CA-TCP: Overtime cost aware data center TCP

被引:0
作者
Deng, Gang [1 ]
Wang, Hong [1 ]
Gong, Zhenghu [1 ]
机构
[1] College of Computer, NUDT, Changsha
关键词
Cloud Computing; Data Center Networks; Deadline; ECN; Overtime Cost; TCP;
D O I
10.1166/jctn.2015.4080
中图分类号
学科分类号
摘要
Latency-sensitive applications in data center usually confront a soft-real-time constraint, which means a soft deadline for the applications. Missing this deadline may cause performance impairment. For this reason, directly using the traditional TCP in data center networks, which is deadline agnostic, may suffers from efficiency problem. Recently, many work improves on TCP to address this problem, e.g., meeting the deadline, minimizing the average flow completion time (FCT), but none of them take the overtime cost into account. However, for different applications, a same overtime may mean different performance influences, namely overtime costs. And for the burst characteristic of flow in data center, overtime is inevitable. Therefore, just try to minimize the FCT or meet their deadlines cannot necessarily minimize overtime cost. In this paper, we present CA-TCP, an overtime cost aware TCP for data center networks. Instead of just trying to meet the deadline or minimize the FCT, CA-TCP employs a novel congestion avoidance algorithm, which targets a reduction in flow's overtime cost by modulating the congestion window via both the extent of congestion and overtime cost. Small-scale implementation and extensive simulation results show that CA-TCP can effectually reduce the overtime cost comparing with TCP, DCTCP and D2TCP under different overtime cost functions. Copyright © 2015 American Scientific Publishers.
引用
收藏
页码:3050 / 3057
页数:7
相关论文
共 20 条
[1]  
Alizadeh M., Greenberg A., Maltz D., Padhye J., Patel P., Prabhakar B., Sengupta S., Sridharan M., Data center TCP (dcTCP), Proc. SIGCOMM (2010), pp. 63-74, (2010)
[2]  
Hoff T., Latency is Everywhere and It Costs You Sales-How to Crush It. December, (2013)
[3]  
Wilson C., Ballani H., Karagiannis T., Rowstron A., Better never than late: Meeting deadlines in datacenter networks, Proc. SIGCOMM (2011), pp. 50-61, (2011)
[4]  
Balajee V., Jahangir H., Vijaykumar T.N., Deadline-aware datacenter TCP (D2TCP), Proc. SIGCOMM (2012), pp. 115-126, (2012)
[5]  
Chi-Yao H., Matthew C., Godfrey P.B., Finishing flows quickly with preemptive scheduling, Proc. SIGCOMM (2012), pp. 127-138, (2012)
[6]  
Munir A., Qazi I.A., Uzmi Z.A., Mushtaq A., Ismail S.N., Iqbal M.S., Khan B., Minimizing flow completion times in data centers, Proc. INFOCOM (2013), pp. 2157-2165, (2013)
[7]  
David Z., Tathagata D., Prashanth M., Dhruba B., Katz Y., DeTail: Reducing the flow completion time tail in datacenter networks, Proc. SIGCOMM (2012), pp. 139-150, (2012)
[8]  
Mohammad A., Shuang Y., Milad S., PFabric: Minimal nearoptimal datacenter transport, Proc. SIGCOMM (2013), pp. 435-446, (2013)
[9]  
Kandula S., Sengupta S., Greenberg A., Patel P., Chaiken R., The nature of data center traffic: Measurements and analysis, Proc. 9th ACM SIGCOMM Conference on Internet Measurement (2009), pp. 202-208, (2009)
[10]  
Shieh A., Kandula S., Greenberg A., Kim C., Saha B., Sharing the data center network, Proc. Usenix NSDI (2011), pp. 309-322, (2011)