"Speeding Up Large-Scale Distributed Computation Using Codes and Scheduling"

Ramtin Pedarsani, Faculty Candidate, UC Berkeley

March 16th (Wednesday), 10:00am
Harold Frank Hall (HFH), Rm 4164 (ECE Conf. Rm.)

Modern large-scale computing platforms face several types of randomness, uncertainty and
system noise. These include stragglers, system failures, maintenance outages, and communication bottlenecks. In this talk, I will introduce two techniques that can speed up large-scale distributed computation, and guarantee high throughput of the processing system in the presence of unpredictability.In the first part of the talk, I visit distributed computing and machine learning algorithms through a codingtheoretic lens, and show how codes can equip them with robustness against system uncertainty. I show that codes can alleviate the effect of stragglers in distributed computing, and significantly reduce communication cost of data shuffling that is a main building block of parallel machine learning algorithms. In the second part of the talk, I focus on efficient resource allocation and job scheduling in data centers. I develop a novel stochastic model for scheduling jobs that are represented as directed acyclic graphs (DAG) on multiple heterogeneous servers. I will then design a throughput-optimal policy for DAG scheduling that is robust to unpredictability of demand and service in the system. I will conclude the talk with an overview of my research agenda and future research directions.

About Ramtin Pedarsani:

photo of Ramtin PedarsaniRamtin Pedarsani is a postdoctoral researcher in the Electrical Engineering and Computer Sciences Department at UC Berkeley. He obtained his Ph.D. in Electrical Engineering and Computer Sciences from UC Berkeley in 2015. He received his M.Sc. degree at EPFL in 2011 and his B.Sc. degree at the University of Tehran in 2009. His research interests include networking, signal processing, information and coding theory, and transportation systems. He is the recipient of the best paper award in the IEEE International Conference on Communications (ICC) in 2014.

Hosted by: Professor B.S. Manjunath