proper network management dictates that an oversupply of bandwidth is necessary anyways. Statistics from the AMS-IX in Amsterdam show that peak traffic is about 50% higher than average and three times higher than the bottom. So next year your average is the same as today’s peak and in 2.5 years even the bottom is at today’s peak. WAN Bandwidth is a problem in some countries like the UK and the USA where the costs of backhaul to and from smaller communities are extremely high because of regulatory and/or competitive problems. Mind you, technical and cost limitations are often not important here, as the costs of installing faster equipment is often not prohibitively high. (DWDM, 10Gbit/s ethernet etc) This also means that you don’t have to build your network in a fashion where everyone can achieve max speeds at the exact same moment. Just carefully planning it. Like the highway system, where we don’t expect all car drivers in the US to show up at the Brooklyn Bridge at the same moment (or to start driving at the same moment at all, regardless of location).

The costs of traffic from the ISP to the rest of the world is governed by the economic laws of peering and transit. For an explanation see my article on Ars Technica. Whether enough is available to the end-user is dependent upon how easy it is for an ISP to get peerings with the most important networks (Google, Microsoft, Yahoo, Akamai etc) and the local costs of transit. Many developing nations find that their biggest problem lie here:
  • The national incumbent monopolizes transit traffic and charges outrages amounts for it.
  • No local internet exhanges to keep local traffic local.
  • No possibilities for local peerings with Google, Microsoft, Yahoo etc meaning that the transit link gets hit harder.) 
Amsterdam, London, New York are places with low costs for transit ($4/mbit/s/month) and many peering opportunities, so any network operating there should be able to get enough traffic for their end-users. Dave Farber once mentioned that traffic costs were only between 1%-5% of a subscription.
So, to conclude:
  • Politicians are right to complain when listed speeds on the local loop cannot be attained because of distance problems. Providers of DSL and wireless should be put in the doghouse for this. The should inform their customers properly of what speeds can really be achieved.
  • Cable networks and wireless networks could be required to publish the mean and median speeds users can attain between 6pm and 10pm.
  • Problems on the WAN and on the interconnect to the rest of the world are either a result of bad investing in backhaul or because of regulatory and competition problems. If that is the case the ISP should inform its customers of the situation, explain why this is the case and show how it deals with distributing a scarce resource among all the users. A good example is the Plusnet DSL network in the UK who are very clear on how they prioritize network traffic between different classes of customers.

No comments:

Post a Comment