I/O Deduplication in Smartphones

Abstract: Previous studies have shown that storage is one of the key factors affecting the overall system performance of the Android-based Smartphones. In details, the storage stack of current Smartphones faces three challenges. First, the performance tends to degrade after repeated usages, particularly writes, due to the physical characteristics of the flash memory, also one of the reasons why Smartphones slow down over time. Second, one of these physical characteristics of flash memory is its limited life cycles, i.e., the number of times each cell can be programmed/written before it fails, and causes the flash storage to get sluggish after repeated usages, which affects the storage reliability of Smartphones. Third, the cost of upgrading the flash capacity from one level to the next level, e.g., from 16GB to 32GB, amounts to nearly 100 USD for most Smartphones. Therefore, these challenges, pointing to the measures of performance, reliability and cost, suggest that it is important to (1) understand the mobile applications and how they interact with the flash-based eMMC; (2) optimize to reduce write traffic to the flash-based eMMC in Smartphones.

Traces: Based on the research objectives, we first develop a low-overhead content-aware trace collection tool in Smartphones: MobileCT. Using MobileCT, we collected traces of 15 popular mobile applications: MobileTraces and the README file. All the 15 mobile applications are running on the Google Nexus 5 smartphone with Android 5.0.1 and Linux Kernel 3.4. The trace file-names are indicating different mobile applications and some file-names also include the running hours. Each record in the trace files is as follows:

[Time in s] [R or W] [LBA] [size in 512 Bytes blocks] [pid] [process] [MD5 per 4096 Bytes]

More details about the mobile applications and how the traces are collected can be found in our published papers, as follows:

Bo Mao, Suzhen Wu, Hong Jiang, Xiao Chen, and Weijian Yang. Content-aware Trace Collection and I/O Deduplication for Smartphones. In Proceedings of the 33rd International Conference on Massive Storage Systems and Technology (MSST'17), Santa Clara, CA, USA, May 15-19, 2017.

Bo Mao, Jindong Zhou, Suzhen Wu, Hong Jiang, Xiao Chen, and Weijian Yang. Improving Flash Memory Performance and Reliability for Smartphones with I/O Deduplication. IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems. 38(6): 1017-1027, June 2019.

Project: Our preliminary study is APP-Dedupe which leverages data redundancy characteristics and I/O deduplication to reduce the write traffic on eMMC storage for Android-based Smartphones. More details about the APP-Dedupe can be found in the papers. The system image of APP-Dedupe is available for research purpose upon request.