Research on improving disk throughput in EC-based distributed file system

Abstract

The development of the Fourth Industrial Revolution resulted in an increase in data type and size, and distributed file systems emerged to store them. Among them, Replication techniques divide the data that you want to store into certain blocks and replicate the divided blocks to store them distributed across multiple nodes. However, there was a problem with increasing the disk's capacity to store the replicated blocks. Thus, the Erasure Coding technique emerged, and the EC-based distributed file system improved the space efficiency issue over the Replication technique because it creates blocks to be stored through encoding and parity blocks to be used for recovery. However, EC-based distributed file system has caused disk disk write throughput problems to access a number of disks, causing system performance degradation. Therefore, this paper proposes Buffering and Combining techniques to improve disk write throughput problems in EC-based distributed file systems.