[PAST EVENT] Pipelined Symbolic Taint Analysis on Multi-core Architectures

February 26, 2016
8am - 9am
Location
McGlothlin-Street Hall, Room 020
251 Jamestown Rd
Williamsburg, VA 23185Map this location
Friday, February 26, 2016
8am - 9am
McGlothlin-Street Hall, Room 020

Colloquium talk by Jiang Ming, Pennsylvania State University

Pipelined Symbolic Taint Analysis on Multi-core Architectures

The multifaceted benefits of taint analysis have led to its wide adoption in security tasks, such as software attack detection, data lifetime analysis, and reverse engineering. However, the high runtime overhead imposed by dynamic taint analysis has severely limited its adoption in production systems. The slowdown incurred by conventional dynamic taint analysis tools can easily go beyond 30X times. One way to improve performance is to parallelize taint analysis. Existing work has dramatically speeded up the analysis but has encountered a bottleneck. A key obstacle to effective parallelization is the strict coupling of program execution and taint tracking logic code. In this talk, I will present TaintPipe, a novel technique for parallelizing taint analysis in a pipeline style to take advantage of ubiquitous multi-core platforms. With the developed techniques, TaintPipe is able to significantly improve the performance of taint analysis and advance the state of the art, enabling broader adoption of information tracking technology. In addition, I will briefly introduce my research on formal program semantics-based methods for obfuscated binary code analysis and outline the future work ahead.

Biography:
Jiang Ming is currently a Ph.D. candidate in the College of Information Sciences and Technology of Pennsylvania State University, where he is a member of the Software Systems Security Research Lab. His research focuses on security, especially software security and malware defense, including secure data flow analysis, software plagiarism detection, malicious binary code analysis, and software analysis for security issues. Jiang Ming has extensive academic and industry experience in computer security. His work has been published in prestigious security and software engineering conferences (USENIX Security, CCS, Euro S&P, and FSE). He is among the first to work on symbolic execution based methods for semantics-based binary code diffing. More recently he has been working on the design of efficient and obfuscation-resilient binary code analysis techniques.