Energy efficient IO stack design for wearable device

Abstract

In this paper, we visit the energy consumption issue of the IO subsystem for wearable device. We argue that the IO stack for wearable device is subject to extreme inefficiency in terms of the IO volume and the number of flush requests. The deficiency of the IO stack leaves a room for improvement from the energy consumption's point of view. In this work, we characterize the IO access patterns of the smartwatch device, develop a model to estimate the energy consumption from the given IO traces, and propose a set of methods to optimize the IO subsystem behavior for energy saving. In smartwatch, the amount of data written daily is 10× as large as the amount of data read daily. The amount of data written to the flash storage each day is approximately as large as the free space in the storage device. To minimize the energy consumption associated with the IO activities in smartwatch, we propose Metadata Embedding and Selective Directory Sync for SQLite DBMS and Flushless Durability Guarantee for the filesystem. We implement the proposed techniques in the commercially available smartwatch product. The proposed techniques reduce the energy consumption associated with the IO activities by 60%. It corresponds to 3% savings in the overall energy consumption. It is achieved solely via software optimization.