communications links from different companies that end up going over the same bridge that is washed away in a flood, or having vandals pour petrol down into underground cable ducts carrying many disparate cables and then setting them on fire . Efforts are being made to improve information about common-mode failures, and customers are increasingly insisting on knowing where fibre actually runs when they pur- chase telecomms circuits. Other lessons are being learnt from 9/11, in particular that systems switched to backup power, but that refuelling arrangements used a small number of companies – who could well have been overstretched, but in the event they couldn’t get permission to enter lower Manhattan anyway. Although there were schemes for getting priority access, the only companies involved were those that were in existence in the 50’s and 60’s when planners were considering nuclear war. The modern ‘dot-com’ companies were completely outside of these systems. In London Docklands there are now regular planning meetings between police, local authorities, data centre operators, Internet com- panies, and so on. In the event of an incident, there may still be difficulties in accessing the Docklands area while it remains a ‘crime scene’, but at least the police have been educated into understanding why that access might be necessary. 7.2.2 Internet exchange points A major concern about single points of failure for the Internet is the growth of Internet Exchange Points (IXPs) such as LINX in London, AMSIX in Amsterdam, DECIX in Frankfurt etc, and the way in which there are tendencies towards one IXP becoming significantly larger than its rivals. ISPs need to be able to provide their customers with connectivity to the whole of the rest of the Internet. They do this by purchasing ‘transit’ from a major networking company, paying for their traffic on a volume basis. To reduce their costs ISPs will attempt to negotiate ‘private peering’ arrangements with other ISPs, where traffic is exchanged ‘settlement free’. This traffic will not be for ‘all possible routes’, but only for the parts of the Internet operated by the other ISP. The largest ‘backbone’ networks (usually called Tier 1 networks) do not purchase transit from anyone, but operate solely on a peering basis. In the past there were only about 5 Tier 1 networks, but there are probably 9 at present, with another 20 or so ‘Tier 2’ networks that have peering-only arrangements in large geographical regions, but use a Tier 1 for remote locations. ISPs often use IXPs to reduce the costs of peering. One of the ISP’s routers is housed at the exchange point and ‘public peering’ traffic (and possibly transit traffic as well) is exchanged with other ISPs over the fabric of the IXP (multiple high speed Ethernet rings at larger IXPs, a switch backplane at the smallest). An ISP with large numbers of customers (or ‘eyeballs’, viz: a data sink) will find it easy to arrange peering with an ISP with a large number of content providers (a data source) because it will be in both their interests to avoid paying for transit. Thus, for example, companies such as Google,
You've reached the end of your free preview.
Want to read all 114 pages?