Extension of S3 REST API for Providing QoS Support in Cloud Storage

In the IaaS (Infrastructure as a Service) cloud environment where virtual machines are deployed, a scalable and reliable storage service is indispensable for hosting virtual machine images, backups, large volumes of application data, etc. Amazon S3 (Simple Storage Service) [3] is one of the storage services to meet such demands for Amazon EC2 [2] users. Swift [9] takes the similar role in OpenStack [7] and provides the S3 compatible interface in addition to its own interface. Although the S3 implementation considers scalability, there is a concern for performance instability when many applications heavily access S3, due to the characteristics of shared resources. Replication and load balancing mitigate the problem but they might not respond every individual request of the applications. Amazon EBS [1] which delivers dedicated performance to the application would not be reasonable for storing large data, in terms of the financial cost and the limitation of shared use. It would be useful that users store large data in S3 for a long term, with low cost, and only when they request with more payment, they can access the data with a certain level of performance. This paper presents an approach which allows users to have specific I/O throughput in use of S3, by making an advance performance reservation. The reservation is intended to be used as a user’s explicit request to S3 in the above. To provide this feature, a storage system which supports the performance reservation is used at the S3 backend, and I/O operations of S3 REST API are extended to be conformed to the reservation-based access to the storage system. Both S3 client and server prototypes are implemented and then evaluated to examine the QoS capability at the S3 data transfer, under the situation where multiple S3 clients access the same resources.