Events Calendar

EE 598 Cyber-Physical Systems Seminar Series

Speaker: Yuankun Xue, University of Southern California

Talk Title: Data-Centers-on-a-Chip as Enablers for Cyber-Physical Systems: A Scalable Model of Computation Guiding the Design Methodologies of Network-on-Chip based Manycore Platforms

Abstract: Synergistic coupling of physical and cyber processes with the goal of enabling a closed-loop control, calls for a paradigm shift in processing and mining the large amounts of cross-device data. One of the fundamental issues to be resolved targets the definition of new models of computation that allows us to integrate, interpret / mine and predict massive amount of multisystem data which requires a wide range of heterogeneous algorithmic description in order to provide accurate decision-making and control.

Towards this end, the complexity of the design-space exploration of large scale networks-on-chip (NoC)-based is exacerbated not only by the ever-increasing number of cores, but also by the increased runtime uncertainties in both the scale and task structure of the emerging applications. As a result, it is crucial to develop rigorous mathematical frameworks for capturing the task dependencies of varied applications to foster the generation of realistic benchmarks that can guide the NoC design. The current NoC benchmark suites either lack portability and poorly scale as they require intensive development efforts on specific architectures and simulation time, or are synthesized based on purely stochastic models that are disconnected from the characteristics of real applications, which may easily lead to biased and/or delayed design choices.

To address this challenge, we present in this talk a benchmark synthesis framework that not only allows extraction of dynamical task dependencies of the application and synthesize traffic workloads spatio-temporally consistent with realistic traffic behavior, but can also be easily scaled by the proposed complex network inspired metrics for large-scale benchmark generation while preserving key structural features that governs application communication behaviors. We validate the proposed framework via a comparative analysis on a realistic simulation environment by running a set of real application benchmarks. We show the synthesized benchmarks respect the traffic patterns of the original applications and preserve key features of application task structures. This newly proposed model of computation enables the efficient and accelerated design of future data-center-on-a-chip architectures for CPS infrastructures.

Biography: Yuankun Xue is a Ph.D student working under the supervision of Professor Paul Bogdan in the Ming Hsieh Department of Electrical Engineering at University of Southern California. He received his B.Sc and M.Sc degree from Fudan University in 2007 and 2011, respectively. His research interests include mathematical approaches for causal modeling, analysis and control of Cyber-Physical Systems, large-scale dynamic networked systems modeling, optimization and control, and design methodologies for high performance manycore platforms for computational biology.

Host: Paul Bogdan

Location: Hughes Aircraft Electrical Engineering Center (EEB) - 248

Audiences: Everyone Is Invited

Contact: Estela Lopez