Research On IND Task Scheduling Model In Intelligent Network Disk Storage System

Thanks to the rapid development of computer technology and the Internet technology, the requirement of the data storage has been changed a lot, including needing massive storage capacity, hoping to quickly access the data and effective management. Many domestic and foreign research institutions and the well-known large enterprises are paying increasing attention to storage technology, especially in recent years the network storage technology has become the "hot spots" of storage area.The intelligent network disk storage systems proposed by our team is commitment to multi-level intelligence and automation. Task scheduling is one of the key problems to reflect the intelligence of IND. Task execution time prediction is a prerequisite of effective task scheduling. In this paper, the task scheduling model mainly consists of two aspects: IND execution time prediction model and the task scheduling algorithm design of single IND. The author's main research work is done by the following:As IND storage system has high reliability, high availability features and the IND applications very similar, the framework of IND execution time prediction model proposed by this paper is mainly established by using historical information. In the specific prediction process, we draw on the vector space model (VSM) ideas and combine with the methods of statistics including the DF, the advantage ofχ2 to obtain the characteristics of the task of metadata items. Exploiting TF-IDF strategy to calculate feature items weight, we improve the TF-IDF in the specific prediction process. In the prediction execution time algorithm, we adopt multiple linear regressions modeling to obtain the most similar task execution time and then correct the execution time by dynamic resources factors. Experiments show that the IND execution time prediction model can achieve a higher accuracy.The adopted scheduling algorithm of the Intelligent Network Disk (IND) storage system can minimize the weighted cost time of tasks, but it results in anther problem that the response time of some tasks is longer. To overcome the starvation phenomena, an improved algorithm named Controlled Scheduling algorithm (CSA) is proposed, which computes theρfactors of tasks by the execution time and weight of step chains, and selects the unexecuted step chain in terms ofρfactor and the executed time of task ofρfactor. Experimental results exhibit that the improved algorithm, compared with the adopted algorithm of IND, decreases the completion time of short tasks and the average turnaround time by sacrificing little weighted cost time. CSA solves the starvation phenomena at a certain extent.