| Summary: | 碩士 === 國立臺灣科技大學 === 資訊管理系 === 102 === Power consumption of smartphones is attached more importance. Most testing tools focus on the overall power consumption of smartphones (system-level power consumption), but few testing tools focus on the power consumption of applications (process-level power consumption). Therefore, constructing a process-level power model to increase the accuracy and efficiency is an important issue. Previous work, Eprof, uses a measurement tool to fetch power consumption, associate with system calls, and manually construct Finite State Machine (FSM), so it is an ineffective way. Our main contribution is automatically constructing FSM to raise construction efficiency and accuracy of FSM. This thesis provides a method, called Automatically Constructing FSM (ACFSM), to measure process-level power consumption. ACFSM fetches system call and maps power consumption, eliminate power noises, filter useless system calls, and solve Tail Power (TP) problem to automatically constructing FSM, where TP is that a hardware component still consumes power during a time period after it stops. Experiment results are divided in two parts: efficiency and accuracy. For efficiency, ACFSM needs 413 seconds but manually constructing FSM needs 1393 seconds, i.e., 70% improvement. For accuracy, the error rate of solving TP is 5.3% but that without solving TP is 10.9%. That is, ACFSM can achieve 51% improvement on the error rate, compared with the approach without considering TP.
|
|---|
