HSAStore - A Hierarchical Storage Architecture for Computing Systems Containing Large-Scale Intermediate Data.

This paper introduces HSAStore, a newly designed storage system whose goal goes to build an efficient storage system for computing systems that containing large-scale intermediate data. HSAStore involves three sub-systems to work together, a distributed file system which is in charge of the intermediate data, a centralized Network Attached Storage (NAS) which stores the raw input data and the results data, and a local file system that serves for the local data. The HSAStore takes full advantage of the network bandwidth of the computing cluster. Moreover, HSAStore adopts the distributed file system, so it is helpful for efficient execution of parallel programs. Experiments show that HSAStore has a significant improvement on efficiency in the computing systems that containing large-scale intermediate data.