Research On Real-Time Tasks Scheduling And Resource Dynamic Adjustment In Cloud Service System

In the face of massive information for information battlefield and combat unit application needs in a high dynamic concurrency, cloud computing shows a promising way for the battlefield information service mode. Cloud computing has become the latest trends for distributed computing. Leveraging the advanced virtualization technology, the large-scale resource, e.g., computing, storage and network etc., are configured into huge pool of resources to provide on-demand service for users. In cloud computing paradigm, users only need to submit tasks to a cloud service system, the cloud service system will automatically analysis tasks' characteristics, prediction their resource requirements, and then schedule them to computing resources according to the status of the underlying resources in cloud service system, finally the result of the task execution will be returned to the users. That is, users only need to submit tasks, quality of service and will receive results of tasks execution, and everything in the middle, the system will be automatically completed by the cloud service system. It is noting that efficient task and resource scheduling mechanism is one of the key techniques to improve the performance of the cloud services system.Currently, there exist many works that exploring the task and resource scheduling in cloud service system. However, the existing works mainly focus on the ideal scheduling environment: 1) the task set is known in advance; 2) the execution times of tasks are deterministic, and are available before scheduling; 3) the resources are immediately available. However, in the actual cloud service system, there are a large number of dynamic, stochastic factors. For example, dramatic changes in the tasks' arrival rate, tasks' execution times are stochastic, resources can be dynamically scaled up and down, and starting the resource causes time overhead. These dynamic and stochastic nature of cloud service system, often make the pre-computed scheduling scheme lost their original advantages and even is no longer feasible. Therefore, the real-time tasks scheduling and resource dynamic adjustment in cloud service system has theoretical and practical value, and challenging.This paper mainly studies the real-time tasks scheduling and resource dynamic adjustment in cloud service system regarding the following three typical cases: tasks arrive dynamically, tasks' execution times are stochastic, the startup times of hosts and virtual machines cannot be ignored. In this paper, the main research work and innovation include the following four points:(1) Propose a scalable host organization mode. For a cloud service system, it consists of large scale hosts, which challenges the traditional host organization mode, e.g., the centralized, hierarchical and distributed. This paper puts forward the cooperative organization mode, where the large-scale hosts are divided into multiple clusters, each cluster has a separate scheduler, who is responsible for task and resource scheduling in this cluster. In addition, schedulers work collaboratively to schedule the tasks and the underlying resources, so as to improve the scalability of the cloud service system.(2) Propose a randomness-aware scheduling framework. In view of the high dynamic, stochastic and high timeliness requirements of real-time tasks in the cloud service system, propose an uncertainty-aware scheduling framework for each cluster. In this framework, most waiting tasks are waiting in the global queue, and at most one task is allowed to wait on the local queue of each virtual machine. When a virtual machine complete the executing task, the wait task in local queue will be executed immediately, and then the task with the higher timeliness requirements in the global queue will be preferentially scheduled to the virtual machine to wait, that avoiding the randomness of the finished task affecting the current scheduling tasks, so as to improve the ability of keeping the scheduling scheme stable and the timeliness of real-time tasks.(3) Propose a randomness-aware scheduling algorithm, named PRS. Aiming at high dynamic, stochastic and high timeliness requirements of real-time tasks in the cloud service system, this paper incorporates both proactive and reactive scheduling methods to make an online algorithm PRS, based on the randomness-aware scheduling framework. Algorithm PRS continuously creates new task and virtual machine scheduling scheme for cloud service system according to its actual situation, so as to ensure the timeliness requirements of real time tasks, and improve the resource utilization of hosts and reduce the energy consumption for cloud service system.(4) Propose a machine startup time aware algorithm STARS to realize the inter-adaptive scheduling of real-time tasks and virtual machines. For cloud service system, the arrival of real-time tasks is random and sudden, after its load suddenly surges, starting the host and creating virtual machine process will cause some time overhead, so that some tasks cannot be timely started, thus delay their deadlines. In view of the above problems, this paper puts forward the machine startup time aware scheduling algorithm STARS. Leveraging the elasticity of a single virtual machine's CPU performance, algorithm STARS can transfer the impact of machine startup time on tasks with shorter deadlines, so as to mitigate the effect of machine startup time on increasing task timeliness, that improve ability of guaranteeing the real-time tasks' timeliness for cloud service system.