Cache-Aware Real-Time Virtualization

Virtualization has been adopted in diverse computing environments, ranging from cloud computing to embedded systems. It enables the consolidation of multi-tenant legacy systems onto a multicore processor for Size, Weight, and Power (SWaP) benefits. In order to be adopted in timing-critical systems, virtualization must provide real-time guarantee for tasks and virtual machines (VMs). However, existing virtualization technologies cannot offer such timing guarantee. Tasks in VMs can interfere with each other through shared hardware components. CPU cache, in particular, is a major source of interference that is hard to analyze or manage. In this work, we focus on challenges of the impact of cache-related interferences on the real-time guarantee of virtualization systems. We propose the cache-aware real-time virtualization that provides both system techniques and theoretical analysis for tackling the challenges. We start with the challenge of the private cache overhead and propose the private cache-aware compositional analysis. To tackle the challenge of the shared cache interference, we start with non-virtualization systems and propose a shared cache-aware scheduler for operating systems to co-allocate both CPU and cache resources to tasks and develop the analysis. We then investigate virtualization systems and propose a dynamic cache management framework that hierarchically allocates shared cache to tasks. After that, we further investigate the resource allocation and analysis technique that considers not only cache resource but also CPU and memory bandwidth resources. Our solutions are applicable to commodity hardware and are essential steps to advance virtualization technology into timing-critical systems