Group-based Energy Adaptation for Storage Systems

Storage systems play a significant role in data centers and there is an urgent need to efficiently store, retrieve and manage the ever increasing volume of data required by a variety of applications in the data center. Much of the stored data often contains a lot of redundancies at the block level that can be removed via de-duplication. The performance and fault-tolerance requirements also need explicit replication of data, but more copies mean higher storage system energy consumption. In our previous work we proposed a flexible storage infrastructure called flexStore that can dynamically control the replication of de-duplicated data based on changing energy budgets for the storage subsystem. In this paper we extend this mechanism with storage policies that allow for differentiated treatment of various applications. In particular, we consider replication of virtual machines belonging to different application groups that are managed independently with respect to both de-duplication and replication. We have built a prototype of the storage system and evaluate the proposed system on an Amazon EC2 cluster. Through this prototype we study the benefits of group based replication both on storage node and on the host side in a data center.