Technical Report Number
Clouds have become appealing platforms for running not only general-purpose applications but also real-time applications. However, current clouds cannot provide real-time performance for virtual machines (VM) for two reasons: (1) the lack of a real-time virtual machine monitor (VMM) scheduler on a single host, and (2) the lack of a real-time aware VM placement scheme by the cloud manager. While real-time VM schedulers do exist, prior solutions employ either heuristics-based approaches that cannot always achieve predictable latency or apply real-time scheduling theory that may result in low CPU utilization. We observe the demand and advantage for co-hosting real-time (RT) VMs with non-real-time (regular) VMs in the same cloud. On the one hand, RT VMs can benefit from the easily deployed, elastic resource provisioning provided by a cloud; on the other hand, regular VMs can fully utilize the cloud without affecting the performance of RT VMs through proper resource management at both the cloud and hypervisor levels. This paper presents RT-OpenStack, a cloud management system for co-hosting both real-time and regular VMs. RT-OpenStack entails three main contributions: (1) integration of a real-time hypervisor (RT-Xen) and a cloud management system (OpenStack) through a real-time resource interface; (2) an extension of the RT-Xen VM scheduler to allow regular VMs to share hosts with RT VMs without jeopardizing the real-time performance of RT VMs; and (3) a VM-to-host mapping strategy that provisions real-time performance to RT VMs while allowing effective resource sharing among regular VMs. Experimental results demonstrate that RTOpenStack can support latency guarantees for RT VMs, and at the same time let regular VMs fully utilize the remaining CPU resources.
Xi, Sisu; Li, Chong; Lu, Chenyang; Gill, Christopher D.; Xu, Meng; Phan, Linh T.X.; Lee, Insup; and Sokolsky, Oleg, "RT-OpenStack: a Real-Time Cloud Management System" Report Number: WUCSE-2014-004 (2014). All Computer Science and Engineering Research.
Permanent URL: http://dx.doi.org/10.7936/K78P5XS0