Font Size: a A A

Key Technologies Of The Mpls Network Dynamic Traffic Engineering

Posted on:2011-11-01Degree:MasterType:Thesis
Country:ChinaCandidate:B HanFull Text:PDF
GTID:2208360308967170Subject:Communication and Information System
Abstract/Summary:PDF Full Text Request
With the rapid development of information technology, demands for multimedia information such as graphics, audio and video increase in high speed, which put a great challenge on bandwidth allocation strategy in communication networks. To face this challenge, Multi-Protocol Label Switching (MPLS) is proposed. In MPLS networks, traffic is aggregated into Label Switching Paths (LSP), which is calculated and deployed in advance as a"tunel"from source router to destination router. These"explicit routing"approach is much more flexible and efficient than traditional hop-by-hop IP routing strategy. To achieve high utilization and fault-tolerance, usually multiple parallel LSP paths are constructed between any node pair in MPLS networks. As a result, the edge router has to determine the splitting of traffic among all these parallel LSPs. To determine the best splitting scheme which makes the network performance optimal is often treated as a subcategory of Traffic Engineering (TE) problem. In this thesis, dynamic TE approach in this context is researched in detail.Major contribution of this thesis is as follow: firstly, the famous MATE algorithm is improved through an off-line path planning phase. Secondly, a dynamic TE scheme, called MS-MATE, is proposed to impove the network performance in multi-service environment. Finally, the granularity of traffic splitting is studied, which turns out to be the most important implementation issue in dynamic TE context.In chapter 2, the MATE algorithm is studied in detail, and some possible improvement is suggested. Then, the off-line path planning phase is proposed as an enhancement to the original MATE algorithm. We considered two strategies for solving path planning problem. The first one use k-shortest path algorithm and the second one formulate the problem in an Integer Linear Programming (ILP) model. The performance of these strategies is compared through extensive simulation.In chapter 3, a dynamic TE algorithm in multi-service environment, which we called it Multi-Service MATE (MS-MATE), is proposed. The parallel paths between a specific node pair are divided into different types of path sets according to their delay, loss rate and jitter. For delay-sensitive traffic (e.g., packet voice), the paths which has path delay less than a given threshold (e.g., 50ms) is used. On the other hand, loss-sensitive traffic is carried on the paths which has end-to-end loss rate lower than a given threshold (e.g, 5%). In this framework, different objective is used for different path set. For example, for the paths carrying packet voice traffic we try to minimize their average delay, and for the paths carrying FTP traffic we try to minimize their average loss rate. In this way, we hope that not only the QoS requirement for each specific traffic type is satisfied, but also the best network performance can be achieved.In chapter 4, an important issue, namely the granularity of traffic splitting, is studied. Firstly, analysis and comparison of some typical strategies, such as packet-based, flow-based, and dynamic-hash-based algorithms, are presented. Based on the insight gained in the analysis, we propose a novel mechanism, called burst-based splitting, to improve the performance in terms of accuracy of splitting ratio and the occurence of reordering events.Finally, we conclude and mention areas of future work in chapter 5.
Keywords/Search Tags:Traffic Engineering, On-line dynamic algorithm, Path Planning, Traffic splitting mechanism
PDF Full Text Request
Related items