Share Email Print

Proceedings Paper

Real-time optical path control method that utilizes multiple support vector machines for traffic prediction
Format Member Price Non-Member Price
PDF $14.40 $18.00

Paper Abstract

An effective solution to the continuous Internet traffic expansion is to offload traffic to lower layers such as the L2 or L1 optical layers. One possible approach is to introduce dynamic optical path operations such as adaptive establishment/tear down according to traffic variation. Path operations cannot be done instantaneously; hence, traffic prediction is essential. Conventional prediction techniques need optimal parameter values to be determined in advance by averaging long-term variations from the past. However, this does not allow adaptation to the ever-changing short-term variations expected to be common in future networks. In this paper, we propose a real-time optical path control method based on a machinelearning technique involving support vector machines (SVMs). A SVM learns the most recent traffic characteristics, and so enables better adaptation to temporal traffic variations than conventional techniques. The difficulty lies in determining how to minimize the time gap between optical path operation and buffer management at the originating points of those paths. The gap makes the required learning data set enormous and the learning process costly. To resolve the problem, we propose the adoption of multiple SVMs running in parallel, trained with non-overlapping subsets of the original data set. The maximum value of the outputs of these SVMs will be the estimated number of necessary paths. Numerical experiments prove that our proposed method outperforms a conventional prediction method, the autoregressive moving average method with optimal parameter values determined by Akaike’s information criterion, and reduces the packet-loss ratio by up to 98%.

Paper Details

Date Published: 13 February 2016
PDF: 6 pages
Proc. SPIE 9773, Optical Metro Networks and Short-Haul Systems VIII, 97730I (13 February 2016); doi: 10.1117/12.2209463
Show Author Affiliations
Hiroshi Kawase, Nagoya Univ. (Japan)
Yojiro Mori, Nagoya Univ. (Japan)
Hiroshi Hasegawa, Nagoya Univ. (Japan)
Ken-ichi Sato, Nagoya Univ. (Japan)

Published in SPIE Proceedings Vol. 9773:
Optical Metro Networks and Short-Haul Systems VIII
Atul K. Srivastava; Werner Weiershausen; Benjamin B. Dingel; Achyut K. Dutta, Editor(s)

© SPIE. Terms of Use
Back to Top