Arts & Sciences Events
This calendar presented by
Arts & Sciences
[PAST EVENT] Zhijia Zhao, Computer Science - Ph.D. Defense
June 29, 2015
1:30pm - 3:30pm
Abstract:
As the speed of uni-processors has reached a barrier, the architectures of modern computing systems, ranging from high performance servers to low-end handheld devices, embrace parallel hardware (e.g. multicore and GPU) for continuous performance gain.
However, whether the applications can take advantages from the underlying hardware parallelism depends on whether they can be effectively parallelized. The parallelization difficulties of applications vary a lot, ranging from trivial to extremely hard. In this dissertation, three important types of computations that lay on the extremely hard end are investigated, including finite state machine-based computations, HTML parsing and just-in-time compilation. The major challenges come from the various internal dependencies carried with these computations. To overcome the challenges, a novel speculative parallelization with systematic and rigorous analysis has been explored, namely principled speculative parallelization. Results show significant performance gain either over previous heuristic-based methods, or over default sequential implementation in cases that no earlier parallel versions is known.
Bio:
Zhijia Zhao is a Ph.D. candidate at the College of William & Mary. He has been advised by Professor Xipeng Shen in the Department of Computer Science. His research lies in the broad field of programming systems and parallel computing, with an emphasis on enabling effective parallel execution through principled approaches. His research includes collaborations with Intel Labs, IBM Research and Mozilla Foundations.
As the speed of uni-processors has reached a barrier, the architectures of modern computing systems, ranging from high performance servers to low-end handheld devices, embrace parallel hardware (e.g. multicore and GPU) for continuous performance gain.
However, whether the applications can take advantages from the underlying hardware parallelism depends on whether they can be effectively parallelized. The parallelization difficulties of applications vary a lot, ranging from trivial to extremely hard. In this dissertation, three important types of computations that lay on the extremely hard end are investigated, including finite state machine-based computations, HTML parsing and just-in-time compilation. The major challenges come from the various internal dependencies carried with these computations. To overcome the challenges, a novel speculative parallelization with systematic and rigorous analysis has been explored, namely principled speculative parallelization. Results show significant performance gain either over previous heuristic-based methods, or over default sequential implementation in cases that no earlier parallel versions is known.
Bio:
Zhijia Zhao is a Ph.D. candidate at the College of William & Mary. He has been advised by Professor Xipeng Shen in the Department of Computer Science. His research lies in the broad field of programming systems and parallel computing, with an emphasis on enabling effective parallel execution through principled approaches. His research includes collaborations with Intel Labs, IBM Research and Mozilla Foundations.
Contact
[[vlthompsondopp]]