ABSTRACT
MPLS TE (Traffic Engineering based on MPLS) and DiffServ (Differentiated Services) can be deployed concurrently in an IP backbone. DiffServ can provide packets with a preferential treatment using different code-points in their headers. MPLS networks can be configured to offer different quality of services (QoSs hereafter) to different paths through the network and enables greater control over routing in packet networks. If the two technologies are combined, then standardized DiffServ service offerings can be made. The combination of MPLS and DiffServ is called DS-TE (DiffServ aware MPLS Traffic Engineering) [2]. Efficient routing of user request is essential in a DS-TE to satisfy QoS and to manage network resources. As proved in [3], it is better to use differentiated LSPs (Label Switched Paths) for different traffic classes. What is important is to select the better LSPs specialized for each class while ensuring the efficient resource utilization. Our proposition focuses on isolation of the various flows on the better paths as long as possible inside the network yields in a better quality of service to the various classes. Our new model, PEMS, has three stages: preprocessing stage, candidate paths computing stage and demand splitting stage for LSP requests. In the preprocessing stage, it extracts good paths in order to avoid online searching overhead. This stage uses only topology information. In the online mode, when link state information is updated, new candidate paths for each class are calculated based on updated information such as measured delay and residual bandwidth. It calculates the splitting probability with different weights of delay and bandwidth depending on the class of requested traffic. When a traffic demand arrives, it performs PER (Prediction of Effective Repartition) algorithm proposed in [1] to select one LSP to carry current flow.
