Arts & Sciences Events
[PAST EVENT] Hao Xu, Computer Science - Dissertation Defense
Abstract:
Modern computer systems have evolved to employ powerful parallel architectures, including multi-core processors, multi-socket chips, large memory subsystems, and fast network communication. Given such powerful hardware, developers rely on performance profiling and modeling to guide their performance optimization. However, performance optimization is facing new challenges on efficiency and accuracy with emerging computer systems. In this dissertation, we propose approaches to address these challenges.
We first study memory contention in Non-Uniform Memory Access (NUMA) architectures. We present DR-BW, a new tool based on machine learning to identify bandwidth contention in NUMA architectures and provide optimization guidance. DR-BW collects performance data with low overhead (<10%), feeds the data into a novel machine learning model to identify contention achieving more than 96% accuracy, and associates the analysis results with both programs and significant data objects.
Next, we study and fix inaccuracy measurements in modern profilers. We investigate multiple modern architectures and quantify the PMU instruction profiling inaccuracy in these architectures with mathematical modeling. Then we design a systematic framework to evaluate the impact of PMU inaccuracy on the profiling results. We propose a software-based technique to rectify the measurement inaccuracy raised by PMU and demonstrate its effectiveness.
Finally, we focus on admission control to avoid overload for low-latency online data systems. Various query types, fast query processing, and long-tail latency distribution bring issues to admission control policy design. In order to meet different requirements (no underuse of the system, no inadvertent starvation of queries, fast rejections and low overhead), we proposed SLO (Service Level Objective) based admission control policy.
Bio:
Hao Xu has been working on his Ph.D. degree in the Department of Computer Science at William & Mary since Fall 2014. He is working with Dr. Xu Liu in the fields of building profiling tools for performance optimizations. Hao Xu got his M.S. in 2014 from University of Chinese Academy of Sciences, China, and B.S. in 2011 from Wuhan University of Technology, China.