Improving the Timeliness of Rate Measurements

Rate measurements are required for many purposes, e.g. for system analysis and modelling or for live systems that react to measurements. For off-line measurement all data is available in advance. Here, time delay between data collection and data analysis is not an issue. On-line measurement, however, measures rates on the fly. Thus, measurement algorithms that provide their output as timely as possible are required. We present three well known algorithms for rate measurement: The Disjoint Intervals method, the Moving Average, and the Exponentially Weighted Moving Average over Disjoint Intervals. We analyze and compare their properties and find problems like heavy time delay or overreaction to random fluctuations. To address these problems, we derive a new algorithm called Time Exponentially Weighted Moving Average as a continuous version of the Exponentially Weighted Moving Average. Finally, we compare this algorithm to the other methods and show that it solves these problems.