Cloud Autoscaling Digital Twin Using Workload Forecasting

Autoscaling a web tier trades server cost against slow responses. This project builds a queueing digital twin of a simulated web tier and feeds it a public trace: 17,544 hours of English Wikipedia page views from the Wikimedia REST API, scaled to a 1 % traffic share. Forecasters predict the next hour, an M/M/c (Erlang-C) model turns a rate into the smallest server count that keeps the mean wait within 100 ms, and static, reactive, HPA-formula and predictive policies are compared on a time-ordered test year. The plain finding: the policy selected on validation was the simple baseline persistence_q90c, not a machine-learning model. Forecasts earn their keep when scale-up takes time: the reactive rule collapses at 1-2 h delays while forecast policies hold.

Category: Information Science and Technology

Information Science & Technology Project IST04 from ProTechSC Academy, with graphical abstract, video demonstration and project guidance.