Death to Dashboards Other Thoughts on Data Visualization: Alarming, Forecasting and Performance Management Based on Variance

摘要

I went to a major metric vendor's users group conference. For almost an hour we talked about different techniques for overriding alerts. One idea, for example, was to tie together the ticketing system and the alerting system so that the alerting system would quit sending out alerts if a ticket was being worked. Great idea and the vendor went on to explain that there were at least three different ways of overriding alerts. The discussion went on and on. This should be an indicator to anyone that there is an elephant in the living room. When we have to spend this much time making threshold-based alerts work, it should indicate to us that thresholding as an idea is broken. What a waste of time and human effort.