Baochun Li
University of Toronto, Kanada
21 May 2015, 04:15 pm
S2|02 Raum C110, Robert-Piloty-Gebäude, Hochschulstr. 10, 64289 Darmstadt
“Scheduling with Multiple Resource Types in Cloud Datacenters”
Abstract:
In the age of big data, it has been the norm for cloud datacenters to run data analytic applications at a large scale. Yet, as multiple applications share resources in these datacenters, it is important to design scheduling disciplines for datacenter resources to be shared in a fair and efficient manner. In this talk, I will present a new class of scheduling disciplines that are specifically designed for sharing multiple resource types in cloud datacenters.
I will first discuss how multiple resource types are to be shared in space, and present a new design that allocates resources to applications by scheduling their computing tasks onto datacenter nodes. I will then focus on the problem of sharing resources over time, and present a new scheduler to multiplex application flows in software routers, sharing both CPU and link bandwidth. In both contexts, I will open with examples, explore their theoretical underpinnings related to fairness and efficiency, and! conclude with challenges and observations in real-world implementations.
Short Bio:
Baochun received the B.Engr. degree from Tsinghua Univ. (CN), and the M.Sc. and Ph.D. degrees from UIUC (US). Since 2000, he is with the ECE Dept. at the Univ. of Toronto (CA), where he currently holds the Bell Canada Endowed Chair in Computer Engineering. Amongst others, he is recipient of the IEEE ComSoc Leonard G. Abraham Award, the IEEE ComSoc Multimedia Communications Best Paper Award, and the
University of Toronto McLean Award.
