Data analysis-based capacity planning of VCL clouds
Ágnes Salánki, Gergo Kincses, László Gönczy, Imre Kocsis · International Journal of Cloud Computing · 2017
Virtual computing labs dramatically changed education methodology with transforming traditional classroom-and lab-based learning models to self paced asynchronous ones. The Apache Virtual Computing Lab (VCL) platform allows students to reserve and use virtual machines (VMs) with a predefined configuration and software setup. In essence, it offers an educational cloud that provides preconfigured lab environments in 'desktop as a service' style. At our university, four courses of a specialisation branch are available in this form. While maintaining VCL, we faced the challenges of short-and long-term capacity planning. We analysed high-level reservation and platform-level monitoring data of five semesters and built mathematical models of workload and resource utilisation based on our observations. The main contribution of this study are data-driven approaches for: 1) predicting reservation patterns of students as course deadlines approach; 2) a regression-based estimate of typical resource utilisation of VMs; 3) elaboration of an optimised schedule of deadlines to avoid rejected reservation queries or a burst out to a public cloud. Applying these methods, fine-tuning of VM configurations and scheduling of upcoming semesters became possible, even in case of methodical/technical educational changes (e.g., modified course schedules, increasing number of attendees).