No abstract

No abstract

Recent years have witnessed two major trends in the development of complex real-time embedded systems. First, to reduce cost and enhance flexibility, multiple systems are sharing common computing platforms via virtualization technology, instead of being deployed separately on physically isolated hosts. Second, multicore processors are increasingly being used in real-time systems. The integration of real-time systems as virtual machines (VMs) atop common multicore platforms raises significant new research challenges in meeting the real-time performance requirements of multiple systems. This paper advances the state of the art in real-time virtualization by designing and implementing RTXen 2.0, a new real-time multicore VM scheduling framework in the popular Xen virtual machine monitor (VMM). RT-Xen 2.0 realizes a suite of real-time VM scheduling policies spanning the design space. We implement both global and partitioned VM schedulers; each scheduler can be configured to support dynamic or static priorities and to run VMs as periodic or deferrable servers. We present a comprehensive experimental evaluation that provides important insights into real-time scheduling on virtualized multicore platforms: (1) both global and partitioned VM scheduling can be implemented in the VMM at moderate overhead; (2) at the VMM level, while compositional scheduling theory shows partitioned EDF (pEDF) is better than global EDF (gEDF) in providing schedulability guarantees, in our experiments their performance is reversed in terms of the fraction of workloads that meet their deadlines on virtualized multicore platforms; (3) at the guest OS level, pEDF requests a smaller total VCPU bandwidth than gEDF based on compositional scheduling analysis, and therefore using pEDF at the guest OS level leads to more schedulable workloads in our experiments; (4) a combination of pEDF in the guest OS and gEDF in the VMM -configured with deferrable server -leads to the highest fraction of schedulable task sets compared to other real-time VM scheduling policies; and (5) on a platform with a shared last-level cache, the benefits of global scheduling outweigh the cache penalty incurred by VM migration.

Abstract-Clouds have become appealing platforms for not only general-purpose applications, but also real-time ones. However, current clouds cannot provide real-time performance to virtual machines (VMs). We observe the demand and the advantage of co-hosting real-time (RT) VMs with non-real-time (regular) VMs in a same cloud. RT VMs can benefit from the easily deployed, elastic resource provisioning provided by the cloud, while regular VMs effectively utilize remaining resources without affecting the performance of RT VMs through proper resource management at both the cloud and the hypervisor levels. This paper presents RT-OpenStack, a cloud CPU resource management system for co-hosting 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) a realtime VM scheduler to allow regular VMs to share hosts with RT VMs without interfering 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 with regular VMs. Experimental results demonstrate that RTOpenStack can effectively improve the real-time performance of RT VMs while allowing regular VMs to fully utilize the remaining CPU resources.

Abstract-Massive computation power and storage capacity of cloud computing systems allow scientists to deploy computation and data intensive applications without infrastructure investment, where large application data sets can be stored in the cloud. Based on the pay-as-you-go model, storage strategies and benchmarking approaches have been developed for cost-effectively storing large volume of generated application data sets in the cloud. However, they are either insufficiently cost-effective for the storage or impractical to be used at runtime. In this paper, toward achieving the minimum cost benchmark, we propose a novel highly costeffective and practical storage strategy that can automatically decide whether a generated data set should be stored or not at runtime in the cloud. The main focus of this strategy is the local-optimization for the tradeoff between computation and storage, while secondarily also taking users' (optional) preferences on storage into consideration. Both theoretical analysis and simulations conducted on general (random) data sets as well as specific real world applications with Amazon's cost model show that the costeffectiveness of our strategy is close to or even the same as the minimum cost benchmark, and the efficiency is very high for practical runtime utilization in the cloud.