Accuracy of link capacity estimates using passive and active approaches with CapProbe

  • Authors:
  • R. Kapoor;Ling-Jyh Chen;M. Y. Sanadidi;M. Gerla

  • Affiliations:
  • Dept. of Comput. Sci., UCLA, Los Angeles, CA, USA;Dept. of Comput. Sci., UCLA, Los Angeles, CA, USA;Dept. of Comput. Sci., UCLA, Los Angeles, CA, USA;Dept. of Comput. Sci., UCLA, Los Angeles, CA, USA

  • Venue:
  • ISCC '04 Proceedings of the Ninth International Symposium on Computers and Communications 2004 Volume 2 (ISCC"04) - Volume 02
  • Year:
  • 2004

Quantified Score

Hi-index 0.00

Visualization

Abstract

CapProbe is an inexpensive and accurate means to estimate capacity. CapProbe combines both dispersion and end-to-end delay to estimate the capacity of the narrowest link on a path. We evaluate in this paper the accuracy of CapProbe estimation, and its dependence on end systems speed, packet sizes, narrow link speeds, and other system parameters. We test kernel and user level implementations of CapProbe and find the kernel implementation to be much more accurate. We also evaluate through experiments the effect of probing packet size on the accuracy of CapProbe estimation. Finally, we explore the idea of a "passive CapProbe" within the context of a TCP flow. Passive here means that the dispersion and delay observed for the TCP flow data and ACK packets, without introducing any additional probing packets. We test active and passive versions of CapProbe with TCP. The active version is found to produce more accurate capacity estimates than the passive version.