DIChirp: direct injection bandwidth estimation

Ideally, network bandwidth estimation algorithms should be independent of the end system performance. If end system capabilities are involved, then the measurement will be of the system throughput and will not indicate a correct assessment of network bandwidth. Packet dispersion-based active bandwidth estimation schemes including Pathload, TOPP and pathChirp use delay correlation where the network-induced delay on packets transmitted at certain rates is translated into bandwidth estimation. Since packet dispersion-based active measurement schemes use delay correlation, bandwidth estimations are distorted by the host protocol stack-induced delay variations. Studies revealed that the host protocol stack-induced delay variations due to context switching are stovepiped in the network-induced delay variations and impact the measurement process. This study explores the delay variations introduced by the host protocol stack in packet dispersion-based techniques. The impact of host protocol delay variations and context switching on bandwidth estimation is analyzed and a new active bandwidth estimation tool minimizing the impact of context switching is proposed. Direct Injection Chirp (DIChirp) bypasses the TCP/IP protocol stack and directly interfaces with the network hardware. It uses the kernel for scheduling the outgoing packets, thus achieving more accurate estimation of bandwidth. Experiments revealed that the host protocol and context switching-induced delay variations can be as high as 800µs and could result in bandwidth estimation errors near 20%. Experiments also revealed that the DIChirp is superior to the pathChirp implementation in performance estimation since the datapath utilized by DIChirp is less prone to delay variations induced by context switching.

[1]  Van Jacobson,et al.  A tool to infer characteristics of internet paths , 1997 .

[2]  B. A. Mar,et al.  pchar : A Tool for Measuring Internet Path Characteristics , 2000 .

[3]  Richard G. Baraniuk,et al.  pathChirp: Efficient available bandwidth estimation for network paths , 2003 .

[4]  Manish Jain,et al.  Effects of Interrupt Coalescence on Network Measurements , 2004, PAM.

[5]  Fulvio Risso,et al.  Profiling and optimization of software-based network-analysis applications , 2003, Proceedings. 15th Symposium on Computer Architecture and High Performance Computing.

[6]  Manish Jain,et al.  End-to-end available bandwidth: measurement methodology, dynamics, and relation with TCP throughput , 2002, SIGCOMM 2002.

[7]  Xiao Zhang,et al.  Anatomy of UDP and M-VIA for cluster communication , 2005, J. Parallel Distributed Comput..

[8]  Mark Crovella,et al.  Measuring Bottleneck Link Speed in Packet-Switched Networks , 1996, Perform. Evaluation.

[9]  Parameswaran Ramanathan,et al.  What do packet dispersion techniques measure? , 2001, Proceedings IEEE INFOCOM 2001. Conference on Computer Communications. Twentieth Annual Joint Conference of the IEEE Computer and Communications Society (Cat. No.01CH37213).

[10]  Zhi-Li Zhang,et al.  Small-time scaling behaviors of Internet backbone traffic: an empirical study , 2003, IEEE INFOCOM 2003. Twenty-second Annual Joint Conference of the IEEE Computer and Communications Societies (IEEE Cat. No.03CH37428).