A novel scheme for fast flow classification in GMPLS-based networks

Abstract

In GMPLS-based networks, data is forwarded in manner of label switching on Label Switching Router (LSR). Ingress LSR must classify different IP flows into a set of Forwarding Equivalence Classes (FECs), which is a typical flow classification process. Ingress LSR will be the bottleneck of GMPLS-based networks if it could not provision fast flow classification. In this paper, we propose a novel fast flow classification scheme, coined Hierarchical Dividing Tree Scheme (HDTS), to improve the switching performance of ingress LSR in GMPLS-based networks. Four important advantages can be achieved by the proposed HDTS. First, the preprocess time in HDTS is reasonable. Second, fast FEC update is supported. Third, memory cost of HDTS is very low. Most important, the key factor that affects flow classification speed is not the number of FECs, but the depth of the search trees in HDTS. Theoretical analyses and simulations are conducted to evaluate performance of the proposed HDTS. Based on the analytical and experimental results, we can conclude that our HDTS improves the switching performance of ingress LSR greatly and is very practical for GMPLS-based networks due to its low cost.