With various applications running, data centers are widely deployed in industry as the essential infrastructure around the world. These applications are interactive in nature with very demanding soft-real-time latency. Data centers execute a cloud application(task) distributedly, and a task is divided to many flows to shuffle a large amount of data. Therefore, low latency is provided to transmit the flows within deadlines in the data center network. Unfortunately, traditional transport protocols in data centers adopt fair sharing to share the link bandwidth equally. Flows cannot be completed within deadlines with these deadline-agnostic protocols, which leads to a waste of link bandwidth. Though some deadline-aware transport protocols are proposed recently to make more flows be completed before deadlines. However, cloud tasks, including financial service, online payment, scientific computation, are useful only if all the flows in the task can be completed before deadlines. Otherwise, the bandwidths consumed by the partially completed flows are wasted because of the failed task.
[1]
Christo Wilson,et al.
Better never than late
,
2011,
SIGCOMM 2011.
[2]
Amin Vahdat,et al.
A scalable, commodity data center network architecture
,
2008,
SIGCOMM '08.
[3]
Brighten Godfrey,et al.
Finishing flows quickly with preemptive scheduling
,
2012,
CCRV.
[4]
Mor Harchol-Balter,et al.
Analysis of SRPT scheduling: investigating unfairness
,
2001,
SIGMETRICS '01.
[5]
Nick McKeown,et al.
Why flow-completion time is the right metric for congestion control
,
2006,
CCRV.
[6]
Vijay Sivaraman,et al.
End-to-end statistical delay service under GPS and EDF scheduling: a comparison study
,
2001,
Proceedings IEEE INFOCOM 2001. Conference on Computer Communications. Twentieth Annual Joint Conference of the IEEE Computer and Communications Society (Cat. No.01CH37213).