STARS: Startup-Time-Aware Resource Provisioning and Real-Time Task Scheduling in Clouds
Xiaomin Zhu, Huangke Chen, Guipeng Liu, Ling Liu · 2016
Green cloud computing has become a major performance measure for many infrastructure as a service (IaaS) cloud data centers. A popular way to reduce energy consumption for virtualized clouds is to dynamically consolidate virtual machines (VMs) and turning off as many idle hosts as possible. Upon the increase of the system's workloads, the closed hosts will be re-started to meet the scale-up demand of resources. However, the time overhead of starting hosts and deploying VMs can delay the start time of real-time tasks, and may cause deadline violation of some real-time tasks. This problem can be further aggravated for heterogeneous physical hosts. In this paper, we propose a novel scheduling architecture that allocates an idle lash-up VM on each active host. Besides, we develop a startup-time-aware scheduling strategy to scale up the resource provisioning for these lash-up VMs to mitigate the performance impact of host machine startup time on real-time tasks. Furthermore, we propose a startup-time-aware scheduling algorithm, named STARS, striving to guarantee deadlines of real-time tasks, while exploiting the optimal operating frequencies and energy efficiencies of heterogeneous hosts to achieve energy conservation. We conduct extensive experiments to validate the efficiency of STARS using Googles workload traces. The experimental results show that STARS outperforms the existing scheduling algorithms in terms of guarantee ratio (up to 19.10%), energy saving (up to 29.40%) and resource utilization (up to 46.69%).