Data Scheduling Based on Data Label in Hybrid Storage Architecture

Carrying out high efficient and rapid analysis of big data is essential to big data application. Due to the poor scalability of DRAM, the performance of big data analysis and related applications is difficult to improve. DRAM/NVM hybrid storage architecture has the advantages of non-volatile and high storage density, which brings an opportunity to optimize big data analysis. Because the task itself depends on the data and does not modify the data, it is possible to solve the problem of operation delay if the data is deployed well on the storage system under the background of hybrid storage architecture. In order to optimize the problem of high latency, this paper discusses the data migration between disk and NVM and proposes a data deployment algorithm based on data label. The validity of labeling is verified by calculating the total time of reading data by tasks in the experiment and the efficiency of task execution is improved.