Final Report - Dynamic Path Scheduling through Extensions to Generalized Multiprotocol Label Switching (GMPLS)

[摘要] The major accomplishments of the project are the successful software implementation of the Phase I scheduling algorithms for GMPLS Label Switched Paths (LSPs) and the extension of the IETF Path Computation Element (PCE) Protocol to support scheduling extensions. In performing this work, we have demonstrated the theoretical work of Phase I, analyzed key issues, and made relevant extensions. Regarding the software implementation, we developed a proof of concept prototype as part of our Algorithm Evaluation System (AES). This implementation uses the Linux operating system to provide software portability and will be the foundation for our commercial software. To demonstrate proof of concept, we have implemented LSP scheduling algorithms to support two of the key GMPLS switching technologies (Lambda and Packet) and support both Fixed Path (FP) and Switched Path (SP) routing. We chose Lambda and Packet because we felt it was essential to include both circuit and packet switching technologies as well as to address all-optical switching in the study. As conceptualized in Phase I, the FP algorithms use a traditional approach where the LSP uses the same physical path for the entire service duration while the innovative SP algorithms allow the physical path to vary during the service duration. As part of this study, we have used the AES to conduct a performance analysis using metro size networks (up to 32 nodes) that showed that these algorithms are suitable for commercial implementation. Our results showed that the CPU time required to compute an LSP schedule was small compared to expected inter-arrival time between LSP requests. Also, when the network size increased from 7 to 15 to 32 nodes with 10, 26, and 56 TE links, the CPU processing time showed excellent scaling properties. When Fixed Path and Switched Path routing were compared, SP provided only modestly better performance with respect to LSP completion rate, service duration, path length, and start time deviation. In addition, the SP routing required somewhat more CPU time than the FP routing. However, when the allowable range for the start time is increased, the CPU time for SP increased less rapidly than FP such that the CPU processing times became comparable. Therefore, SP routing may scale better than FP routing. The Path Computation Element Working Group is a complementary working group to the GMPLS working group (CCAMP) in the IETF and is developing standards to discover, manage, and access elements that compute routes for GMPLS for LSPs. The algorithms used to compute the routes are not subject to standardization. Since the PCE supports only standard GMPLS LSPs, it does not support scheduled LSPs. Therefore, it is a natural extension of our Phase I work to introduce this enhancement into the PCE. In addition, the PCE adds another dimension to our product line because the PCE by itself may be a product with a standard interface.

[发布日期] 2009-05-22

[学科分类] 数学（综合）