Lucky (crypto-leprechaun) wrote:
At 18:15 8/11/96, Dave Farber wrote:
The avalanche behavior of power systems is still not well understood and techniques to prevent such failures are not obvious. Same can be said of telephone and computer networks at different levels.
Does somebody here have a pointer to literature on this topic? A system in which small localized disturbances can amplify, propagate through the system, leading to catastrophic failure is the worst of all possible designs. I fail to understand why a system as important as the power grid would display this type of behavior. Why is the grid negatively dampened?
Because the system designed to fail non-destructively rather than risk a power surge. Part of the problem is that the power being controlled moves as fast as any information about the state of the network: all decisions must be made locally at the switch level. When a major line fails the power generation stations can not just "turn off the juice" at the speed required, the power must either be redistributed or else the switch overloads and shuts down to prevent a massive power spike. The power companies would rather face the ire of consumers without power for an evening than the outcry that would happen if every toaster, microwave, and computer on the west coast was fried by the spike caused by 3 gigawatts being dumped into grids that could not handle the load (not to mention the long-term blackouts caused by local switching equipment getting fried.) In this case the system must fail to prevent damage. The great east coast blackout was caused by a $50 switch which wore out. The switch happened to control a chunk of load which was re-routed on to another line, causing a cascade failure as the excess load caused other switches to fail when it was dumped on to those lines (and the excess load caused by these other switches tripping accellerated the failure.) Power distribution is not like telecom, if a phone switch dies the calls end; the failure does not endanger the upstream switch with overload from bits not going anywhere. Imagine that cars on highways had no brakes until they reached an offramp but the DOT could apply a kill-switch to segments between offramps (tying this thread into another non-crypto thread on the list :),if a failure in the roadway caused traffic to stop then all of the upstream traffic would have to be dumped off the road via offramps that were only one or two lanes. This would cause further bottlenecks and the kill-switch would need to be applied to more and more segments upstream to prevent the traffic stopped or being slowly bled off from causing more problems. No imagine that offramp-to-offramp messages were transmitted via messengers travelling upstream in the carpool lane :) Even the power company's secret experiments in the cores of nuclear reactors have not found a way to get around the 186,000 km/s speed limit on everything. jim