Performance Evaluation of Fast Startup Congestion Control Schemes

The Transmission Control Protocol (TCP) uses the Slow-Start mechanism at the beginning of a connection and after idle times. The Slow-Start delays the transport of data in particular if the round-trip time is large, which is undesirable for interactive applications. In order to speed up transfers, several alternatives have been proposed recently. This paper evaluates the performance and robustness of new fast startup congestion control schemes. We compare both end-to-end approaches as well as protocols that rely on additional feedback from the routers, using implementations in the Linux stack. Both testbed measurements and simulation studies quantify the potential performance improvement, the risk of packet loss, and the benefits of additional router support. Our results, which are also partly verified analytically, reveal that end-to-end fast startup mechanisms would not cause too much performance degradation if they are selectively used and carefully tuned. Additional router support would improve the fairness at the cost of a higher complexity.

[1]  Peter Steenkiste,et al.  Improving TCP startup performance using active measurements: algorithm and evaluation , 2003, 11th IEEE International Conference on Network Protocols, 2003. Proceedings..

[2]  Michele C. Weigle,et al.  Tmix: a tool for generating realistic TCP application workloads in ns-2 , 2006, CCRV.

[3]  Nick McKeown,et al.  Rate control protocol (rcp): congestion control to make flows complete quickly , 2008 .

[4]  Injong Rhee,et al.  Towards a common TCP evaluation suite , 2008 .

[5]  Dimitri Papadimitriou,et al.  Open Research Issues in Internet Congestion Control , 2011, RFC.

[6]  Michael Scharf,et al.  Performance Evaluation of Quick-Start TCP with a Linux Kernel Implementation , 2008, Networking.

[7]  Mark Handley,et al.  TCP Congestion Window Validation , 2000, RFC.

[8]  Van Jacobson,et al.  Congestion avoidance and control , 1988, SIGCOMM '88.

[9]  Limin Wang,et al.  Congestion Control Without a Startup Phase , 2010 .

[10]  Pasi Sarolahti,et al.  Congestion Control in Linux TCP , 2002, USENIX Annual Technical Conference, FREENIX Track.

[11]  Sally Floyd Limited Slow-Start for TCP with Large Congestion Windows , 2004, RFC.

[12]  Anthony McGregor,et al.  Simulation with real world network stacks , 2005, Proceedings of the Winter Simulation Conference, 2005..

[13]  Nick McKeown,et al.  RCP-AC: Congestion Control to Make Flows Complete Quickly in Any Environment , 2006, Proceedings IEEE INFOCOM 2006. 25TH IEEE International Conference on Computer Communications.

[14]  Injong Rhee,et al.  CUBIC: a new TCP-friendly high-speed TCP variant , 2008, OPSR.

[15]  Michael Scharf,et al.  Performance analysis of the Quick-Start TCP extension , 2007, 2007 Fourth International Conference on Broadband Communications, Networks and Systems (BROADNETS '07).

[16]  Dimitri Papadimitriou,et al.  Open Research Issues in Internet Congestion Control draft-irtf-icc rg-welzl-congestion-control-open-research-01.txt , 2007 .

[17]  Amitabha Das NETWORKING 2008, Ad Hoc and Sensor Networks, Wireless Networks, Next Generation Internet , 7th International IFIP-TC6 Networking Conference, Singapore, May 5-9, 2008, Proceedings , 2008, Networking.

[18]  Sally Floyd,et al.  Increasing TCP's Initial Window , 1998, RFC.

[19]  Sally Floyd,et al.  Determining an appropriate sending rate over an underutilized network path , 2007, Comput. Networks.

[20]  S. Ha,et al.  Hybrid Slow Start for High-Bandwidth and Long-Distance Networks , 2008 .

[21]  Sally Floyd,et al.  Quick-Start for TCP and IP , 2007, RFC.