For many disaster or threatening situations we have developed reference scales to calibrate the severity of individual events. From earthquakes, hurricanes, storms, pollen count or pollution we have simple and singular metrics. But there are events that are threatening and stressful, both natural and self inflicted, that remain uncalibrated in any way. Birth, death, illness, being mugged, getting divorced and changing jobs are common examples. Recently, IT has added a raft of new mechanisms to the menu; like changing or upgrading computers and/or their software, operating bugs and crashes. But, in an increasingly virtual world of electronic commerce, perhaps the most critical calamity will be the network crash. When nets crash, virtual organisations stop working - little or nothing can be done. So, how might we calibrate their severity?

So far there have been relatively few network failures on a scale that have caused catastrophic economic disruption. One exception was 1991 when 40M people lost their telephone service for 19 hours on the East Coast of the USA. Computer networks have only recently assumed great economic importance and quantifying the impact of failures may become vital if future network design is to be correctly focused. The key difficulty is the diversity of the failure types, causes, mechanisms, avoidance and customer impact. What you are doing when there is a crash is important. In the worst case valuable work and information can be lost or corrupted. So we might contemplate a simple means of ranking network failures so they can be readily understood by non-specialists. Perhaps the Richter scale for earthquakes could be modified to meet that need. After all, it brings the advantage of an established familiarity and intuitive feel across a broad range of potential users. For any simple linear measure, network outages vary over many orders of magnitude, so a logarithmic, Richter like scale, seems highly appropriate.

Whilst absolute accuracy may be important to those investigating outages or seeking to protect networks from them in future, it may not be a strong general requirement An outage in the early hours of the morning has arguably far less impact on people than the identical event at peak working times. Also, the level of attention to an outage is likely to diminish with distance from the affected area, so a measure needs to be independent of these effects.

When Steven Hawking was writing A Brief History of Time', he was advised that book sales would see a halving for every mathematical equation included. So he opted for just one equation - and I decided to do the same here. A NetQuake measure based on down time and number of people affected seems a sensible first step. So, following the approach of Richter, we define the customer impact of a NetQuake as: Q = log10 NT. Where N is the number of computer terminals affected, and T is total down time.

On the earthquake scale, a magnitude 6.0 event has special significance, because it marks the fuzzy boundary between minor and major events. So we might calibrate a magnitude 6.0 NetQuake to be represented by 100 computer terminals off line for 10,000 seconds (2.8 hours). But notice 10 terminals for 100,000 seconds (28 hours), or 1,000 terminals for 1,000 seconds (16.8 minutes) and so on, give the same result.

Earthquakes in excess of magnitude 7 are considered major events, and in a local geographical sense are rare. The same is true of electronic networks, but the notion of geography is different. Computer terminals can be distributed across the planet. So NetQuakes can be distributed and concentrated in business or sector space - and not necessarily geographical space.

Peter Cochrane holds the Collier Chair for the Public Understanding of Science & Technology at the University of Bristol. His home page is:
http://cochrane.org.uk