Your location now is: Home > Lectures > Content

(August 17)Practical Flow Scheduling for Commodity Datacenters
Sep 6, 2016


Practical  Flow Scheduling for Commodity Datacenters


Prof.  Kai Chen


9:30, August 17,  2016


609, New  Science and Technology Building, North Campus

Lecturer  Profile

Kai  Chen is an Assistant Professor with CSE department at HKUST. He received his  BS/MS from University of Science and Technology of China in 2004/2007 and PhD  from Northwestern University in 2012, all in Computer Science. His research  interests include networked systems design and implementation, data center  networking, data centric networking, and cloud and big data systems. His work  has been published in various top conferences and journals such as SIGCOMM,  NSDI, IEEE/ACM TON and so on. One of his papers was a SIGCOMM’10 best paper  candidate, fast-tracked to IEEE/ACM TON. More information can be found  at:

Lecture  Abstract

Our group has been  working on datacenter networks for years. In this talk, I will first overview a  lab-scale 100-server datacenter cluster we built and research works we  conducted based on it in the last 4 years. Then, I will pick one of them called  PIAS, Practical Information-Agnostic flow Scheduling, and discuss the details.

PIAS was motivated by  the observation that many prior datacenter flow scheduling schemes minimize  flow completion times (FCT) assuming prior knowledge of flows and custom switch  functions. This makes them superior in performance, but hard to use  in practice. By PIAS, we present a scheme to minimize FCT with no prior  knowledge and existing commodity switch functions. At its heart, PIAS leverages  multiple priority queues available in existing commodity switches to implement  a Multiple Level Feedback Queue (MLFQ), in which a PIAS flow is gradually  demoted from higher-priority queues to lower-priority queues based on the  amount of bytes it has sent. Consequently, short flows are likely to be  completed in the first few high-priority queues and thus be prioritized over  long flows in general, which enables PIAS to emulate SJF without knowing flow  sizes in advance. PIAS is readily implementable with commodity switches and  backward compatible with legacy TCP/IP stacks. We have shown its performance  through both testbed experiments and ns2 simulations. To make our work easy to  reproduce, we made the implementation code online at:

Previous:(August 29)High-Repetition-Rate Sub-Cycle Waveform Synthesis
Next:(August 2)Bringing Cores Closer Together: The Wireless Revolution in On-Chip Communication

South Campus
Add: 266 Xinglong Section of Xifeng Road, Xi’an, Shaanxi 710126
Tel: 86-29-81891818
North Campus
Add: No. 2 South Taibai Road, Xi’an, Shaanxi 710071
Tel: 86-29-88202212