Task-aware Scheduling for Data Center Networks

PhD Qualifying Examination


Title: "Task-aware Scheduling for Data Center Networks"

by

Miss Jingjie JIANG


Abstract:

As the most prevalent computing platform for data parallel applications, data 
centers have received extensive attentions recently. Leveraging the scale of 
economy, data centers are much more cost-efficient than traditional enterprise 
networks. The ambulant redundancy of both computation servers and network 
devices guarantee the reliability of data centers. The symmetric network 
typologies in most data centers further provide many candidate paths to ensure 
network connectivity. In addition, data centers usually adopt uniform 
management strategies across the network, which can elastically control the 
network and improve the security of stored data.

Tasks in a data center, such as web search queries or MapReduce jobs, usually 
consist of plenty of flows transferring data among different virtual machines 
across the network. The completion time of these flows significantly influence 
the user perceived performance and network efficiency. In this paper, we 
investigate the state-of-art flow scheduling schemes in data center networks. 
Before the introduction of task-awareness, both academic research and industry 
standard have focused on reducing the completion time of each individual flow, 
aiming to minimize the average completion time of all the flows in a network. 
This is equivalent to maximizing the overall throughput of a network. 
Nevertheless, different flows within the same task are correlated with one 
another, in the sense that a task is not considered to be completed till the 
completion of all its constituent flows. Therefore, the task completion time is 
far more relevant than the completion time of any individual flow within. We 
need to coordinate and schedule the flows with the task semantics into 
consideration.

Task-aware scheduling is a novel concept, and there exists many challenges and 
open problems to solve. In this paper, we comprehensively investigate existing 
literature regarding flow scheduling in data center networks. By comparing the 
scheduling strategies with and without task-awareness, we reiterate the 
importance of task-aware scheduling. We further examine the feasibility and 
limitations of several tentative task-aware scheduling schemes. Based on the 
in-depth analysis, we discuss some promising techniques that can be integrated 
to further improve the task performance in data center networks.

Date:			Tuesday, 10 June 2014

Time:                  	4:30pm - 6:30pm

Venue:                  Room 3501
                        Lifts 25/26

Committee Members:	Prof. Bo Li (Supervisor)
  			Dr. Lei Chen (Chairperson)
  			Dr. Kai Chen
  			Dr. Lin Gu


**** ALL are Welcome ****