Capacity Metric for Chip Heterogeneous Multiprocessors


Mof Otoom, Yarmouk University, Jordan -- 26-06-2013


Metrics are required in order to evaluate any system, including computer systems. A lack of appropriate metrics can lead to ambiguous or incorrect results. For many decades, computer architects have focused on techniques that reduce latency and increase throughput. The change in modern computer systems built around Chip Heterogeneous Multiprocessors (CHMs) that process multiple, variable heterogeneous workloads in the service of single users calls this focus into question. Modern computer systems are expected to integrate tens to hundreds of processor cores onto single chips, often used in the service of single users, potentially as a way to access the Internet. The design goal is to integrate as much functionality as possible during a given time window. 

To address performance evaluation challenges of the next generation of computer systems, such as multicore computers inside of cell phones, I introduce in this presentation a new metric, Capacity, which evaluates the performance of CHMs that process multiple, variable heterogeneous workloads, which we refer to as demands. In contrast to single-valued metrics such as throughput, Capacity is a shape, a surface in ndimensions and a curve in 2-dimensions. We show how Capacity is a successor to throughput, through an automobile production analogy, thus motivating how multiprocessors should be viewed as plants, rather than production pipelines. For the analysis of Capacity curve shapes, we propose the development of a Demand Characterization Method (DCM) to be used in conjunction with the capacity metric to identify optimal CHM designs for specific demands. We include experimental results finding that Capacity is a better predictor of optimal designs than single-valued metrics. 


Mof Otoom is currently an Assistant Professor in the Computer Engineering Department at Yarmouk University, Jordan. Dr. Otoom earned his Ph.D. in Computer Engineering from Virginia Tech under the supervision of JoAnn M Paul, his M.Sc. in Computer Information Systems from Arizona State University, and his B.Sc. in Computer Engineering from Yarmouk University. His research interests are in the areas of performance evaluation and workload characterization of CHMs, embedded systems design, and Workload Specific Processors. Dr. Otoom is a member of IEEE, Eta Kappa Nu, Gamma Beta Phi, and Golden Key International Honour. 


CE Tweets