IP網路與分波多工網路之最佳化選徑與資源配置演算法

(1)

國立交通大學電資學院

資訊科學與工程研究所

博

士

論

文

IP 網路與分波多工網路之最佳化選徑

網路與分波多工網路之最佳化選徑

與資源配置演算法

Optimization-based Approaches for Routing and

Resource Provisioning in IP and Optical WDM etworks

研究生：陳春秀

指導教授：楊啟瑞博士

中

華

民

國

九

十

九

年

七

月

(2)

IP 網路與分波多工網路之最佳化選徑

網路與分波多工網路之最佳化選徑

與資源配置演算法

Optimization based Approaches for Routing and

Resource Provisioning in

IP and Optical WDM etworks

研究生：陳春秀

Student：Alice C. S. Chen

指導教授：楊啟瑞博士

Advisor：Dr. Maria C. Yuang

國立交通大學資訊學院

資訊科學與工程研究所

博士論文

A Dissertation

Submitted to Institutes of Computer Science and Engineering Department of Computer Science

College of Computer Science National Chiao Tung University in partial Fulfillment of the Requirements

For the Degree of Ph.D

in

Computer Science and Information

July 2010

Hsinchu, Taiwan, Republic of China

(3)

IP 網路與分波多工網路之最佳化選徑與

網路與分波多工網路之最佳化選徑與

資源配置演算法

學生：陳春秀

指導教授：楊啟瑞博士

國立交通大學資訊科學與工程研究所

Abstract in Chinese

網路的發展由單一功能垂直整合(如: 電話網路、行動網路)逐漸走

向網路融合與水平分工的下世代網路，未來多樣化的服務都可以在這個

融合的 IP 下世代網路(Next Generation Network; NGN)上發展。而融合的

網路平台除了必須提供高服務品質(QoS)和高存活度(Survivability)的網

路服務之外，下世代網路也需要有極高的網路訊務(traffic)承載能力。而

網路服務業者在提供高品質與高存活度的網路平台的同時，也必須兼顧

網路的建置成本(CPEX)與運營成本(OPEX)的降低，因此如何將網路資

源進行最佳化的配置成為重的課題。

光封包交換技術(Optical Packet Switching; OPS)因為可以直接在光

訊號領域進行資料的高速交換，不需要將資料封包轉回電訊號處理，因

此避免目前在高速路由器上所遭遇的超高速電路的技術瓶頸。未來 OPS

將能改變下世代網路基本的運作模式，提供數十 Gbps 的超高速的網路

訊務傳輸功能。但目前 OPS 光封包交換系統的設計仍受到光交換器以及

光儲存器功能尚未成熟的限制，因此光封包交換系統的設計仍是光通訊

研究上重要挑戰。相對來說，光路交換(Optical Circuit Switching; OCS)

可以提供穩定的網路傳輸服務，是目前 WDM 核心網路中最常被使用的

光交換模式。要使光網路達到最佳的使用效率，光路(lighpath)最好能在

要使用之前才建立。除了傳統上長時間固定使用的光路的服務之外，提

供光路預約的服務模式可以使網路服務業者提升運營效率，也可以讓使

用者享有更好的服務。但是如何同時考量光路的預約是否被接受

(4)

(admission) 以及光路的路徑規劃 (routing) 與波長使用 (wavelength

assignment)是相當具挑戰性的問題。

另一個光路交換模式所衍生的問題是光路的容量與其所需承載的

資料量需求有落差的現象，目前在核心網路(core network)與都會網路

(metro network) 最常被使用的 SONET/SDH 網路上也存在類似的問

題。SONET/SDH 網路的傳輸容量級距(granularity)規範是考量傳送電話

網路中的語音話務所規劃訂定的，並不適合目前資料網路所產生的資料

傳送的頻寬需求。例如：資料網路上最常用的 100Mbps 乙太網路與

SONET STS-3 的 155Mbps 就存在 55Mbps 的落差。再者，SONET 傳

輸容量級距需要以四倍方式成長不能分割，非常不適合資料網路的傳輸

需求，造成網路資源使用的沒有效率。此外，為了能夠提供資料的傳送

保護，SONET 提供了 APS 保護機制，其中 1+1 protection 模式提供了

最佳的保護機制，但同時也更加造成網路頻寬浪費。NG-SONET 為了

改善 SONET 的在傳送數據資料傳輸容量級距過大的問題，增加了新的

VCAT 功能，讓點對點的大容量傳輸電路可以由數條容量較小的電路組

成，但仍能維持資料的同步。有了這個新功能，讓兼顧網路資源的使用

同時也能達成高存活度的網路傳輸服務的問題有了新的解法，這也是網

路最佳化選徑與資源配置的一項值得研究的課題。

傳統上要提供高存活度的網路傳輸服務大都透過提供與工作路徑

不同(disjoint)的額外的保護路徑達成，需要使用較多的頻寬來達成。新

近被提出的網路編碼(Network Coding)技術，改變傳統網路資料直接轉送

的模式，選定部分的節點將收到的資料進行編碼再轉送出去，接收端由

不同的路徑接收到資料之後，依照原來編碼的方法反向操作，解出所需

要的資料。因此若是選定某些路徑做為資料的備用路徑，將所要備用的

資料與此路徑上原來傳送的資料編碼後傳送，將可以不增加頻寬的使

用，但是又可以達到資料在有鏈路中斷(link failure)時，仍能順利送達的

目標。

根據上述的問題分析，以及新的技術進展，我們進行下列四項下世

代網路最佳化選徑與資源配置的問題研究，提升網路的運作效能與存活

度。包括 WDM 核心網路中光封包交換系統的設計以及光路預約許可

(admission)與路徑的規劃；以及 NG-SONET 網路以及 IP 網路，以最佳

化的網路路徑規劃與資源配置來達成高網路服務存活度(survivability)

所衍生的路徑規劃配置的問題。本論文相關章節內容說明如下：

在第一章，先簡要介紹下世代網路，說明在 WDM 網路、SONET

網路以及 IP Multicast 網路相關的技術進展，並指出在這些網路中有關

網路資源規劃與配置最佳化的問題。

在第二章，介紹目前在 WDM 網路中重要的關鍵技術元件的功能

與限制，並說明目前在多波長交換網路的光封包交換系統(OPS)所面臨

(5)

的一些研究議題。同時提出克服相關問題的新式 OCPS 交換模式，以

及相關的實驗網路- OPSINET。接著提出一個新的具有 buffer 能力的

OPS 系統架構設計，運用 WDM 多波長的性質、AWG 的交換能力以

及 Cyclic Demux 分單元的特性，設計出 non-blocking 的交換器，同時

將其後所介接的 FDL 運用不同的波長擴充成為多個同樣時間長度的

FDL，大幅降低達成特定 packet loss probability 所需要使用的 FDL 的

數量，並提出此一設計的效能分析。

在第三章，首先介紹光路預約問題的特性，目前在 WDM 網路中，

靜態的光路規劃配置的問題，被稱為 RWA problem，其特性是沒有配

置波長轉換功能的節點中，光路所經過的 link 上都需要使用同一個頻率

的光波。因為這個同一光路上光波連續的限制 (Wavelength

Continuality) ， RWA problem 已經被證明為是一個 NP-Complete

Problem。光路預約需要考量光路許可、路徑規劃以及光波配置，想要

達到最佳化的配置，必須同時這三項因素，基本上光路預約問題也是一

個 NP-Complete Problem。我們運用網路最佳化方法來解決這一個網路

資源配置的問題。

在第四章，先介紹 SONET 網路以及 NG-SONET 網路的新功能；

並討論如何用用最少的網路資源來達成使用者對於網路存活度的期望

的問題。依據數據資料傳送先天上可以因應網路頻寬變化調適的特性，

提出一個在 NG-SONET 網路上的新的網路存活度需求的概念-網路存

活度品質(Quality-of-Survivability)，讓使用者可以定義在網路正常運作

模式以及面臨 link failure 或是 node failure 情況下所需要使用的傳輸

頻寬。配合 NG-SONET VCAT 點對點的大容量傳輸電路可以由數條容

量較小的電路組成，但仍能維持資料的同步的特性，同時考量傳輸電路

的路徑規劃與路徑的存活度需求，將相關的傳輸電路分散配置，降低所

使用的電路因為 link failure 或是 node failure 所造成的影響，同時達成

運用最少的網路頻寬來達成使用者對於傳輸頻寬與存活度的需求。

在第五章，先簡要介紹網路編碼(Network Coding)技術，以及其在提

升網路頻寬的使用效率與網路存活度上相關的研究成果，並介紹運用網

路編碼技術來達成高存活度的網路群播(multicast)的研究基礎。我們分析

了運用網路編碼以及樹狀結構(Tree-based)模式來進行網路群播服務所

需要使用的網路頻寬，並探討其在疏密度不同的網路上的適用性。

在第六章，回顧本研究相關的研究成果，並提出未來可以再進一步

探討的方向。

(6)

Optimization based Approaches for Routing

and Resource Provisioning in IP and Optical

WDM etworks

Student：Chun-Shiow Chen

Advisor：Dr. Maria C. Yuang

Institutes of Computer Science and Engineering

National Chiao Tung University

Abstract

Next Generation Network (“NGN”) shifts from separate vertically integrated application-specific networks to a single network being capable of carrying all services. In addition to providing a technology independent network platform for emerging services, the NGN needs to support ever-increasing traffic demands by a high efficiency and survivability way. One key issue of the next generation network is how to maintain Quality of Service (QoS) and survivability across a wide range of network services while lowering overall network costs (CapEx and OpEx).

Optical Packet Switching (OPS) allows forwarding of ultrahigh bit rate data packets directly in the optical domain and has been proposed as a solution to overcome the “electronic bottleneck”. It will further bring fundamental changes in the design of the Next Generation Network. However, high-speed switching and optical buffering are challenging problems of the OPS system implementation. On the contrary, Optical Circuit Switching (OCS) offers explicit transport guarantees is an important operation paradigm for many network applications. At the current stage most WDM applications follow the OCS paradigm. To get the best network usage, an optical path should be setup just before it is needed.

(7)

Providing a lightpath reservation service to users can increase network operators’ revenue and provide users with better services. How to jointly determine call admission control as well as Routing and Wavelength Assignment is a significant problem to network operators.

SONET/SDH has been dominating transport in metro and backbone networks for decades due to its superior survivability and short failure recovery time. But legacy SONET/SDH only supports contiguous concatenation transport switching over the overall path and its coarse granularity rates are not a good match to packet traffic. NG-SONET VCAT enables forming a high-order end-to-end large-size path by grouping multiple smaller lower-order paths. Based on the VCAT capability, an intelligent path provisioning algorithm can be used to achieve flexible bandwidth usage in NG-SONET networks. Conventional network protection approaches employ extra network resources and precompute backup paths to bypass the failure link or node. It consumes much bandwidth to provide protection. Network coding allows the intermediate nodes not only to forward packets but also encode/decode incoming packets using algebraic primitive operations [17]. By transmitting combinations of incoming data on a backup path enables each receiver node to recover a copy of the data transmitted on the working path if the working path fails.

According to the advances mentioned above, we do some research on the routing and resource provisioning problems of the next generation network to improve the network efficiency and survivability. We deal with four Routing and Resource Provisioning problems in next generation networks. The first two problems are related to transport functions of core networks in how to design a WDM OPS system and the Advance Lightpath Reservation problem in WDM Networks. The third one is about NG-SONET networks to find an optimal solution for Quality-of-Survivable multi-path routing and provisioning problem. The last one correlates to a survivable multicast IP network. This dissertation is organized as next described.

(8)

technology progresses in WDM, SONET, and IP multicast networks. We also point out several routing and resource provisioning problems in these networks.

In Chapter 2, we first give a brief introduction to OPS enabling technologies, discuss the design issues of multi-wavelength optical packet switching networks and propose a new switching architecture to route packets and resolve contentions in both the wavelength and space dimensions together.

In Chapter 3, we focus on the routing and resource allocation issues of prescheduled lightpath provisioning problems and give a Lagrangean relaxation based near-optimal algorithm for advance lightpath reservation in WDM networks. The major challenge is that we need to determine request admission, as well as Routing and Wavelength Assignment jointly.

In Chapter 4, we investigate the problems of how to meet the survivability requirements which users expect while lowering network resources consumed and propose a Quality-of-Survivability concept benefit by a phenomenon that data services are tolerant of bandwidth degraded gradually as the available bandwidth reduces. The goal of routing and resource provisioning is to satisfy bandwidth requirements of different states and minimize total bandwidth consumption at the same time.

In Chapter 5 we briefly introduce the emerging network coding fundamentals first. Based on the observations, network coding has been proposed as a new technique to enhance network throughput and survivability in the literature, we study the problem of optimal routing and bandwidth provisioning for survivable multicast communications using network coding.

(9)

Acknowledgements

完成此論文，要感謝的人很多。謝謝我的良師益友李詩偉教授，給我最佳的指導與督促，以及無窮盡的耐心支持我完成學業。謝謝 Susan Lynn Clevenger，給我最好的英文指導以及很多的關懷。謝謝我的指導教授楊啟瑞老師，給我最大的包容，讓我有足夠的空間找到自己可以投入的研究方向。謝謝我的家人、朋友與公司的同事，在我專心撰寫論文的期間給我最大的支持與鼓勵。

(10)

List of Figures

Figure 1 - Next Generation Network ... 1

Figure 2 - ITU-T Y.2011 – Separation of services from transport in NGN ... 2

Figure 3 - An example of a NGN network configuration ... 3

Figure 4 - SONET multiplexing... 4

Figure 5 - Optical packet switching system ...11

Figure 6 - Arrayed waveguide grating (AWG)... 13

Figure 7 - Ingress router architecture ... 17

Figure 8 - Optical label switched router architecture... 18

Figure 9 - OPSINET testbed configuration... 19

Figure 10 - OPSINET: a snapshot ... 20

Figure 11 - FSOB system architecture. ... 23

Figure 12 - Packet loss probability (PLP) of 2 by 2 system under traffic load 0.8... 32

Figure 15 - Packet loss probability (PLP) of various switching scale settings ... 35

Figure 16 - Packet loss probability (PLP) of various switching scale and traffic load settings... 36

Figure 17 - A new architecture of FSOB system... 36

Figure 18 - Example of an ALR problem... 39

Figure 19 - Lagrangean relaxation algorithm (LGR) ... 45

Figure 20 - Primal heuristic algorithm ... 46

Figure 21 - Simulation results of ALR ... 49

Figure 22 - Illustration of multi path provisioning with survivability ... 53

Figure 23 - Illustrations of graph transformation ... 58

Figure 24 - The USA Network (24 nodes and 86 OC-48 bi-directional links) ... 62

Figure 25 - Simulation results of MP-QoS (on USA network, single-link failure protection) ... 66

Figure 26 - Simulation results of MP-QoS (under the network with various connection degrees, single-link failure protection)... 67

Figure 27 - Simulation results of MP-QoS (on USA network, single-node failure protection) ... 68

Figure 28 - Simulation results of MP-QoS (under the network with various connection degrees, single-node failure protection) ... 69

Figure 29 - Simulation results of MP-QoS (on USA network, single link/node failure protection) ... 70

Figure 30 - Simulation results of MP-QoS (under the network with various connection degrees, single link or node failure protection)... 71

Figure 31 – An example of multicast network architecture ... 72

Figure 32–An example of network coding... 73

Figure 33 - Protection schemes with/without network coding... 77

Figure 34 - Illustrations of graph transformation ... 78

(13)

Acronyms

ALR Advance Lightpath Reservation problem APS Automatic Protection Switching

AWG Arrayed Waveguide Grating

ATM Asynchronous Transfer Mode

BTL Bundle Tree-Based Link Protection Scheme CAM Content Addressable Memory

CapEx Capital Expenditure

CoS Class of Service

CSC Core Switch Controller

DF Deadline First

DSL Digital Subscriber Line FCFS First-Come-First-Serve

FDL Fiber Delay Line

FOWC Fixed Optical Wavelength Converters FSOB Full shared Output Buffer

FWM Four-Wave Mixing

GE Gigabit-Ethernet

GFP Generic Framing Procedure

GMPLS Generalized Multi-Protocol Label switching GPON Gigabit Passive Optical Network

ICL Information and Communications Research Laboratories IP over WDM IP-over Wavelength-Division Multiplexing

IPTV Internet Protocol television

ISIS-TE Intermediate System to Intermediate System with Traffic Engineering ITL Individual Tree-Based Link Protection Scheme

ITRI Industrial Technology Research Institute

JIT Just-In-Time

JET Just-Enough-Time

LCAS Link Capacity Adjustment Scheme

LGR Lagrangean Relaxation

LMP Link Management Protocols MEMS Micro-Electro Mechanical Systems

MP-QoS Multi-Path Provisioning for NG-SONET Networks with Quality-of-Survivability Constraints

(14)

NGN Next Generation Network

NG-SONET/SDH Next Generation Synchronous Optical Networking OADM Optical Add/Drop Multiplexer

OBS Optical Burst Switching OCS Optical Circuit Switching

OCPS Optical Coarse Packet Switching OLSR Optical Label Switched Router OLSP Optic Label Switched Path OPS Optical Packet Switching

OPSINET Optical Coarse Packet Switched IP-over-WDM Network OSPF-TE Open Shortest Path First with Traffic Engineering OPEX Operating Expenditure

OTN Optical Transport Network

OXC Optical Cross-Connect

PLP Packet Loss Probability

QCP QoS Control Processor

QoS Quality of Service QT Quiescence Threshold

ROADM Reconfigurable Optical Add/Drop Multiplexer

RSVP-TE Resource ReSerVation Protocol with Traffic Engineering RWA Routing and Wavelength Assignment

SASK Superimposed Amplitude Shift Keying SASK Superimposed Amplitude Shift Keying

SD Source-and-Destination

SDH Synchronous Digital Hierarchy SHR Self-healing Ring

SLA Service Level Agreement

SOA Semiconductor optical amplifiers SONET Synchronous Optical Networking STS Synchronous Transport Signal TDM Time-Division Multiplexing

TOWC Tunable Optical Wavelength Converters

UB Upper Bound

USHR Unidirectional Self-healing Ring

WC Wavelength Conversion/Wavelength Converter WiMAX, Worldwide Interoperability for Microwave Access WDM Wavelength Division Multiplexing

XGM Cross-Gain Modulation

(15)

Chapter 1. Introduction

Nowadays the IP-based Internet supports various types of services, such as voice, video, interactive games, and emerging cloud computing technology. In order to provide Internet users broadband access and better Quality of Service (QoS), a new network framework called the Next Generation Network (NGN) is proposed [1]. A NGN is an enhanced IP-based network. As shown in [1], it shifts from separate vertically integrated application-specific networks such as the PSTN and the IP network to a single network capable of carrying all services. It is an NGN objective to support services and applications independently of the technologies concerning access networks and core networks. ITU-T Y.2001 provides a general definition of NGN [1], as follows:

A Next Generation Networks (NGN) is a packet-based network able to provide Telecommunication Services to users and able to make use of multiple broadband, QoS-enabled transport technologies and in which service-related functions are independent of the underlying transport-related technologies. It enables unfettered access for users to networks and to competing service providers and services of their choice. It supports generalized mobility which will allow consistent and ubiquitous provision of services to users. [ITU-T Recommendation Y.2001 (12/2004) - General overview of NGN]

Current Networks NGN Network

M o b il e N e tw o rk s P S T N /I S D N D a ta /I P N e tw o rk s C a b le T V N e tw o rk s Voice Services Data Services Video Services Wireline Access Wireless Access Multi-Service IP Backbone Network Control

(QoS, Admission Control, Security%)

Existing & Emerging Services

Single/simple/cost-effective network infrastructure For existing & Emerging services Separate vertically integrated

application-specific networks

(16)

The concept of Next Generation Network is to provide a new service-independent network infrastructure with QoS-enabled features and broadband transport capabilities that support the provision of value-added multimedia services over multiple and heterogeneous QoS-enabled transport technologies. The most significant change is the independence of the data transportation and the service. The NGN functions are divided into service and transport strata according to Recommendation Y.2011 [2], as shown in Figure 2. The service stratum makes requests to transport stratum to get the required network resource and service reliability. NGN transport stratum is required to use the IP protocol for general, ubiquitous and global public connectivity. The IP protocol may be carried over various underlying transport technologies of the transport stratum (e.g., cable access, xDSL, wireless access, Ethernet, optical access, or OTN) according to the operator's environment.

NGN Transport

(cable access, xDSL, wireless access, Ethernet, optical access, OTN)

Point to Point, Point to Multipoint, Multipoint to Multipoint NGN Services

Point to Point, Point to Multipoint, Multipoint to Multipoint

Telephone Services

Data Services (WWW, E-mail, %) Video Services (TV, movie,%)

Figure 2 - ITU-T Y.2011 – Separation of services from transport in NGN

An example of a NGN network configuration is shown as Figure 3. End-user equipment may be either mobile or fixed. End-user networks can be networks within homes or enterprise networks. Access network functions collect and aggregate the traffic from end-user networks to the core network. Usually, the access network functions are performed by access networks and access transport networks. The core network function is responsible for ensuring information coming form the access networks transport throughout the core network. It links access transport networks and connects with other core networks.

(17)

Users/ User Network Access Transport Core Transport NGN Service Stratum OXC OXC OXC OXC OU y OU x OU z OLT splitter 1 : 16~64 3G/HSPA FTTH WDM Ring Network WDM Mesh Network Metro Ethernet OADM _OADM ADM Metro Ethernet ADM Ether Switch OU y OU x OU z OLT splitter 1 : 16~64 3G/HSPA FTTH VoIP Server Location/ Present Server Application Server Message Server IPTV Video Server Server Farm SONET SONET/ NG-SONET NGN Transport Stratum Access

Figure 3 - An example of a NGN network configuration

The access networks connect business and residential subscribers to central offices of their service provider. It spans a distance of a few kilometers perhaps up to 20 kilometers. Diversified technologies, such as xDSL, Cable Modem, Passive Optical Network and WiMAX, are deployed to allow much more flexible use of the access network. However, numerous researchers are working on the emerging access network technologies to provide fully converged services, ubiquitous access and diverse users’ devices. 4G Wireless Systems, GPON (Gigabit Passive Optical Network) and the hybrid wireless-optical network are emerging as a promising technology to provide economical and scalable broadband access.

An access transport network usually spans a city to connect those access networks in part or all of a city and covers distances of a few ten to a few hundred kilometers. The major functions of an access transport network include traffic aggregation and routing. SONET/SDH is the most common technology used in transport networks. It is capable of carrying data from different access networks through a synchronous, flexible, optical

(18)

hierarchy. SONET/SDH is designed to optimize TDM-based traffic. It was initially deployed to carry circuit originated traffic (such as T1 and T3 TDM) over fiber, but it quickly evolved mapping and concatenation capabilities to also carry ATM, Frame Relay, IP and Ethernet traffic. SONET/SDH is a circuit-switched transport and supports contiguous concatenation transport switching over the whole path. The basic units of transmission in SONET are STS-1 (51.84 Mbps), STS-3 (155.52 Mbps), STS-12 (622.08 Mbps), and STS-48 (2.488 Gbps). As shown in Figure 4, multiple lower order signals can be adapted into a higher order signal.

Low-speed mapping function DS1 DS2 E1 STS-1 51.84 Mbps Medium speed mapping function DS3 44.736 STS-1 High-speed mapping function E4 139.264 STS-1 STS-1 STS-1 STS-3c _MUX OC-n Scrambler E/O STS-n ATM or POS STS-3c High-speed mapping function STS-1 STS-1 STS-1 . . . . . .

SONET digital hierarchy

STS-n : (Electrical) Synchronous Transport Signal level n

OC-n : Optical Channel level n

Frame time is constant, size increases in proportion to n

Bit Rate( (Mbps) Electrical Signal 274.176 DS4 44.736 DS3 6.312 DS2 1.544 DS1

North America T- System

Figure 4 - SONET multiplexing

Compared to the requirements for transport networks, SONET/SDH falls short in inefficient payload mapping and lack of framing protocol [3]. The inefficient payload mapping is attributed to concatenation, which has strict payload size restrictions and requires contiguous payload elements. One of the main problems perceived in the SONET/SDH system is its inefficient transport of current Ethernet which runs at 100 Mbps and 1 Gbps. The virtual concatenation (VACT) is a new feature of NG-SONET/SDH. VCAT enables forming a high-order end-to-end large-size path by grouping multiple smaller lower-order paths [4].

(19)

With VCAT, flexible bandwidth usage can be achieved using intelligent path provisioning in NG-SONET/SDH networks.

Ethernet is successful in local area networks. Efforts to extend its boundaries beyond LAN to the carriers' backbone networks are in progress. Metro Ethernet [5] is another new solution for access transport networks. It is based on the Ethernet standard and concerns such issues as CoS, SLAs and management. Metro Ethernet products are widely used in service provider networks, such as mobile and broadband backhaul.

The core network is the backbone of modern IP networks. It spans a distance of a few hundred to a few thousand kilometers in length. The core network provides two major functions. The first one is longhaul data transportation and the other is the exchange of information between different worldwide sub-networks. The technologies currently used in the core and backbone network facilities are WDM, OADM, OXC, and submarine cable systems.

Wavelength Division Multiplexing (WDM) [6] is basically a modern fiber optical transmission technique which multiplexes various optical carrier signals on a single optical fiber by using different wavelengths to carry different signals. The capacity of a given link can be multiplied by simply upgrading the WDM multiplexers and demultiplexers at each end. Since the WDM technique is capable of providing data capacity in excess of hundreds gigabits per second, modern transport networks increasingly employ this technology to utilize the vast transmission bandwidth of fiber to accommodate unprecedented, accelerating demand for bandwidth. With the availability of optical fiber amplifier technologies and the WDM multiplexing technique, optical networking is an immediate success owing to its obvious merits; gracious capacity is increased by adding a wavelength at a time without having to install additional fibers.

(20)

important network elements in WDM optical transport networks. Through configuring these two network elements, network operators can setup lightpaths. The network managerial and reconfiguration capabilities of OADMs and OXCs evolve from fixed to configurable continuously. There are three generations of the optical networking technique evolvement. In the initial phase, OADM or OXC are not configurable, that is, they are fixed. In fixed OADMs, the add/drop and through channels are predetermined and can only be manually rearranged after installation. The second generation of optical networking investigates the reconfigurable aspect of all-optical multi-wavelength networking and the viability of transparent networking due to no electronics element is involved in the data plane. The reconfiguration is applied to each wavelength. An end-to-end optical circuit between a node pair called lightpath can be setup through configuring the optical network elements properly. The configuration can be set through network management or based on a short optical label which includes information related to source, destination, and others. We call the whole wavelength switching paradigm as the Optical Circuit Switching (OCS) paradigm. Most IP over WDM network applications follow the OCS paradigm now. In the IP over WDM network, an optical path is a large pipe to transport data from one end to the other. This makes relatively static utilization of individual WDM channels. The packet routing proceeds in the electronic IP routers which are connected to OXCs/OADMs. These architectures rely on Optical-to-Electrical-to-Optical (O/E/O) conversions since the data transportation is the optical domain, but all packets processing and routing are done in the electrical domain.

The optical networking technology has come to the third generation in recent years. Reconfigurable OADMs(ROADMs), Reconfigurable OXCs and GMPLS (generalized multi-protocol label switching) have been proposed in order to automate lightpath setup procedures. GMPLS [7] is an extension of MPLS and used as the control mechanism for configuring not only packet-based paths, but also optical-based paths. It consists of several

(21)

protocols, including routing protocols (OSPF-TE or ISIS-TE), link management protocols (LMP), and a reservation/label distribution protocol (RSVP-TE). GMPLS serves as a control mechanism for ROADMs and OXCs allowing the creation or termination of label switched lightpaths in the optical network to adapt to changing loads.

Some emerging technologies are developed for enhancing QoS-enabled features and broadband transport capabilities of the NGN network. The most significant technology improvement happening in optical communication is Wavelength Division Multiplexing (WDM). WDM, a modern fiber optical transmission technique, can scale the capacity of a single optical fiber deeply into the terabit per second range. As mentioned before, NGN transport stratum is required to use the IP protocol. However, the scalability of electronic IP routers and their ability to match the rising transmission capabilities of WDM in the optical layer is difficult. This situation led to research interest in optical packet switching (OPS) [8],[9]. In OPS, packets are directly switched in the optical domain in order to bypass the electronic switching bottleneck. OPS paradigm can advocate efficient sharing of wavelength channels among multiple connections satisfying a multitude of applications with diverse Quality of Service (QoS) requirements flexibly and cost-effectively. Current applications of WDM mostly follow the Optical Circuit Switching (OCS) paradigm by making relatively static utilization of individual WDM channels.

Within the NGN architecture, the resource and admission control functions within access and core networks determine the demand admission control, bandwidth reservation and allocation as well as priority handling upon the request from the service stratum [1]. The transport service provision is based on transport subscription information, SLAs, network policy rules, service priority, and transport resource status and utilization information. Most IP over WDM network applications follow the whole wavelength switching paradigm now. The emerging network application like cloud computing relies deeply on the ready-availability of

(22)

broadband and grid computing. Cloud computing applications such as enterprise cloud may need to setup high-speed lightpaths in order to synchronize the distributed database located in diverse campuses periodically. A major feature of such applications is that traffic demands are requested to the network in advance before the connections are set up [35]-[38] and last in a pre-scheduled time period. It will be a new type of transport service request coming from the service stratum in NGN networks. One major challenge arising in these Advance Lightpath Reservation problems has been to jointly determine call admission control as well as Routing and Wavelength Assignment (RWA) [11].

One key issue of the next generation network related to resource and admission control functions is how to maintain quality of service (QoS) and survivability across a wide range of network services while lowering overall network costs (CapEx and OpEx). The survivability refers to a network’s capability to provide continuous service in the presence of failures. How to prevent service interruption, and keep service loss to a minimum if a network failure is inevitable, becomes a critical issue. New technologies like NG-SONET and network coding provide new capabilities to improve service survivability. Virtual concatenation, a new function of NG-SONET, enables forming a high-order, end-to-end, large-size path by grouping multiple smaller lower-order paths. Those lower-order paths may individually take different routes to reduce the damage caused by a link failure and improve service survivability. Network coding allows the intermediate nodes not only to forward packets but also encode/decode incoming packets using algebraic primitive operations [17]. By transmitting combinations of incoming data on a backup path enables each receiver node to recover a copy of the data transmitted on the working path if the working path fails. According to the advances mentioned above, some research is needed on the routing and resource provisioning problems of the next generation network to improve the network efficiency and survivability.

(23)

In this dissertation, we deal with four Routing and Resource Provisioning problems in next generation networks. The first two problems are related to transport functions of core networks in how to design a WDM OPS system and the Advance Lightpath Reservation problem in WDM Networks. The third one is about NG-SONET networks to find an optimal solution for Quality-of-Survivability multi-path routing and provisioning problem. The last one correlates to a survivable multicast IP network. The remainder of this dissertation is organized as next described. In Chapter 2, we first give a brief introduction to OPS enabling technologies, discuss the design issues of multi-wavelength optical packet switching networks and propose a new switching architecture to route packets and resolve contentions in both the wavelength and space dimensions together. In Chapter 3, we focus on the routing and resource allocation issues of prescheduled lightpath provisioning problems and give a Lagrangean relaxation based near-optimal algorithm for advance lightpath reservation in WDM networks. The major challenge is that we need to determine request admission, as well as Routing and Wavelength Assignment jointly. In Chapter 4, we investigate the problems of how to meet the survivability requirements which users expect while lowering network resources consumed and propose a Quality-of-Survivability concept benefit by a phenomenon that data services are tolerant of bandwidth degraded gradually as the available bandwidth reduces. The goal of routing and resource provisioning is to satisfy bandwidth requirements of different states and minimize total bandwidth consumption at the same time. In Chapter 5 we briefly introduce the emerging network coding fundamentals first. Based on the observations, network coding has been proposed as a new technique to enhance network throughput and survivability in the literature [13]-[16],[43]-[46]. We study the problem of optimal routing and bandwidth provisioning for survivable multicast communications using network coding. Finally, concluding remarks and future work are made in Chapter 6.

(24)

Chapter 2. Multi-wavelength Optical Packet Switching

etworks

The ever-growing demand for Internet bandwidth and recent advances in optical communication technologies brings about fundamental changes in the design and implementation of the next generation core networks. In conventional IP over WDM network, optical signals are converted into electronic ones for packet switching inside an electronic switch. The packets are transformed to optical format again for being carried in optical fiber. Such O/E/O conversion incurs high cost and technical difficulty. Furthermore, as data transmission rates are ever-increasing, it is more and more difficult for electronics switching to meet such high-speed requirements. Besides, bandwidth requirements driven by the deployment of new IP services and the increasing penetration of existing services are constantly changing. OCS paradigm is not optimally bandwidth-efficient for transporting traffic from these IP-based services. The OPS provides a packet-based optical switching solution that is, packets are directly switched in the optical domain through an OPS node from any input port to any output port. It is capable of achieving high statistical multiplexing gains, better packet loss performance, and Quality of Service (QoS) differentiation. It has been envisioned as the ultimate solution for the data-centric optical Internet.

A generic functional block of an OPS node, shown as Figure 5 - Optical packet switching system, consists of a multiplexer/demultiplexer pair, an input interface, a switching fabric, a buffer, an output interface, and a control unit. The demultiplexer separates the incoming multi-wavelength optical signal into several single wavelength optical signals. These optical signals are forwarded to the input interface where headers and payloads are decoupled. Then the headers are sent to the control unit which performs electric header processing in order to obtain routing information. The information is used to determine the routing of the switching

(25)

fabric so as to deliver the signal to the right destination. Payloads of these packets are maintained in optical format inside the switching system. They are exchanged by the switching fabric and put into optical buffers if contentions occur. Finally, the new headers will be generated and combined with the original optical packets. The designs of switching fabrics and optical buffers are important problems in optical packet switching system.

Header Processing

Header Regeneration Electrical Header Processing

New Packet Header

Optical Buffers (Fiber Delay Lines)

E/O O/E Packet Header Payload 1 ₅ 1 WDM channel Demux Switching Fabric 2 3 Demux Mux Mux Input Interface Control Unit output Interface 4

Figure 5 - Optical packet switching system

Optical switches which perform switching functionalities to route the incoming packets to the correct output ports in a very short time period are crucial to the design of an OPS system. There are several versatile technologies used to fabricate optical switches, such as micro-electro mechanical systems (MEMS) switches, thermal optical switches, electro-optical switches and others [51]. The characteristics of optical MEMS are low crosstalk, wavelength insensitivity, polarization insensitivity, and scalability. Its switching speeds range from millisecond to sub-millisecond. The advantages of thermal optical switches are polarization-insensitive operations and switching speeds on the order of milliseconds. Electro-optical switches like LiNbO3 switches and semiconductor optical amplifiers (SOA)-based switches offer relatively faster switching speeds. They can switch a packet within a few nanoseconds. Each optical switching technology has unique performance characteristics. To meet the switching requirements of an OPS system, switching speeds of

(26)

optical switch fabrics for packet switching should be in nanosecond order and optical switches need to be strictly non-blocking. Arrayed waveguide grating (AWG) can switch fast, is scalable to large size and consumes little power there for it is promising for constructing high-speed large-capacity switching fabric. Using limited range wavelength converters and arrayed waveguide grating routers to construct a strictly non-blocking optical switching fabric has been proposed in the literature [56].

An AWG provides a fixed routing of an optical signal from a given input port to a given output port based on the wavelength of the signal [50]. Generally, it consists of two star couplers joined together with arms of waveguides of unequal lengths as shown in Figure 6(a). A useful characteristic of the AWG is its cyclical wavelength routing property illustrated by the table in Figure 6 - Arrayed waveguide grating (AWG)(b). Signals of different wavelengths coming into an input port will each be routed to a different output port. Different signals using the same wavelength can be input simultaneously to different input ports, and still not interfere with each other at the output ports. If the multi-wavelength input is shifted to the next input port, the demultiplexed output wavelengths also shift to the next output ports accordingly. An AWG with N input and N output ports is capable of routing a maximum of N2 connections. If an ‘‘out-of-range’’ wavelength is sent to the input port, that wavelength is simply lost or ‘‘blocked’’ from reaching any output port. Because the AWG is an integrated device, it can easily be fabricated at low cost. The disadvantage of the AWG is that it is a device with a fixed routing matrix which can not be reconfigured.

(27)

AWG Input/Output Waveguide λ λλ λN-2 λ λλ λ0 λ λλ λN-1 λ λλ λ0 λ λλ λN-1 λ λλ λ0 λ λλ λ1 λ λλ λ2 λ λλ λN-1 λ λλ λ0 λ λλ λN-1 λ λλ λ1 0 λ λλ λ 0 λ λλ λ 0 λ λλ λ 0 λ λλ λ 0 λ λλ λ 1 λ λλ λ 1 λ λλ λ 1 λ λλ λ −−−1 − λ λλ λ 1 −−−− λ λλ λ 1 −−−− λ λλ λ 1 −−−− λ λλ λ 2 λ λλ λ 2 λ λλ λ 2 λ λλ λ 3 λ λλ λ 3 λ λλ λ 2 −−−− λ λλ λ 2 −−−− λ λλ λ 2 −−−− λ λλ λ 3 −−−− λ λλ λ 3 −−−− λ λλ λ 1 λ λλ λ 4 −−−− λ λλ λ 4 λ λλ λ Wavelength:w Output: v Input: u w=(N-u+v) mod N

(b) Wavelength routing table (a) Arrayed waveguide grating

Figure 6 - Arrayed waveguide grating (AWG)

Wavelength conversion plays a major role in providing the wavelength flexibility in WDM networks. By the introduction of wavelength converters, the data modulated on an incoming wavelength can be transfer to a different outgoing wavelength. Thus, wavelength converters combined with AWG can construct a switching fabric. The design proposed in [49] used a single stage of AWGs is blocking. In [56], novel constructions of strictly non-blocking and rearrangeably non-blocking switching fabrics are given. Various approaches for realizing wavelength conversion have been proposed including cross-gain modulation (XGM), cross-phase modulation (XPM) and four-wave mixing (FWM). It has been shown that the XGM WC’s bit rate can come close to 100 Gb/s. The major drawback of the XGM WC is the degradation of the extinction ratio when converting from shorter to longer wavelengths. However, the XGM WC is very popular due to its simplicity, polarization independence and insensitivity to input wavelength. The XPM scheme generally exhibits better conversion efficiency than the XGM scheme. It has high conversion efficiency, polarization immunity, and no increase in phase noise, but also linear signal up-conversion with a low optical power requirement. Four-wave mixing (FWM) in SOA is an attractive mechanism for wavelength

(28)

conversion since it preserves amplitude, frequency, and phase information; it is generally format independent and also largely bit-rate-independent, thus offering the best transparency. It is superior owing to its ultrafast response. It is also the only approach that allows simultaneous conversion of multiple wavelengths [50].

The all-optical buffer being used for contention resolution is an enabling technology for all-optical packet switched networks. In the optical buffer, data would be kept in optical format (i.e., in the form of light) throughout the storage time without being converted into the electrical domain. All-optical buffers are currently achieved by either fiber delay lines (FDLs), or slow-light technologies. However, the slow-light technologies [54] have been shown to have limited capacity and a delay-bandwidth product; in addition, it is too sensitive for wavelength accuracy that makes it not a feasible solution to support optical buffers. The optical fiber delay line (FDL) is currently the practical way to implement optical buffering. Nevertheless the properties of fiber delay lines differ significantly from properties of electronic buffers. The FDLs can not delay packets for an arbitrary period of time but only for multiples of a basic unit, called the granularity of the FDL. That is, only a discrete set of delays can be provided for contention resolution. Feed-forward and feedback are two kinds of FDL structures in optical buffering [52]. In the feed-forward structure, the packets heading for the same output port at the same time are fed into fiber delay lines of different lengths to resolve contention. A packet coming out of the FDL should be routed to an output port immediately. In the feedback method, a packet may re-circulate in the switch several times until an output port becomes available again. However the feedback architecture leads to larger switch fabric and more crosstalk from which the signal could suffer from significant power loss and noise.

2.1 Optical Coarse Packet Switching

(29)

processing and optical buffer technologies, as well as large switching overhead. In light of this, while some works [9],[10] directly confront the OPS limitations, others attempt to tackle the problem by exploiting different switching paradigms, in which Optical Burst Switching (OBS) [17]-[24] has received the most attention. OBS [17] was originally designed to efficiently support all optical bufferless [18],[19] networks while circumventing OPS limitations. By adopting per-burst switching, OBS requires IP packets to be first assembled into bursts at ingress nodes. Essentially, major focuses in OBS have been on one-way out-of-band wavelength allocations (e.g., Just-In-Time (JIT) [20], and Just-Enough-Time (JET) [18], and the support of QoS for networks without buffers [18],[19] or with limited Fiber-Delay-Line (FDL)-based buffers [21]. In particular the JET-based OBS scheme is considered most effective, wherein a control packet for each burst payload is first transmitted out-of-band, allowing each switch to perform a just-in-time configuration before the burst arrives. Accordingly, a wavelength is reserved only for the duration of the burst. Without waiting for a positive acknowledgment from the destination node, the burst payload follows its control packet immediately after a predetermined offset time, which is path (hop-count) dependent and theoretically designated as the sum of intra-nodal processing delays.

However, its just-in-time-based design results in several complications [25]. These OBS design complications have been the primary motivators behind the design of the OCPS paradigm. To circumvent OPS limitations, a new Optical Coarse Packet Switching (OCPS) paradigm is proposed. Similar to OBS, OCPS supports per-burst switching, which is labeled-based, QoS-oriented, and either bufferless or with limited FDL-based buffers. Being different from OBS which uses out-of-band control, OCPS adopts in-band control in which the header and payload are modulated and transported via the same wavelength.

An OCPS switching network comprises ingress/egress routers, Optical Label Switched Routers (OLSR), and GMPLS controllers. IP packets in an OCPS network belonging to the

(30)

same loss class and the same destination are assembled into bursts. The header of a burst payload carries forwarding (i.e., label) and QoS (e.g., priority) information. A label is the network control information that is swapped at each switching node. The header and the payload of a burst are time-aligned. They are modulated based on a Superimposed Amplitude Shift Keying (SASK) technique [27]. A burst is assembled at an ingress router then forwarded along a pre-established Optical Label Switched Path (OLSP). At each switching node, the header and payload are first SASK-based demodulated. While the header is extracted and electronically processed, the payload remains transported optically in a fixed-length FDL achieving constant delay. Provided with no buffer and that there is more than one burst payload at the switch destined for the same wavelength output, contention occurs and resolution is required. Each burst payload is then SASK-based re-modulated with the new header, and switched according to the label information in the header. Finally at egress nodes, the reverse burstification process is performed and IP packets are extracted from bursts.

The ingress router simply performs burstification. It consists of five major components: Scheduler/Shaper, Gigabit-Ethernet (GE) Interface, Header/Payload Generator, Optical Transmitter, in addition to the GMPLS controller interface, as shown in Figure 7. All label and wavelength information have been downloaded in advance from the GMPLS controller to the ingress router through the GMPLS controller interface and saved.

The Scheduler/Shaper performs QoS-enabled packet aggregation. A burst is generated and transmitted either when the burst size reaches its maximum or the maximum burst assembly time expires, respectively. After having determined a burst to be generated, the Header/Payload Generation module aggregates packets and in turn performs framing for packet delineation in addition to generating header information. Through the GE interface, the header and payload ultimately pass in parallel to the Header/Payload Generator. The payload is then encoded via the 8B/10B Encoder. At the optical transmitter, the header and payload are

(31)

SASK-based modulated and transmitted via a preconfigured wavelength. GMPLS Controller Header/ Payload Generation Scheduler/ Shaper Header Generator Payload Generator 8B/10B Encoder SASK Optical Transmitter G E I n te rf a c e IP Packets GMPLS Interface Optical Burst (header, payload) Ingress Router

Figure 7 - Ingress router architecture

The Optical Label Switched Router (OLSR) (see Figure 8) performs packet/burst switching functions mentioned above. It consists of three major components for each input port (fiber), and one cyclic-frequency AWG switch for the entire node. The three components are: Header Extractor/Eraser, Burst Mode Receiver for Header (BMRH), and Core Switch Controller (CSC). First, the Header Extractor/Eraser extracts the header, and erases it for the payload. While the payload continues traveling optically along the internal FDL, the header is received and recovered in amplitude by BMRH. With the recovered header, CSC performs label swapping, QoS control, and laser tuning control. Notice that owing to the use of an AWG switch, once an OLSP is established, the path is determined locally via the binding from an old label to a new (label, wavelength) pair. All label and wavelength information has been downloaded in advance from a GMPLS Controller and saved in the Content Addressable Memory (CAM).

(32)

F4 F3 F2 F1 F1 Header Extractor/ Eraser FTLS FPGA λ λλ λ1 λ λλ λ2 λ λλ λ3 λ λλ λ4 Switching Fabric( LiNbO3 Switch Or AWG) F1 F2 F3 F4 λ λλ λ3 λ λλ λ1 λ λλ λ4 λ λλ λ2 D E M U X EPDL Burst Mode Receiver

O/E Limiting_Amp.

TH Stb. (1R) Digital Splitter CAM Label Swapping Burst-Mode CDR GMPLS Interface QoS Processing Clock Header Payload Tuning Signal Legends: : Optical Line : Electronic Line Header/Payload Synchronizer (SCM-based) C O M B I N E R Header Generator Payload Regenerator Label Update PC Linux GMPLS Control

Plane Digital Comm.Network

Traffic Engineering

OSPF RSVP-TE LMP

Routing Signaling Control Channel Mgmt

Routing and Wavelength Assignment, Constraint Based Routing,

Call Admission Control

Fiber Delay Line

Core Switch Controller (CSC)

Figure 8 - Optical label switched router architecture

The QoS Control Processor (QCP) is responsible for prioritized contention resolution and header integrity assurance. It is worth noting that, due to AWG, any two bursts arriving from different input ports never contend. On the contrary, contention will occur for bursts arriving from the same input port but carried by different wavelengths, and destined for the same output port. Basically, to switch a burst to the destined output port, an idle wavelength is selected. If all wavelengths are busy, higher priority bursts receive absolute precedence over lower-priority bursts. That is, owing to bufferless, one of the lower-priority bursts being served is preempted and discarded. Finally, with the new (label, wavelength) pair read from CAM, CSC generates the new header and sends a tuning signal to the gated tunable laser. The new header is re-synchronized with the payload having traveled within the FDL.

2.2 Optical Coarse Packet Switched IP-over-WDM etwork (OPSIET)

Based on OCPS, we construct an experimental optical IP-over-WDM network. It is a collaborative project between National Chiao Tung University and the Information and

(33)

Communications Research Laboratories (ICL)/Industrial Technology Research Institute (ITRI). The experimental optical IP-over-WDM network is referred to as OPSINET. The main objective is to examine and resolve fundamental OPS transport and QoS challenges from both the system and network-layer perspectives.

OPSINET consists of three types of nodes - edge routers, optical lambda/fiber switches (OXCs), and Optical Label Switched Routers (OLSRs), with multi-granularity switching capabilities, as shown in Figure 9. While lambda(λ) and fiber OXCs are layer-1 optical devices that switch on a single lambda and an entire fiber, respectively, the OLSRs are layer-3 optical nodes that route and switch packets on a label basis. The label-based routing and switching in OPSINET is managed by the control plane implemented by an out-of-band GMPLS network. The GMPLS network [26] connects a number of GMPLS controllers, each of which governs the routing/switching of an OPSINET node.

Legend:

L2SC: Layer 2 Switch Capable L3SC: Layer 3 Switch Capable

λ λλ

λSC: λλλ- Switch Capableλ

FSC: Fiber Switch Capable Ingress Ingress Router Router (L3SC) (L3SC) OLSR Fiber Switch λ λλ λ Switch L2SC L2SC Bridge Bridge Mac@1 Mac@2 λ λλ λ1 λ λλ λ2 λ λλ λ2 λ λλ λ1 λ λλ λ1 λλλλ₂ λ λλ λ Switch FSC FSC λ λλ λ λ λλ λSSCC λ λλ λ λ λλ λSSCC L2SC L2SC Bridge Bridge L3SC L3SC Edge Edge Router Router Egress Egress Router Router (L3SC) (L3SC) Smart Bit Performance Test Smart Bit OLSR OLSR (L3SC Router) (L3SC Router) Packet Generator _ControllerGMPLS

Figure 9 - OPSINET testbed configuration

A snapshot of OPSINET is displayed in Figure 10. In the basic transport, OPSINET performs efficient per-burst switching by means of the time-aligned design and SASK-based modulation of the header and burst payload. Through this experiment, we perceive that the

(34)

data-centric optical Internet can become a reality based on the OPS technology. Traffic Generator Layer 2 Bridge Ingress Router Egress Router 50KM Fiber Core Switch Lambda Switch Fiber Switch Network Processor Optical Spectrum Analyzer

Figure 10 - OPSINET: a snapshot

2.3 Fully Shared Output Buffer Switch using Cyclic DeMUX

We have implemented the OLSR system, shown in Sec. 2.2, that could perform header/payload mux/demux and optical label swapping. We used LiNbO3 to be the switching fabric. However, LiNbO3 is highly polarization dependent. In addition, due to the small port count number of LiNbO3, the switch size is quite limited. That makes the architecture not scalable. In this section, we further present an AWG based switching architecture with shared output buffers aims to resolve the scalability and packet contention problems of OPS.

The basic requirements of an OPS system are capable of minimizing packet loss probability and achieving QoS differentiation. Although using a large-size non-blocking optical switch or equipped many optical buffers can reduce packet loss probability, it results in poor system scalability. Furthermore the functionality of the optical switches and optical buffers are still under development, therefore the design of the optical-buffered switch architecture and the corresponding scheduling and routing algorithms are still in its early stage.

(35)

In general, the switching subsystem can be categorized as being non-blocking or blocking. For the blocking switches, the Banyan switch is the most scalable and economic architecture but suffers for internal blocking. The non-blocking switching subsystem can always connect input and output ports without affecting other existing connections. However, the non-blocking optical switches are less scalable and economically infeasible due to using a large number of switching elements. There is another type of non-blocking switch constructed by limited range wavelength converters and arrayed waveguide grating (AWG) [53], which converts each packet to an appropriate wavelength thus establishing a path to the required output port according the routing properties of the AWG. The AWG is fast switching, scalable and low power consumption, but the control algorithm to properly decide the wavelength of each packet is a challenge.

According to the position of the buffer, buffered-packet switches are essentially classified as input buffering, output buffering, shared buffering, and recirculation buffering [29]. While input (output) buffering has a separate buffer for each input (output) port, shared buffering allows buffers to be shared among multiple inputs and/or outputs. Recirculation buffering can support dynamic buffering durations at the expense of additional hardware to maintain signal quality. Output buffering has been shown to be effective for packet switching. It is profound by its performance on low packet loss without suffering the head-of-line problem arising in an input buffered switch.

There have been several optical-buffered switch architectures proposed in the literature. Chiaroni et al. proposed a broadcast-and-select optical packet switching architecture [30] that can easily perform many-to-many switching the employment of optical splitters and couplers results in significant power loss. Danielsen et al. [31] proposed three output-buffered optical packet switching architectures which can resolve contentions in the wavelength dimension, but the scalability of space-switches makes the architectures hard to implement. Hunter et al.

IP網路與分波多工網路之最佳化選徑與資源配置演算法

國立交通大學電資學院

資訊科學與工程研究所

博

博

博

博

士

士

士

士

論

論

論

論

文

文

文

文

IP 網路與分波多工網路之最佳化選徑

網路與分波多工網路之最佳化選徑

網路與分波多工網路之最佳化選徑

網路與分波多工網路之最佳化選徑

與資源配置演算法

與資源配置演算法

與資源配置演算法

與資源配置演算法

Optimization-based Approaches for Routing and

Resource Provisioning in IP and Optical WDM etworks

研 究 生：陳春秀

指導教授：楊啟瑞 博士

中

中

中

中

華

華

華

華

民

民

民

民

國

國

國

國

九

九

九

九

十

十

十

十

九

九

九

九

年

年

年

年

七

七

七

七

月

月

月

月

IP 網路與分波多工網路之最佳化選徑

網路與分波多工網路之最佳化選徑

網路與分波多工網路之最佳化選徑

網路與分波多工網路之最佳化選徑

與資源配置演算法

與資源配置演算法

與資源配置演算法

與資源配置演算法

Optimization based Approaches for Routing and

Resource Provisioning in IP and Optical WDM etworks

研究生：陳春秀

指導教授：楊啟瑞博士

IP and Optical WDM etworks

研究生：陳春秀

指導教授：楊啟瑞博士

國立交通大學資訊學院

博士論文

指導教授：楊啟瑞博士

用者享有更好的服務。但是如何同時考量光路的預約是否被接受

(admission) 以及光路的路徑規劃 (routing) 與波長使用 (wavelength