Exponential Moving Maximum Filter for Predictive Analytics in Network Reporting

摘要

In networking industry, there are various services that are mission critical. For example, DNS and DHCP are essential and are common network services for a variety of organizations. An appliance that provides these services comes with a reporting system to provide visual information about the system status, resource usage, performance metrics, and trends, etc. Furthermore, it is desirable and important to provide prediction against these metrics so that users can be well prepared for what is going to happen and prevent downtime. Among the predictive measures, there are multiple metrics to reflect peak or maximum values such as peak volume or resource usage in networking. The peak value prediction is critical for the IT managers to ensure its organization is ahead of the cycles in terms of the network capacity and disaster recovery. There have been many algorithms and methods for prediction of trended time series data. However, peak values often do not fall into a trend by nature. The traditional trend prediction methods do not perform well against this type of data. In this paper, we present a novel filtering algorithm named "Exponential Moving Maximum" (EMM), this filter is used before applying a prediction algorithm against peak time series data. We also provide some experimental results on real data as a comparison to show that the prediction method has better accuracy when EMM filtering is applied to certain categories of networking data.