Is minimizing flow completion time the optimal way in meeting flow's deadline in datacenter networks

In modern datacenters, the most common method to solve the network latency problem is to minimize flow completion time during the transmission process. Following the soft real-time nature, the optimization of transport latency is relaxed to meet a flow's deadline in deadline-sensitive services. However, none of existing deadline-sensitive protocols consider deadline as a constraint condition of transmission. They can only simplify the objective of meeting a flow's deadline as a deadline-aware mechanism by assigning a higher priority for tight-deadline constrained flows to finish the transmission as soon as possible, which results in an unsatisfactory effect in the condition of high fan-in degree. It drives us to take a step back and rethink whether minimizing flow completion time is the optimal way in meeting flow's deadline. In this paper, we focus on the design of a soft real-time transport protocol with deadline constraint in datacenters and present a flow-based deadline scheduling scheme for datacenter networks (FBDS). FBDS makes the unilateral deadline-aware flow transmission with priority transform into a compound centralized single-machine deadline-based flow scheduling decision. In addition, FBDS blocks the flow sets and postpones some flows with extra time until their deadlines to make room for the new arriving flows in order to improve the deadline meeting rate. Our simulation results on flow completion time and deadline meeting rate reveal the potential of FBDS in terms of a considerable deadline-sensitive transport protocol for deadline-sensitive interactive services.

[1]  Bo Li,et al.  On meeting deadlines in datacenter networks , 2013, Tsinghua Science and Technology.

[2]  Thierry Turletti,et al.  A Survey of Software-Defined Networking: Past, Present, and Future of Programmable Networks , 2014, IEEE Communications Surveys & Tutorials.

[3]  Nick McKeown,et al.  Why flow-completion time is the right metric for congestion control , 2006, CCRV.

[4]  Shu Yang,et al.  Delay-differentiated scheduling in optical packet switches for cloud data centers , 2015 .

[5]  Chunming Qiao,et al.  CoLoR: an information-centric internet architecture for innovations , 2014, IEEE Network.

[6]  Bi Jun,et al.  Allocation and scheduling of network resource for multiple control applications in SDN , 2015, China Communications.

[7]  Alex X. Liu,et al.  Friends, not Foes – Synthesizing Existing Transport Strategies for Data Center Networks , 2014 .

[8]  T. N. Vijaykumar,et al.  Deadline-aware datacenter tcp (D2TCP) , 2012, CCRV.

[9]  Nick McKeown,et al.  Deconstructing datacenter packet transport , 2012, HotNets-XI.

[10]  Brighten Godfrey,et al.  Finishing flows quickly with preemptive scheduling , 2012, CCRV.

[11]  D. Kahneman,et al.  Experimental Tests of the Endowment Effect and the Coase Theorem , 1990, Journal of Political Economy.

[12]  Randy H. Katz,et al.  DeTail: reducing the flow completion time tail in datacenter networks , 2012, SIGCOMM '12.

[13]  Jianping Wu,et al.  TAPS: Task-aware preemptive flow scheduling , 2014, 2014 IEEE 20th International Workshop on Local & Metropolitan Area Networks (LANMAN).

[14]  Hongbin Luo,et al.  An Improved Deadline-aware Scheme with Low Overhead in Datacenter Networks , 2015 .

[15]  Sheng Wang,et al.  Rapier: Integrating routing and scheduling for coflow-aware data center networks , 2015, 2015 IEEE Conference on Computer Communications (INFOCOM).

[16]  Zhang Hong,et al.  Resource intensity aware job scheduling in a distributed cloud , 2014, China Communications.