欢迎访问广东工业大学数据挖掘与信息检索实验室!

UTSA Tongping Liu博士学术讲座预告

作者:DMIR    来自:    发表时间:2015-12-15    浏览量:1062


报告题目:  Deterministic Systems to Defeat Reliability Problems

报告时间20151216日(周三)上午1000

报告地点:  大学城校区工学一号馆725室(DMIR实验室)

报告人Tongping Liu 博士

报告人简介:

      Tongping Liu is an Assistant Professor at the University of Texas at San Antonio. He got his Ph.D. from the University of Massachusetts Amherst in 2014. His research spans runtime systems, operating systems, programming languages, compiler, and distributed systems. His primary research goal is to practically improve the reliability and performance of parallel software. His work appeared in those most prestigious system conferences, such as SOSP, OSDI and OOPSLA. He has been awarded a 2015 Google Faculty Research Award for his work in improving the performance of multithreaded programs.

报告摘要:

    The advent of multi-cores drives the biggest revolution of software development - parallel programming. Inherent non-determinism inside parallel programs can greatly complicate the debugging, testing, and reproducing of program errors. We develop an efficient deterministic system, Dthreads, to defeat this non-determinism problem of multithreaded programs. Dthreads enforces that a program will always generate the same executions if fed with the same input, even in the face of data races. Dthreads explodes multithreaded applications into multiple processes with private mappings, then uses standard virtual memory protection to track writes, and deterministically orders updates by each thread. Dthreads substantially outperforms the existing deterministic systems, and provides a great foundation for all future research in this field. Dthreads appeared at SOSP 2011, and has been cited over 150 times after its publication.
      Programs written in unsafe languages like C and C++ often suffer from errors like buffer overflows, dangling pointers, and memory leaks. Dynamic analysis tools like Valgrind can detect these errors, but their high overhead makes them too heavyweight for use in deployed applications. We develop an evidence-triggered dynamic analysis (called as "DoubleTake") with less than 5\% overhead, making it practical in deployed settings for the first time. DoubleTake allows a program to execute at nearly full speed until errors are found. It then employs the deterministic execution, with more heavy instrumentation, to pinpoint the locations of errors. DoubleTake is practical and convenient to deploy, requiring neither custom hardware, compiler, nor operating system support.




欢迎广大教师和学生参加!