Automated performance attack discovery in distributed system implementations

Hyojeong Lee, Purdue University

Abstract

Security and performance are critical goals for distributed systems. The increased complexity in design, incomplete expertise of developers, and limited functionality of existing testing tools often result in implementations with vulnerabilities and make the debugging process difficult and costly. The deployed vulnerabilities are often exploited by adversaries preventing the system from achieving its design goals. We refer to attacks that slow down the performance of a system as performance attacks. In the past, finding performance attacks has been a painstaking manual process that involved an expert of the target implementation. Given the cost associated with each vulnerability that occurs in the production, there is a need for tools to automatically check that the implementation of a protocol achieves its performance goals with respect to malicious components in the system. In this dissertation, we find performance attacks automatically from implementations of distributed systems. We do not try to show that an implementation is free from all attacks. Our goal is to find attacks and report them to the user in a timely manner. We first investigate how to find attacks automatically from implementations under a simulated environment. A simulated approach, however, has a fundamental limitation in terms of applicable target systems, as certain assumptions are made about languages, operating systems or libraries used. Therefore, we next investigate challenges and requirements to automatically find attacks in implementations of distributed systems under an emulated environment where no limiting assumptions are made.

Degree

Ph.D.

Advisors

Killian, Purdue University.

Subject Area

Computer science

COinS