The RSVP-TE protocol has been widely applied to the lightpath establishment in ASON. Thus, the Optical Cross Connect (OXC) and wavelength lightpath are regarded as the Label Switched Router (LSR) and Label Switched Path (LSP), respectively. Shorter signaling processing time provides better services of real-time Resource on Demand (RoD) and rapid failure restoration, so a novel software solution, based on a Planar-Chain Approach (PCA) instead of the traditional Linear-Chain Approach (LCA), is proposed to reduce the number of PATH State Blocks (PSBs) and RESV State Blocks (RSBs) maintained at each LSR. Hence, the RSVP-TE processing time complexity is reduced from O(N^2) to O(N), where N is the number of LSRs. This work has been implemented by Linux-OS in the RSVP-TE testbed of Bell Labs and Tsinghua University.For the soft state refresh, the information of PATH/RESV message is maintained in PSB/RSB with the matching triple of SESSION object, SENDER_TEMPLATE/FILTER_SPEC object and InIf (Incoming Interface)/OutIf (Outgoing Interface). A given LSP exists in the state blocks corresponded with a pair of PSB and RSB with the same SESSION. We improved the LCA to obtain the PCA, in which two separate linear chains for PSB and RSB are integrated into one planar chain with a transverse linear chain based on SESSION and two longitudinal linear chains based on PSB and RSB, as seen in the figure. The head of the SESSION chain is served as the planar chain head. Each unit in the SESSION chain records a SESSION and the corresponding PSB and RSB chain heads. The PCA with the planar structure is suitable to the rapid searching operations so as to shorten the signaling processing time at each LSR. The PSB/RSB access is a procedure of searching the planar chain according to the matching triple. After searching the SESSION chain for a matched SESSION, the PSB/RSB chain is hunted for the matched SENDER_TEMPLATE/ FILTER_SPEC and InIf/OutIf in order to obtain the matched PSB/RSB.We made the simulations on NSFNET and two star topologies (with 30 or 50 LSRs) to obtain the signaling processing time along one LSP in NSFNET and in the central LSR of star with different traffic load. Numerical results demonstrate that:1) - As a whole, the time in the PCA is shorter than that in the LCA and the advantage is more visible for heavier load, because more PSBs and RSBs are maintained at LSR with the increasing traffic load. When the load is 200 Erlang, the time in the PCA is decreased by one order of magnitude in the star topologies and by 49% for only PATH and 62% for both PATH and RESV in NSFNET.2) - In the star topologies, the time in the LCA is only dependent of the number of traffic requests, because all the traffic connections traverse the central LSR and are kept linearly; the time in the PCA almost fully depends on the number of LSRs, because the length of searching SESSION chain and PSB/RSB chain corresponds to the number of source and destination LSRs.
© 2006 Optical Society of America