Authors:
Tomokazu Moriyama
;
Ryo Yamamoto
;
Satoshi Ohzahata
and
Toshihiko Kato
Affiliation:
Graduate School of Informatics and Engineering, University of Electro-Communications, 1-5-1, Chofugaoka, Chofu, Tokyo 182-8585 and Japan
Keyword(s):
TCP, Congestion Control, Packet Loss Classification, IEEE 802.11 WLAN, K-Means Clustering.
Related
Ontology
Subjects/Areas/Topics:
Data Communication Networking
;
Internet Technologies
;
Network Architectures
;
Network Monitoring and Control
;
Network Protocols
;
Telecommunications
Abstract:
Recent IEEE 802.11 wireless LANs provide high speed data transfer using the newly introduced physical and
MAC technologies. Although packet losses over a wireless link are also decreased by the help of new MAC
technologies, some packet losses still occur randomly. Those packet losses invoke TCP congestion control,
which reduces the TCP level throughput, even if congestion does not occur at al. In order to resolve this
problem, some machine learning based approaches have been proposed, which use K-means clustering in
order to discriminate congestion triggered packet losses and wireless error triggered packet losses. However,
those proposals use only delay related parameters, but delay may increase due to non-congestion reasons, in
which case the conventional proposals fail discrimination. This paper proposes a method to classify packet
losses by the K-means clustering focusing on congestion window size and round-trip delay, and to stop
decreasing congestion window when losses
are triggered by wireless errors. We develop the proposed method
as a Linux kernel module and show the performance evaluation results that the throughput increases by 40%
without increasing unnecessary packet losses.
(More)