Sliding window-based multi-priority metadata task scheduling method

Abstract

The invention discloses a sliding window-based multi-priority metadata task scheduling method. The method comprises the following steps of setting three task priorities of metadata, namely high priority, medium priority and low priority; setting expected response time delay of metadata tasks with high priority, medium priority and low priority; calculating metadata task expected scheduling time, and sorting the metadata task expected scheduling time from long to short; setting a sliding window, preferentially scheduling high-priority metadata tasks in the sliding window, and preferentially scheduling tasks with short expected scheduling time for metadata tasks with the same priority; shifting the sliding window, and completing task scheduling if a buffering queue is empty. According to the method, the metadata tasks with high priority are scheduled early, and the metadata tasks with the same priority are scheduled according to a first-in first-out scheduling policy; as the priority scheduling is limited in the range of the sliding window, the advantage that high-priority metadata are scheduled preferentially is reflected and the scheduling equity is considered.

Description

technical field [0001] The invention relates to a scheduling method for metadata tasks in a cloud computing environment, in particular to a sliding window-based multi-priority metadata task scheduling method. Background technique [0002] Cloud computing has achieved rapid development in recent years. Based on the scale effect and technical support provided by top IT companies, it provides enterprises and users with economical, reliable, and convenient storage and computing services. With the popularization of cloud computing services, various cloud computing platforms have emerged. Typical cloud computing platforms such as Google's GFS, Microsoft's Windows azure, and Amazon's Amazon all provide cloud computing services, and the number of cloud users is also increasing year by year. At present, most cloud computing platforms adopt a master-slave structure similar to HDFS (Hadoop Distribute File System). The master node NameNode is the metadata server, which manages the metad...

AI Technical Summary

Problems solved by technology

The first-in-first-out scheduling algorithm is suitable for single-buffer queues. It is simple to implement, but there is no classification; the priority scheduling algorithm is strictly scheduled according to the priority level, which realizes QoS distinction, but may cause low-level requests to be unresponsive for a long time, resulting in "fake death". ", unfair

Images