Project Description

Software systems continuously running over a long time may suffer gradual performance degradation or failure rate increasing. This phenomenon, known as ‘Software Aging’, has become a great challenge to dependability-critical software systems. Researchers have made remarkable achievements in predicting resource exhaustion time and designing optimal rejuvenation plan. However, limited works focus on measuring aging and characterizing aging progress. And currently the only widely used tool is Sen’s slope estimator. We pinpoint some drawbacks of this approach: (1) a not unified estimator which needs periodicity test and period length inference in advance (2) an oversimplified linear description of aging progress (3) with no ability to distinguish between abrupt change and the “aging-like” gradual degradation. We design an enumersative Hodrick-Prescott filter to overcome all these shortcomings. And we also propose a new metric AS based on the nonlinear trend estimated by Hodrick-Prescott filter to dynamically measure severity of aging. Our approach and metric are validated on real aging time series collected from a VOD (video- on-demand) server. The results shows our approach improve the Sen’s slope estimator a lot.

Results in Real Data