Virtualization of Micro-architectural Components Using Software Solutions

Cloud computing has become a dominant computing paradigm in the information technology industry due to its flexibility and efficiency in resource sharing and management. The key technology that enables cloud computing is virtualization. Essential requirements in a virtualized system where several virtual machines (VMs) run on a same physical machine include performance isolation and predictability. To enforce these properties, the virtualization software (called the hypervisor) must find a way to divide physical resources (e.g., physical memory, processor time) of the system and allocate them to VMs with respect to the amount of virtual resources defined for each VM. However, modern hardware have complex architectures and some microarchitectural-level resources such as processor caches, memory controllers, interconnects cannot be divided and allocated to VMs. They are globally shared among all VMs which compete for their use, leading to contention. Therefore, performance isolation and predictability are compromised. In this thesis, we propose software solutions for preventing unpredictability in performance due to micro-architectural components. The first contribution is called Kyoto, a solution to the cache contention issue, inspired by the polluters pay principle. A VM is said to pollute the cache if it provokes significant cache replacements which impact the performance of other VMs. Henceforth, using the Kyoto system, the provider can encourage cloud users to book pollution permits for their VMs. The second contribution addresses the problem of efficiently virtualizing NUMA machines. The major challenge comes from the fact that the hypervisor regularly reconfigures the placement of a VM over the NUMA topology. However, neither guest operating systems (OSs) nor system runtime libraries (e.g., HotSpot) are designed to consider NUMA topology changes at runtime, leading end user applications to unpredictable performance. We presents eXtended Para-Virtualization (XPV), a new principle to efficiently virtualize a NUMA architecture. XPV consists in revisiting the interface between the hypervisor and the guest OS, and between the guest OS and system runtime libraries so that they can dynamically take into account NUMA topology changes