大学サイト   学部サイト   学科サイト   EnglishPages
 
旧尾家・塚本研究室です.尾家祐二 現・九州工業大学学長の情報は こちら.
   
研究業績


Studies on Performance Analyses of Resource Management in Multiprotocol Label Switching Network
著者
  Shogo Nakazawa
 
雑誌名/会議名
  九州工業大学大学院 情報工学研究科 情報システム専攻, 博士学位論文 2002 年 1 月
 
アブストラクト
  For many years, the Internet has worked quite well without providing Quality of Service (QoS). Today, the Internet mainly offers best-effort service. A best-effort service makes no guaran- tees regarding when or whether a packet is delivered to the receiver since packets are usually dropped at some IP routers due to network congestion. However, as the Internet continues to grow, applications other than traditional data transfer, such as Voice over IP (VoIP) and video- conferencing, are envisioned. To support transmission of multimedia such as voice, video, and data application traffic with varying service requirements from the network, all major Internet service providers (ISPs) and vendors are providing products that have the QoS features. Multiprotocol Label Switching (MPLS) is an Internet Engineering Task Force (IETF) stan- dard for a new switching paradigm that enables packet switching at layer 2 while using layer 3 forwarding information. To improve the performance of IP router which consults with the routing tables for every IP packet, MPLS forwards packets based on a short fixed length label without address resolution on the layer 3. This mechanism is called cut-through and it allows datagrams to be quickly transmitted. The major objective of this dissertation is to analyze some performances of MPLS networks in terms of resource management. In order to realize the data transmission with its QoS require- ment such as latency and packet loss, we should estimate required network resources based on the performance measures. In particular, I will focus on the number of labels required, the bandwidth allocation among the label switched paths (LSPs), and the buffer size of the label switching router (LSR). First, we investigate the performance comparison between two label mapping policies of LSR; one is the data-driven label mapping policy and the other is the control-driven one. In gigabit networks, the number of IP flows through a router would be a few millions, so that LSPs set on layer 2 become one of most important resources. Considering the scalability of the label switching schemes is also important, since they are to become extensively used in the Internet to provide cost effective high speed IP packet forwarding. In Chapter 3, we focus on the number of labels required for the LSR with different label mapping methods. In order to evaluate the LSR, we derive three performance measures, the cut-through rate, the average IP datagram processing delay, and the mis-order rate. The cut-through rate is the ratio of the average number of datagrams transmitted by cut-through to that of datagrams generated from whole sources. The mis-order rate indicates the ratio of out-of-order transmission of datagrams caused by cut-through. By assuming LSR with these label mapping policies as queueing model, we analyze these performance measures and evaluate the number of labels required under these label mapping policies. In addition, we compare their performances in both cases of LSR. From numerical results, the data-driven label mapping method can reduce the number of consumed labels to achieve the same cut-through rate in comparison with the control-driven one. Furthermore, we observe that the datagrams in the data-driven model are transmitted about ten times faster than in the control-driven one. Thus, the data-driven model will enable us to utilize network resources efficiently. Second, in order to investigate bandwidth allocation for real time traffic, we evaluate the impact of the bandwidth allocation for cut-through transmission on its delay performance. In Chapter 4, we assume the label mapping method as the data-driven scheme, in which IP flows up to the maximum number of LSPs are transmitted by cut-through at layer 2, but exceeded flows are lifted up the routing kernel of layer 3. Therefore, the bandwidth allocation of LSR between the hop-by-hop LSP and cut-through LSPs will become important in improving delay performance. Through numerical results, we obtain that the cut-through rate is increasing as the maximum number of LSPs, it may approach w, where w is the bandwidth ratio for cut-through transmission in layer 2 and is the average total arrival rate of IP datagrams. Moreover, we find that the LSR has the optimal number of cut-through LSPs which minimizes the average processing delay in LSR. Third, we investigate the guide for QoS provisioning in MPLS networks. In actual networks, the LSR accommodates both mission-critical and non-mission-critical traffic. In Chapter 5, we consider the case that both real time and non-real time traffic come into LSR under the assumption that the cut-through transmission is applied only to real time traffic. Thus, in the situation, we analyze the finite buffer model to evaluate the packet transmission delay and the packet loss rate in layer 3 routing kernel. Since mission-critical traffic has stringent delay constraints, LSR transmits it fast by cut-through forwarding while non-mission-critical traffic is transfered via layer 3 with legacy forwarding method. Therefore, it becomes important to allocate transmission bandwidth among some cut-through LSPs and hop-by-hop LSP in delay constraints. On the other hand, packet loss rate should be kept relatively low for non-mission- critical traffic, so that the buffer dimensioning in layer 3 routing kernel is essential. Therefore, by investigating the impact of both the bandwidth allocation and the buffer dimensioning to meet the requirement with respect to their processing delay and packet loss rate, we represented the guideline for QoS provisioning in MPLS networks. Finally, I would like to hope that this dissertation will be helpful for further study in this field.
 
キーワード
  MPLS
 
記述言語
  English
 
 

web_admin [ at ] infonet.cse.kyutech.ac.jp /  Copyright (c) 2016 Oie & Tsukamoto Laboratory, All Right Reserved.