Saga: A Cost Efficient File System Based on Cloud Storage Service

摘要

Cloud Storage service providers such as Amazon Simple Storage Service (S3) and Google Storage for Developers offer low-cost and highly available scale storage resource with a simple pay-as-you-go charging model. The cost of running storage systems on such a Cloud Storage service mainly depends on occupied storage space, number of requests and amount of data transfer. Traditional design of storage stack based on disk driver or tape didn't consider cost as a system metric, hence it brings considerable optimization space for the design of storage system based on Cloud Storage. In this paper we propose Saga, a user mode file system based on Cloud Storage service, that is designed to support POSIX interface with the goal of minimizing cost. Saga is specially designed under the cost efficient principle that minimizes occupied storage space by store-one-copy and copy-on-write strategies and minimizes number of requests by distinguishing objects loaded by write or read requests. Saga is also efficient from a performance perspective and utilizes parallel characteristics of Cloud Storage to boost the performance. Experimental results show that Saga is cost efficient and works well with general-purpose I/O workloads.