McMaster ECE Distinguished Seminar: Optimizing Job Performance Within and Across Datacentres

Share

Superchapter: Joint Chapter of Communications, Information Theory, and Signal Processing Societies

 


In the era of big data, resources in the cloud need to be carefully scheduled to optimize the performance of data analytic jobs, so that large volumes of data can be processed efficiently. It is typical that solving optimization problems may not be practical enough for making online scheduling decisions, and other alternatives (such as stable matching) may need to be considered. Contrary to such an intuition, we show that optimization can still be used as a practical tool for scheduling resources within and across datacenters in the cloud. Within a datacenter, we designed and implemented a new utility optimal scheduler to allocate resources across competing coflows with different degrees of sensitivity, while still maintaining max-min fairness. Across geographically distributed datacenters, we designed new task schedulers to improve job performance, again by formulating and solving optimization problems. In this talk, we show how these optimization problems can be efficiently solved, and make the case for their practicality with our experiences in real-world implementations.



  Date and Time

  Location

  Contact

  Registration


  • 1280 Main Street West
  • McMaster University
  • Hamilton, Ontario
  • Canada L8S 4K1
  • Building: ETB
  • Room Number: 535


  Speakers

Prof. Baochun Li

Prof. Baochun Li of University of Toronto

Topic:

Optimizing Job Performance within and across Datacenters

In the era of big data, resources in the cloud need to be carefully scheduled to optimize the performance of data analytic jobs, so that large volumes of data can be processed efficiently. It is typical that solving optimization problems may not be practical enough for making online scheduling decisions, and other alternatives (such as stable matching) may need to be considered. Contrary to such an intuition, we show that optimization can still be used as a practical tool for scheduling resources within and across datacenters in the cloud. Within a datacenter, we designed and implemented a new utility optimal scheduler to allocate resources across competing coflows with different degrees of sensitivity, while still maintaining max-min fairness. Across geographically distributed datacenters, we designed new task schedulers to improve job performance, again by formulating and solving optimization problems. In this talk, we show how these optimization problems can be efficiently solved, and make the case for their practicality with our experiences in real-world implementations.

Biography:

Baochun Li is a Professor in the Department of Electrical and Computer Engineering at the University of Toronto. He is cross-appointed to the Department of Computer Science. He is a member of the Computer Engineering group. Starting August 2005, he holds the Bell Canada Endowed Chair in Computer Engineering. He leads the iQua research group.

Baochun received his B.Engr. degree in 1995 from Tsinghua University, China, and his M.S. and Ph.D. degrees in 1997 and 2000 from the Department of Computer Science, University of Illinois at Urbana-Champaign.

Baochun’s research interests are in the area of large-scale distributed systems, cloud computing, datacenter networks, applications of network coding, mobile computing, and wireless networks. He enjoys the process of bridging the gap between theory and practice, and of bringing theoretical results to practical implementations. In the past decade, he has worked on applications of control theory, game theory, microeconomics, optimization theory, as well as network coding to address practical challenges. He also loves writing actual code, building real systems from scratch, and making them work efficiently.