Quantifying Performance Determinism in Virtualized Mixed-Criticality Systems
Andrew Jerome Hughes, Amro Awad · 2019
Virtualization has been a leading technology for managing shared infrastructures. Virtualization reduces infrastructure cost, enhances security, provides isolation and flexibility for migrating/upgrading systems. However, as virtualization is being pushed into defense and mission-critical applications, new challenges are emerging. For instance, U.S. Department-of-Defense (DoD)s recent initiative to deploy a cloud-based Joint-Enterprise Defense Infrastructure (JEDI) brings in several questions, mainly, how to ensure deterministic performance for mission-critical applications in a virtualized environment. Similar to avionic systems with mixed-criticality tasks, mission-critical applications and systems require lengthy software certifications that assess the determinism and guaranteed execution time of such applications. Unfortunately, we observe several challenges on leveraging cloud for time-critical jobs. Thus, in this paper, we characterize and investigate the determinism of performance for applications on cloud systems and compare between state-of-the-art schedulers and real-time implementations.