ABOUT THE CALENDAR
TRUST Security Seminar: Scheduling with Privacy Constraints
Seminar | November 29 | 1-2 p.m. | Soda Hall, Wozniak Lounge
Negar Kiyavash, University of Illinois at Urbana-Champaign
In systems where a finite resource is to be shared, a scheduler dictates how the resource is divided among competing processes. Examples of such systems include, a computer where the CPU is shared between the different threads running, a cloud computing infrastructure with shared computing resources, a network router serving packets from different streams etc.. In such situations, when a processor is shared by multiple users, the delays experienced by jobs from one user are a function of the arrival pattern of jobs from other users, and the scheduling policy of the server. Consequently, a scheduling system creates a timing side channel in which information about arrival pattern from one user is inadvertently leaked to another. In this talk, this information leakage is studied for a two user scheduling system. We demonstrate that commonly used scheduling policies such as the first come first served and round-robin fare poorly in terms of preserving a user's privacy. That is, a malicious attacker can learn the arrival pattern from the other user accurately if one of these scheduling policies is used. Furthermore, we prove that no scheduler can provide maximum privacy without idling/taking vacations, and consequently no policy can simultaneously be delay and privacy optimal. We also propose a scheduling policy called the accumulate and serve, a parametrizable throughput optimal scheduling policy that trades-off delay for enhanced privacy.
EECS Home | Contact WebTeam |
Copyright © 2014 UC Regents