Period Estimation for Linux-based Edge Computing Virtualization with Strong Temporal Isolation

Virtualization of edge nodes is paramount to avoid their under-exploitation, allowing applications from different tenants to share the underlying computing platform. Neverthe-less, enabling different applications to share the same hardware may expose them to uncontrolled mutual timing interference, as well as timing-related security attacks. Strong timing isolation through SCHED_DEADLINE reservations is an interesting solution to facilitate the safe and secure sharing of the processing platform: nevertheless, SCHED_DEADLINE reservations require proper parameter tuning that can be hard to achieve, especially in the case of highly dynamic environments, characterized by workloads that need to be served without knowing any accurate information about their timing. This paper presents an approach for estimating the periods of SCHED_DEADLINE reservations based on a spectral analysis of the activation pattern of the workload running in the reservation, which can be used to assign and refine reservation parameters in edge systems.