Dynamic And Parallel Scheduling Of Dependent Tasks In Cloud

Now cloud computing has become a hot research in industry and academia, and gradually developed into the most promising business computing model. Cloud computing relies on virtualization technology and bases on distributed computing and grid computing. Its purpose is to provide reliable, personalized and QoS-guaranteed services. The cloud computing environment provides such a wide range of services, so how to schedule or execute services according to the needs of users and to ensure the quality of service is one of the key issues to be resolved.Most applications in cloud system must be broken down into several sub-tasks, and the decomposed sub-tasks have dependency between each other. How to improve the parallelism, real-time and dynamic nature of dependent task scheduling, and to improve system utilization, make a reasonable dependent task scheduling and deployment plan become one of the key issues in distributed and cloud computing. For dependent task scheduling problem, the various existing algorithms usually based on different model assumptions. These early scheduling model assumptions do not meet cloud environment characteristics. And most of the scheduling algorithms are based on the classical list scheduling ideas. They can not do dynamic scheduling based on real-time system resources' information; and ignored parallelism of dependent tasks, making system resources can not be fully utilized to improve system utilization.In order to better solve the dependent task scheduling in cloud, this paper presents a system model for cloud environment, which allows calculating nodes to join and leave system dynamically and improve self-organization and scalability of the whole system. It is more suitable for practical applications. Then based on this model, a dynamic and parallel scheduling mechanism TDDPS for dependent tasks is built. Firstly, the whole model is prescribed, including the scheduling problem description and resource matching. And in order to improve resource matching efficiency through the scheduling process, a multi-pattern matching automata approach is provided. Then introduce the dynamic and parallel scheduling mechanism in detail. It decoupled tasks'dependency through the O in-degree of tasks and dynamically built a ready task set to describe tasks can be parallel scheduled. And then use real-time system resource information to make a distributed consultation scheduling in parallel, effectively improve parallelism of scheduling process. When distribute tasks, we also take into account execution and communication overhead (E/C) between tasks to decide whether to replace part of data transfer with task replication in order to reduce communication overhead. This scheduling mechanism can make dynamical and parallel distributed multi-objective consultation scheduling. It considers the real-time, communication overhead and load balance performance, and can improve overall system performance through dynamic scheduling policy.