階層式行動網路下之位址分配、封包繞送、以及資訊存取機制

全文

(1)國立交通大學資訊工程學系博士論文. 階層式行動網路下之位址分配、封包繞送、以及資訊存取機制 Addressing, Routing, and Information Provisioning Mechanisms for Hierarchical Mobile Networks. 研究生：徐元瑛指導教授：曾建超. 教授. 中華民國九十四年六月.

(2) 階層式行動網路下之位址分配、封包繞送、以及資訊存取機制 Addressing, Routing, and Information Provisioning Mechanisms for Hierarchical Mobile Networks. 研究生：徐元瑛. Student：Yuan-Ying Hsu. 指導教授：曾建超. Advisor：Chien-Chao Tseng. 國立交通大學資訊工程學系博士論文. A Dissertation Submitted to Department of Computer Science and Information Engineering College of Electrical Engineering and Computer Science National Chiao Tung University In Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy in. Computer Science and Information Engineering June 2004 Hsinchu, Taiwan, Republic of China. 中華民國九十四年六月.

(3) 階層式行動網路下之位址分配、封包繞送、以及資訊存取機制研究生: 徐元瑛. 指導教授: 曾建超博士國立交通大學資訊工程學系博士班. 摘要由於無線網路以及無線裝置的技術更新，依不同的需求已有各式各樣的無線網路因應而生，例如行動隨意網路（mobile ad-hoc networks）、無線感測網路（wireless sensor networks ）、無線網狀網路（ wireless mesh networks ）、以及移動網路（ mobile networks）。一無線網路可以整體移動，也可與其他無線網路重疊進而形成階層式移動網路。階層式移動網路提供移動節點一個彈性的方式使其可以利用適合的無線存取技術連接到基礎網路。但是，當移動節點在階層式移動網路中漫遊時，仍有許多技術問題需要解決。在這篇論文中，我們提出了一些在階層式移動網路中位址分配、封包繞送、以及資訊存取的方法。. 現有的位址分配技術通常採用廣播的方式來尋求位址或檢查是否有重複使用，但廣播會在多點跳躍的網路中造成相當大的負擔。我們提出了一個稱為 Prime DHCP 的位址分配方式使得在分配位址的過程中不需要在整個網路中廣播訊號。Prime DHCP 將網路中每個節點設為 DHCP 代理伺服器，並且獨自執行提出的質數位址配置演算法計算出各自獨有的位址群。DHCP 代理伺服器以及質數位址配置演算法可以共同排除廣播的必要性。. 根據 Prime DHCP 分配的位址結果，我們提出了一個基於質數原則的自我組態繞送協定，此協定可讓每個節點自行找到繞送路徑到其他內部節點。此方法讓每個節點只需要依據目的節點的位址即可找到繞送路徑，不需要定期與其他節點交換路由資訊或發送訊息給目的節點詢問路徑。當有封包需要送到外部網路時，我們在每一個網路中設定至少一個閘道伺服器用以負責繞送進出外部網路的封包。除此之外，我們修改了 Mobile IP 的方法以支援節點的移動。同時我們也提出了一個負擔平衡的繞送協定來平衡多個閘道伺服器的外部網路流量。. i.

(4) 除了網路位址以及繞送的機制以外，我們也提出了在階層式網路下資訊存取的方法。當一個閘道伺服器需要服務許多移動節點時，閘道伺服器外部的網路頻寬勢必會被所有的節點一同分攤而形成瓶頸。為此我們提出了一個雙模網路架構以及一考量負載的排程機制以安排傳輸順序。此外，使用者的資訊有可能在階層式行動網路中存在各式各樣的裝置中。因此我們學習記憶體階層的方式提出了個人資訊階層（ Personal Information Hierarchy，PIH）以及相對應的存取機制來存放使用者的個人資訊。. 我們已針對所有提出的方法進行效能評估。評估的結果顯示 Prime DHCP 可以明顯地降低在分配位址時的時間以及所造成的訊號花費；基於質數原則的自我組態繞送協定可以減少繞送路徑的設定時間；PIH 的架構以及機制可以增加存取的空間以及使用者存取資料的速率。. 關鍵字: 多點繞送無線網路、行動隨意網路、移動網路、位址分配、封包繞送、漫遊、自動組態、資訊階層。. ii.

(5) Addressing, Routing, and Information Provisioning Mechanisms for Hierarchical Mobile Networks. Student: Yuan-Ying Hsu. Adviser: Dr. Chien-Chao Tseng. Department of Computer Science and Information Engineering National Chiao Tung University. ABSTRACT With the advance of wireless and terminal technologies, various wireless networks, such as mobile ad-hoc networks, wireless sensor networks, wireless mesh networks, and mobile networks, are designed for different purposes. A wireless network may move as a whole and furthermore may overlay with one another to form a hierarchical mobile network. Hierarchical mobile networks provide a flexible approach for mobile nodes to access Infrastructure networks with any appropriate wireless technologies. However, many technical issues need to be resolved for mobile nodes to roam within a hierarchical mobile network environment. In this thesis, we propose several mechanisms for network addressing, routing and information provisioning in hierarchical mobile networks. Current address allocations usually involve broadcasting, which introduces huge overhead in multi-hop environments, for address solicitation or duplicate address detection. We propose a Prime DHCP scheme that can allocate addresses to hosts without broadcasting over the whole network. Prime DHCP makes each host a DHCP proxy and run a prime numbering address allocation algorithm individually to compute unique addresses. The concept of DHCP proxies and the prime numbering address allocation algorithm together eliminate the needs for broadcasting. Based on the address allocation result by Prime DHCP, we propose a prime-based self-configured routing protocol for each node to route data packets to other local. iii.

(6) nodes within the same network. With the proposed routing protocol, each node can derive a routing path to a local node according to the node’s address without periodically exchanging routing information with other nodes. Furthermore, the node need not send a routing request to the destined node before forwarding packets to the local node, either. For packets destined to external networks, we configure at least one gateway in each wireless network, and have the gateways responsible for routing packets from/to external networks. To support host mobility, we adopt mobile IP with minor modifications, and we also propose a load-balanced routing protocol to balance external traffic between multiple gateways. Besides network addressing and routing mechanisms, we also propose information provisioning mechanisms for hierarchical mobile networks. When a gateway needs to serve a lot of mobile nodes, the external bandwidth would be shared by all the mobile nodes beneath the gateway. We propose a two-tier proxy architecture and a load-based scheduling mechanism to schedule traffic according to data sizes. Furthermore, personal information of a user might be stored in various devices. Therefore we also propose a personal information hierarchy (PIH) to store personal information and corresponding information accessing policies for PIH by adapting the successful experience of memory hierarchy. We have conducted performance evaluation for all proposed mechanisms. Performance results show that prime DHCP can significantly reduce the signal overhead and the latency for hosts to acquire addresses; prime-based self-configured routing protocol can significantly decrease path setup time and signal overhead; and the PIH architecture and accessing policies together can significantly increase the storage capacity with a negligible decrease in access speed a user can experience in personal information management.. Keywords: multi-hop wireless network, MANET, network mobility, addressing, routing, roaming, auto configuration, information hierarchy.. iv.

(7) Acknowledgments First of all, I would like to thank my dearest Lord Jesus for His eternal love that helps me escape from shadows and continuously have confidence in my research work. I thank my advisor Dr. Chien-Chao Tseng for his guidance and support in the past six years. He always patiently discusses with me and gives me suggestions according to his professional experience in mobile computing. Without his assistance, I can not complete this dissertation so smoothly. I am also grateful to Prof. Yu-Chee Tseng for his aid in system evaluation. His positive and deliberate working attitudes impress me a lot, and these characteristics are exactly what I need to learn. I appreciate my committee, Prof. Chung-Ta King, Prof. Chyi-Ren Dow, Prof. Li-Hsing Yen, Prof. Shiao-Li Tsao, and Dr. Jen-Shun Yang, for their carefully reading my thesis and giving me valuable advice and comments. Special thanks is due to Prof. Yi-Bing Lin for his kindly encouragement and directions of my future life. My lovely colleagues, especially my dear classmate RT and juniors, ML, Chien-An, Yoda, and Welkin, in the Wireless Internet Laboratory are also thanked. They are the greatest team that I have ever met. Thank them to accompany me with all the joy and sorrow. At last, I would love to thank my sweet family for their constant love and trust throughout these years. They are my most powerful back-ups.. v.

(8) Contents. Abstract in Chinese. i. Abstract in English. iii. Acknowledgments. v. Table of Contents. vi. List of Figures. ix. List of Tables. xii. 1 Introduction. 1. 1.1. Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 2. 1.2. Overview of proposed mechanisms . . . . . . . . . . . . . . . . . . . . . . .. 2. 1.3. Synopsis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 3. 2 System Architecture. 4. 2.1. Mobile ad-hoc networks . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 4. 2.2. Wireless sensor networks . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 5. 2.3. Wireless mesh networks . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 6. 2.4. Network mobility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 7. 2.5. Hierarchical mobile networks . . . . . . . . . . . . . . . . . . . . . . . . . .. 8. 3 Addressing Mechanisms for Hierarchical Mobile Networks 3.1. 10. Previous works . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 11. 3.1.1. 11. Best effort allocation . . . . . . . . . . . . . . . . . . . . . . . . . .. vi.

(9) 3.1.2. Centralized allocation . . . . . . . . . . . . . . . . . . . . . . . . .. 11. 3.1.3. Decentralized allocation . . . . . . . . . . . . . . . . . . . . . . . .. 12. 3.2. DHCP relay method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 12. 3.3. DHCP proxy method: prime DHCP . . . . . . . . . . . . . . . . . . . . . .. 14. 3.3.1. Address allocation tree . . . . . . . . . . . . . . . . . . . . . . . . .. 14. 3.3.2. Address allocation procedures . . . . . . . . . . . . . . . . . . . . .. 17. 3.3.3. Exception handling . . . . . . . . . . . . . . . . . . . . . . . . . . .. 21. 3.3.4. Performance evaluation . . . . . . . . . . . . . . . . . . . . . . . . .. 22. Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 24. 3.4. 4 Internal Routing Mechanisms for Hierarchical Mobile Networks. 25. 4.1. Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 25. 4.2. Prime-based self-configured routing protocol . . . . . . . . . . . . . . . . .. 25. 4.2.1. Stateless prime-based self-configured routing protocol . . . . . . . .. 27. 4.2.2. Stateful prime-based self-configured routing protocol . . . . . . . . .. 30. 4.2.3. Performance evaluation . . . . . . . . . . . . . . . . . . . . . . . . .. 32. 4.2.4. Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 37. Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 41. 4.3. 5 External Routing Mechanisms for Hierarchical Mobile Networks. 42. 5.1. External Routing Architecture . . . . . . . . . . . . . . . . . . . . . . . . .. 42. 5.2. Load-balanced routing protocol . . . . . . . . . . . . . . . . . . . . . . . . .. 43. 5.2.1. Protocol design . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 44. 5.2.2. Performance evaluation . . . . . . . . . . . . . . . . . . . . . . . . .. 45. Mobile IP Adaption . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 48. 5.3.1. Problems of roaming in public networks . . . . . . . . . . . . . . . .. 50. 5.3.2. Problems of roaming in private networks . . . . . . . . . . . . . . .. 50. Location-assisted routing enhancements . . . . . . . . . . . . . . . . . . . .. 52. 5.4.1. Localization algorithm . . . . . . . . . . . . . . . . . . . . . . . . .. 52. 5.4.2. Location-based fast handoff . . . . . . . . . . . . . . . . . . . . . .. 53. 5.3. 5.4. vii.

(10) 5.5. Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 6 Information Provisioning Mechanisms for Hierarchical Mobile Networks 6.1. 6.2. 6.3. 54 56. Load-based scheduling mechanism . . . . . . . . . . . . . . . . . . . . . . .. 56. 6.1.1. Two-tier architecture . . . . . . . . . . . . . . . . . . . . . . . . . .. 57. 6.1.2. Load-based scheduling scheme . . . . . . . . . . . . . . . . . . . . .. 58. 6.1.3. Performance evaluation . . . . . . . . . . . . . . . . . . . . . . . . .. 63. Hierarchical personal information management . . . . . . . . . . . . . . . .. 70. 6.2.1. Personal information hierarchy . . . . . . . . . . . . . . . . . . . . .. 72. 6.2.2. Information accessing policies . . . . . . . . . . . . . . . . . . . . .. 73. 6.2.3. Service adaptation strategies . . . . . . . . . . . . . . . . . . . . . .. 77. 6.2.4. Performance evaluation . . . . . . . . . . . . . . . . . . . . . . . . .. 78. Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 86. 7 Conclusion and Future Works. 88. Bibliography. 90. viii.

(11) List of Figures 2.1. Two options of wireless networks: infrastructure and ad hoc. . . . . . . . . .. 5. 2.2. An example of wireless sensor network. . . . . . . . . . . . . . . . . . . . .. 5. 2.3. An example of wireless mesh network. . . . . . . . . . . . . . . . . . . . . .. 6. 2.4. An example of network which is mobile (NEMO). . . . . . . . . . . . . . . .. 7. 2.5. System architecture of hierarchical mobile network. . . . . . . . . . . . . . .. 9. 3.1. Address resolution and DHCP operations for host n to join the MANET. . . .. 12. 3.2. The message flow of DHCP in our architecture. . . . . . . . . . . . . . . . .. 13. 3.3. An example of address allocation tree. . . . . . . . . . . . . . . . . . . . . .. 16. 3.4. Message flows of the address allocation procedure of a new coming node. . .. 18. 3.5. An Example of MANET topology with a new coming host. . . . . . . . . . .. 19. 3.6. An example of address allocation result. . . . . . . . . . . . . . . . . . . . .. 20. 3.7. The address allocation tree. . . . . . . . . . . . . . . . . . . . . . . . . . . .. 20. 3.8. The last two steps of Figure 3.6 when address space=20. . . . . . . . . . . .. 22. 3.9. The effect of recycle period on the address utilization. . . . . . . . . . . . . .. 23. 3.10 The effect of address space on the address utilization. . . . . . . . . . . . . .. 23. 4.1. An example of setting routing path: (a) Address allocation tree; (b) Mesh topology; (c) Routing paths while applying stateless routing protocol; (d) Routing paths while apply stateful routing protocol. . . . . . . . . . . . . . .. 4.2. The procedure of setting routing path in the source routing method of stateless routing protocol. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 4.3. 27. 28. The procedure of setting routing path in the hop-by-hop routing method of stateless routing protocol. . . . . . . . . . . . . . . . . . . . . . . . . . . . .. ix. 29.

(12) 4.4. The procedure of setting routing path in the stateful routing protocol. . . . . .. 30. 4.5. Two network topologies when network size=5*5: (a) Link=4 and (b) Link=8.. 33. 4.6. The signal overhead. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 33. 4.7. Four address allocation types: type0: from central node, type1: from left-top corner node, type2: from center-top node, and type3: from right-top corner node. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 34. 4.8. The average hop count. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 35. 4.9. The hop count difference when link=4 and type=0. . . . . . . . . . . . . . .. 35. 4.10 The hop count difference of different address assignment type when link=4. .. 36. 4.11 The hop count difference of different address assignment type when link=8. .. 37. 4.12 The address allocation tree with virtual links. . . . . . . . . . . . . . . . . .. 37. 4.13 Example of address allocation tree with virtual links. . . . . . . . . . . . . .. 39. 4.14 The hop count difference with limited address space. . . . . . . . . . . . . .. 40. 4.15 The effect of levels for help on the address space requirement. . . . . . . . .. 41. 5.1. External routing architecture. . . . . . . . . . . . . . . . . . . . . . . . . . .. 43. 5.2. An example of load-balanced routing protocol. . . . . . . . . . . . . . . . .. 45. 5.3. The testing environment. . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 46. 5.4. (a) selection of gateway by host N5, and (b) traffic loads at gateway N1 and N2. 46. 5.5. Trace of N5’s throughput over time. . . . . . . . . . . . . . . . . . . . . . .. 47. 5.6. Mobile IP operation scenarios. . . . . . . . . . . . . . . . . . . . . . . . . .. 48. 5.7. An example of ping-pong routing for an MN attaches to a hierarchical mobile network. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 49. 5.8. Operations of NAT traversal. . . . . . . . . . . . . . . . . . . . . . . . . . .. 51. 5.9. Illustration of locating an MN. . . . . . . . . . . . . . . . . . . . . . . . . .. 53. 5.10 An example of selecting candidate points for pre-registration. . . . . . . . . .. 54. 6.1. The hierarchical-proxy architecture for two-tier wireless networks. . . . . . .. 58. 6.2. The load-based scheduling scheme without SPs. . . . . . . . . . . . . . . . .. 59. 6.3. Sequence diagram when the MONET is moving (without SPs). . . . . . . . .. 60. x.

(13) 6.4. Sequence diagram when the mobile network is moving (with SPs). . . . . . .. 63. 6.5. Completion Rates in two-tier and single high-tier networks. . . . . . . . . . .. 65. 6.6. Average waiting times in two-tier and single high-tier networks. . . . . . . .. 65. 6.7. Average waiting times with different hit ratios of the GP. . . . . . . . . . . .. 66. 6.8. Average waiting times with and without scheduling. . . . . . . . . . . . . . .. 67. 6.9. Completion rates with and without scheduling. . . . . . . . . . . . . . . . .. 68. 6.10 Performance for different ratios of moving periods to staying periods. . . . .. 68. 6.11 Performance for different percentages of heavyweight services. . . . . . . . .. 69. 6.12 Performance for different percentages of low-tier bandwidth for lightweight services. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 70. 6.13 Example of four-layered repositories. . . . . . . . . . . . . . . . . . . . . .. 71. 6.14 Example of information division. . . . . . . . . . . . . . . . . . . . . . . . .. 74. 6.15 Memory blocks in mobile devices. . . . . . . . . . . . . . . . . . . . . . . .. 74. 6.16 Algorithm of three-level replacement policy by taking e-mail for example. . .. 77. 6.17 Effect of information hierarchy on number of mails with different memory space sizes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 80. 6.18 Effect of the size of IIMB on miss rates of PIB and IIMB. . . . . . . . . . . .. 81. 6.19 Miss rates of PIB and IIMB with different percentage of the size of memory blocks and fixed number of mails. . . . . . . . . . . . . . . . . . . . . . . .. 81. 6.20 Effect of different access behaviors on miss rates of PIB , IIMB, and CIB. . .. 83. 6.21 Miss rates of PIB and IIMB with different percentage of the size of memory blocks. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 84. 6.22 Effect on average access time. . . . . . . . . . . . . . . . . . . . . . . . . .. 85. xi.

(14) List of Tables 3.1. Qualitative analysis of address allocation mechanisms . . . . . . . . . . . . .. 22. 4.1. Virtual links recorded by nodes in Figure 4.12 . . . . . . . . . . . . . . . . .. 38. 5.1. The impact of traffic fluctuation to load-balancing protocol. . . . . . . . . . .. 47. 6.1. Simulation Parameters. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 64. 6.2. Example of abstraction for e-mail, calendar, and business card. . . . . . . . .. 78. 6.3. Entry sizes of IIMB, PIB and CAHB of a cellular phone. . . . . . . . . . . .. 79. xii.

(15) Chapter 1 Introduction Nowadays, there are various wireless networks developed for different purposes. The most common one is IEEE 802.11 Wireless LAN (WLAN), which constructs the infrastructure by access points. By deploying WLAN access points, users now may access Internet services in anytime at anywhere and with any device having wireless communication capability due to the flexibility and convenience of wireless connectivity. Mobile ad hoc network [42] further provides higher flexibility to free the necessity of base stations. Mobile hosts communicate with one another in a multi-hop manner. Besides two basic wireless communication modes, infrastructure and ad-hoc, there are some other variations for different purposes. Network mobility [57] is addressed to provide Internet access ability for group moving like public transportation. Taking train for example, all mobile hosts on a train form a subnetwork and move together. A mobile router is deployed on the train to take responsibility of attaching to the infrastructure and serving all hosts on the train for Internet accessing services. Another variation is wireless mesh network [26], which is recently designed to represent a promising alternative for broadband Internet access. In the past, wireless networks are limited at some particular hot spots, like coffee shops, airports, or hotels due to the difficulties of installing and maintaining a wired network backhaul connection. Wireless mesh networks replace traditional wired backhaul routers with wireless ones and all of them form a mesh topology. Mesh topology can make such network have good fault-tolerant potential, and is suitable to provide broadband Internet access services.. 1.

(16) 1.1 Motivation All wireless networks above may integrate together to form a hierarchical mobile network to provide users ubiquitous wireless access. Users may roam between various wireless networks. However, the most important problem is how to provide transparent Internet connectivity without user involvements. Two major issues of providing transparent connectivity are address auto-configuration and seamless roaming capability. Since IP addresses represent relative logical locations of mobile hosts, mobile hosts need to change IP addresses when they roam into a new network. Address auto-configuration are required to help mobile hosts configure an address automatically without users involvements when they join in a new network. The other requirement is seamless roaming ability, which helps mobile hosts maintain connections if they switch to another network. Again, this should be achieved without disturbing users. In addition to transparent Internet connectivity, information storage philosophy might also need some minor modifications under the environment of hierarchical mobile networks. In hierarchical mobile networks, users have high flexibility of movement and selecting mobile devices. This might cause information to be stored anywhere or in any device. It is important to information service providers to efficiently organize information for users.. 1.2 Overview of proposed mechanisms In this thesis, we proposed effective addressing and routing mechanisms in hierarchical mobile networks. The proposed addressing mechanism is originated from the canonical factorization theorem of positive integers. Each positive integer can be presented as a product of a unique sequence of prime numbers. Based on this characteristic of positive integer, proposed addressing mechanism can guarantee uniqueness of addresses assigned by different nodes. The proposed routing mechanism utilize addressing results, which can generate a unique address allocation tree to help mobile hosts find routing paths. Since the address allocation tree is unique, each host can easily realize the logical position of itself and destination on the tree. This can enable each host to find routing paths without exchanging routing information or sending routing requests.. 2.

(17) We also proposed a personal information hierarchy in this thesis. As mentioned earlier, information might be distributed all over the network while wireless accesses provides high flexibility and convenience to users. However, this may cause problems for users to manage their personal information. The proposed personal information hierarchy learns successful experience of memory hierarchy in computer systems to help users manage their personal information with both advantages of portability and storage space.. 1.3 Synopsis The remainder of this thesis is organized as follows. Chapter 2 describes the system architecture of hierarchical mobile networks. In Chapter 3, we present two proposed addressing allocation methods, DHCP relay method and DHCP proxy method. Chapter 4 and Chapter 5 describe internal and external routing mechanisms respectively for mobile hosts communicate with other hosts inside or outside the same network. We then describe the proposed personal information hierarchy and information accessing policies in Chapter 6. At last, we conclude the thesis in Chapter 7. 3.

(18) Chapter 2 System Architecture With the advance of embedded computing technologies, portable devices, such as laptops, Personal Digital Assistants (PDAs), and cellular phones, have been widely used. A portable device may even have several wireless interfaces, such as IEEE 802.11 Wireless LAN (WLAN), General Packet Radio Service (GPRS), Personal Handy-phone System (PHS), and/or Bluetooth. In this Chapter, we introduce dsome wireless networks and an integrated hierarchical mobile networks.. 2.1 Mobile ad-hoc networks Wireless communications are typically supported in two models: infrastructure and ad hoc, as illustrated in Figure 2.1. Among these two options, forming a mobile ad hoc network (MANET) is more flexible since it is independent of the availability of base stations. A MANET is a network consisting of a set of mobile hosts, which can roam around at their own will. Since no base stations are supported in such an environment, hosts may have to communicate with each other in a multi-hop manner. Applications of MANETs occur in situations like battlefields, disaster areas, and outdoor assemblies. Hence, intensive research has been dedicated to MANET [31, 42, 49, 61]. A working group called manet has been formed by the Internet Engineering Task Force (IETF) to study the related issues and stimulate research in MANET [50, 4, 54, 46]. The routing protocols in MANETs can be divided into two categories: Reactive MANET Protocols (RMPs) and Proactive MANET Protocols (PMPs). RMPs are on-demand routing protocols, in which mobile hosts send routing requests only when necessary. PMPs are table-driven routing. 4.

(19) Infrastructure. Ad hoc. b d. a. b. a d c g. c Laptop. f f. e. g. e Laptop. PDA Laptop. PDA. Laptop. Figure 2.1: Two options of wireless networks: infrastructure and ad hoc. Wireless Sensor Network Sensor node. Sensor node. Fusion Center Sensor node. Sensor node. Figure 2.2: An example of wireless sensor network. protocols, which make mobile hosts exchange routing information periodically and maintain routing paths to other hosts. Since MANET is a milti-hop wireless network and mobile hosts can move around freely, researchers need to design dynamic protocols with limited signal overhead.. 2.2 Wireless sensor networks A wireless sensor network (WSN) [44, 12, 8, 10, 13, 25, 33, 66] is a wireless network for observing some phenomenon such as temperature, pressure, or relative humidity. As shown in Figure 2.2, a WSN is a highly distributed networks consisting of a large number of small sensor nodes, which are densely deployed either inside the phenomenon or very close to it, and at least one data fusion center. WSNs like MANETs are multi-hop wireless networks. The main purpose of WSNs is to collect sensed data and send these to the fusion center. Instead of sending the raw data to the nodes responsible for the fusion, sensor nodes use their processing abilities to locally carry out simple computations and transmit only the required and partially. 5.

(20) Internet. Figure 2.3: An example of wireless mesh network. processed data. Through wireless network, fusion center will deal with the transferred data from sensors and report useful information to observer. The power consumption of sensor nodes is always critical since it is almost impossible to recharge each sensor node. In order to save power, the transmission range of each sensor node is usually limited to several meters. As a result, WSN applies multi-hop forwarding if fusion center is far away from some sensors. Due to this limitations, there are lots of problems, like power consumption, scalability, reliability or fault tolerant problems, need to be solved in WSNs.. 2.3 Wireless mesh networks Although IEEE 802.11 WLAN wireless broadband networks has become very popular in recent years, the limited reach of signal propagation and high cost of installing and maintaining a wired network backhaul connection have limited WLAN network deployments to homes, offices, public hot spots (in coffee shops, airports, hotels, and other similar locations), and some wide-area hot zones. In addition, installing, managing, and scaling multiple hot spots is very difficult. A wireless mesh network [26, 36, 35, 1] has potential to overcome these limitations to. 6.

(21) Internet. Figure 2.4: An example of network which is mobile (NEMO). create truly unwired cities. As shown in Figure 2.3, backhaul routers in a wireless mesh network have multiple wireless links to other backhaul routers. These backhaul routers can be deployed without wiring. A mesh network could be defined as a network that employs one of two connection arrangements, full mesh topology or partial mesh topology. In the full mesh topology, each node is connected directly to each of the others. In the partial mesh topology, nodes are connected to only some, not all, of the other nodes. A wireless mesh network is considered as a promising alternative for broadband Internet access. To provide reliable Internet access, there are still some problems need to be solved before it could be deployed in reality. These problems include efficiently QoS routing, fault tolerance, auto-configuration, ... etc.. 2.4 Network mobility Although lots of research has been done on host mobility supports that aim to provide continuous Internet connectivity to mobile users [5, 14], it is possible that an entire network may move as a unit and change its point of attachment to the Internet dynamically. A network which is mobile (NEMO) [55, 40, 59, 57] consists of a mobile router (MR) and all its attached nodes, through either wired or wireless interfaces. The MR changes its point of. 7.

(22) attachment to the Internet dynamically while it is moving. Similar to the IP addressing method used in fixed networks, all IP addresses of the nodes in the MONET have the same IP prefix as the MR does. Besides, the nodes attached to the NEMO may themselves be mobile with respective to the MR. Figure 2.4 gives an example of NEMOs that may possibly deployed in a mass transportation like a train. All nodes, including MR, local mobile nods (LMNs), and local fixed nodes (LFNs), in the train form a NEMO and move together. MR could be equipped with multi-tier wireless interfaces such as GPRS, Wireless LAN and/or Bluetooth. LMNs may attach to the NEMO with a low-power but high-bandwidth wireless interface such as wireless LAN. Both LMNs and LFNs may access the Internet through the MR. Furthermore, an LMN or LFN may itself be a router to form a hierarchy of NEMOs. There is another smaller scale NEMO, which is personal area network (PAN). A user may bring several mobile devices, and these devices can form a PAN and one of devices can help all devices in a PAN access the Internet. The concept of mobile networks could extend the reachability of hosts in NEMOs, and possibly help Internet Service Providers (ISPs) in providing seamless services to more users (with fixed or mobile devices). For example, a NEMO deployed in an airplane, a boat, a train or a bus could provide the Internet connectivity to the passengers with fixed or mobile devices such as desktops, laptops, pocket PCs, PDAs, or mobile phones. Furthermore, although LMNs or LFNs move with the NEMO, they are static relative to the NEMO. Therefore, they do not change their points of attachment and need not make any location update while the NEMO is moving. In addition, the MR of the NEMO can connect to the Internet with whatever external wireless interface that is appropriate. The local nodes of the NEMO need not be aware of the changes in the external wireless interfaces of the MR. Although NEMOs can extend our accessibility to the Internet, some problems [56, 60] still remain to be solved before the NEMOs can be widely deployed.. 2.5 Hierarchical mobile networks In previous sections, we introduce several wireless networks with different topologies and purposes. Among these wireless networks, MANETs, WSNs, and wireless mesh networks are. 8.

(23) A WSN. Internet An MN. A cell phone. A NEMO. A NEMO. An MN A MANET A MANET. A PAN. A NEMO. Figure 2.5: System architecture of hierarchical mobile network. multi-hop wireless networks and NEMOs are a single-hop networks. These networks might together form a hierarchical network. As shown in Figure 2.5, a mobile node (MN) could attach to an infrastructure directly or attach to a wireless mesh network, MANET, or NEMO. A MANET, NEMO, or PAN could also attach to the Internet, a wireless mesh network, or even another NEMO. When a network attaches to another network, there must be a node acting as bridge of two different networks. We define such nodes as gateways. For example, MRs are gateways of NEMOs. If there are more than one gateway in a MANET, we define subMANET to represent a sub-network consisting of a gateway and all mobile nodes attaching to such gateway. Similarly, there might be sub-NEMOs or sub-WSNs. In a hierarchical mobile network, a mobile node or a NEMO can freely roam between different networks. This could provide high flexibility to mobile hosts, but it also causes problems in network management to support seamless roaming capability to mobile hosts. In the following chapters, we propose some network and information management mechanisms in hierarchical mobile networks.. 9.

(24) Chapter 3 Addressing Mechanisms for Hierarchical Mobile Networks We know that a mobile host needs to set an IP address before it starts to communicate with other hosts. Since prefix of an IP address represents logical position of a device, each network usually manages their addresses locally. Traditionally, a network can manage their addresses statically or dynamically. In static address management, users need to have an address in advance and configure the address manually. In dynamic address management, a network can apply some dynamic address allocation methods, such as Dynamic Host Configuration Protocol (DHCP) [45] to assign addresses. In our architecture of hierarchical mobile networks, mobile hosts need to acquire an address whenever they roam into a new network. We can divide all networks in our architecture into two categories: single-hop and multi-hop networks. MANETs, WSNs, and wireless mesh networks are multi-hop networks and NEMOs belong to single-hop networks. It is impossible for mobile hosts to configure addresses manually in hierarchical mobile networks, so we need to find proper address allocation methods for each network. DHCP involves three rounds of broadcasting, which may cause huge signal overhead or even broadcast storm problem [51] in a multi-hop environment. Therefore, single-hop networks like NEMO in our architecture can apply DHCP directly, and we propose efficient addressing allocation mechanisms in multi-hop environments in this Chapter. We will describe prior researches on this problem in the following section 3.1. We then propose two addressing allocation mechanisms for multi-hop environments in this Chapter. The first one is DHCP relay method in Section 3.2, which can reduce signal overhead. The. 10.

(25) second one is DHCP proxy method in Section 3.3 to reduce both signal overhead and latency of address acquiring process.. 3.1 Previous works Several dynamic address allocation schemes have previously been proposed for MANETs. According to the prior research [53], they can be classified into three categories:. 3.1.1 Best effort allocation Schemes in this category can not guarantee address uniqueness. Prophet scheme [21] proposes a complex address generation function for each host to generate a sequence of addresses to be assigned to new coming neighbors. The proposed function tries to use huge address space to degrade the percentage of generating a same address from two nodes. When a new node joins a network, it can simply acquire an address from its neighbors. However, the proposed function in Prophet method still can not guarantee the uniqueness of addresses assigned by different nodes. In other words, there still exists probability that two nodes can generate a same address. Therefore, even with a large address space, prophet scheme may still needs some mechanisms, such as Duplicate Address Detection (DAD) or weakDAD [37], to resolve address conflicts. DAD will cause broadcast storm problem and weakDAD will introduce extra packet overhead by adding MAC address into IP headers for all data packets.. 3.1.2 Centralized allocation A centralized server is deployed to manage all addresses in this category of address allocation schemes. DHCP [45] is a typical example, but it needs broadcasting for both server discovery and DAD. ODACP [53] attempts to reduce broadcasting by having the server broadcast advertisement periodically so a new host can directly register its addresses to the server. A longer advertisement interval does help to reduce the overhead of broadcasting, but it also results in longer latencies for hosts to obtain addresses.. 11.

(26) DHCP server B DHCP server C. DHCP server A. f. DHCP Relays. i. g. l d. h j. DHCP_discover (Unicast). k m. e. DHCP_discover (Broadcast) n. Figure 3.1: Address resolution and DHCP operations for host n to join the MANET.. 3.1.3 Decentralized allocation A host could acquire an address by itself or from a neighbor and then performs DAD to ensure the uniqueness of the address. In AAA [9], hosts randomly select an address in the range of 169.254/16. In MANETconf [34], each host stores all addresses used in the MANET, and a new coming host acquires an address from one of its neighbors. The neighbor then broadcasts a query, on behalf of the new host, for DAD throughout the network. Mohsin [32] employs a buddy system for address allocation, but it is difficult to manage address blocks among all MANET hosts.. 3.2 DHCP relay method The first address allocation method we propose is DHCP relay method [23]. As shown in Figure 3.1, we install at least one DHCP server in each MANET. If there are more than one DHCP server in a MANET, we define a sub-MANET consisting of a DHCP server and all nodes with addresses assigned by such DHCP server. To avoid confusion, we assign an exclusive section of IP addresses to each DHCP server. Note that we also allow a host to use its old IP address after roaming into a new sub-MANET.. 12.

(27) Mobile Host Boot. DHCP Relay. DHCP Server. DHCP_Discover DHCP_Offer DHCP_Request DHCP_Ack. Configured. DAD (by ARP) Lease is going to expire. DHCP_Request DHCP_Ack. Lease is extended. Broadcast. Unicast. Figure 3.2: The message flow of DHCP in our architecture. Taking Figure 3.1 for example, when a new mobile host n joins the MANET, it first broadcasts a DHCP discover. Three nodes, j, k, and m, belonging to two sub-MANETs receive the request and help forwarding the request to their DHCP servers. To avoid the broadcast storm problem [51], the forwarding is done by unicast. This can be achieved by configuring each internal mobile host as a DHCP relay. Figure 3.2 illustrates the related DHCP message flows. The DHCP Discover is forwarded by the DHCP relay to the DHCP server via DSDV routing. The DHCP server then replies a DHCP Offer by including an available IP address. Note that the host may receive multiple offers from several servers. So the host will broadcast a DHCP Request to notify all DHCP servers the IP address that it selects. Again, the notification will be supported by unicast. The selected server then replies a DHCP Ack if the IP is still available. Afterward, the Duplicate Address Detection (DAD) procedure will be executed to ensure that no other host is using the same IP address. There is also a parameter, called lease, after which the mobile host has to renew its temporary IP address with the same DHCP server. After obtaining an IP address, the host will turn itself into a DHCP relay of the DHCP server that it selects. By configuring hosts except DHCP servers as DHCP relays can make messages be sent by one-hop broadcasting instead of whole-network broadcasting. Therefore, the proposed DHCP relay method can reduce signal overhead. Unfortunately, a host still needs to wait until DHCP. 13.

(28) server sending back the response. In the next section, we will propose a DHCP proxy method to further reduce latency.. 3.3 DHCP proxy method: prime DHCP In this section, we propose a Prime DHCP scheme for address allocation without broadcasting in the whole MANET during the address allocation process. In the proposed prime DHCP, each host serves as a DHCP proxy that can assign addresses to new hosts by running a proposed Prime Numbering Address Allocation (PNAA) algorithm individually to compute unique addresses for address allocation. The use of DHCP proxies and the PNAA together eliminate the need for broadcasting in the whole MANET. As mentioned in the Section 3.1, almost all previous address allocation schemes for MANETs rely on broadcasting for server discovery or DAD. We propose Prime DHCP to achieve these two functionalities without broadcasting. Prime DHCP configures each host as a DHCP proxy, so all hosts are eligible to assign addresses and a new host can acquire an address simply from its neighbors. Besides, each DHCP proxy runs PNAA individually to compute unique addresses for address allocation so that DAD is not required in prime DHCP.. 3.3.1 Address allocation tree In order to eliminate the necessity of DAD, we propose a PNAA algorithm to guarantee the uniqueness of addresses. PNAA is originated from the canonical factorization theorem of positive integers, that is, every positive integer can be written as a product of prime numbers in a unique way. Before describing proposed mechanism, we first define prime factorization sequence of an integer in Defination 1. According to the canonical factorization theorem [17], each integer can be expressed as a production of prime numbers. For example, integer and thus. pfSeq (12) =. 12 is equal to 2 2 3,. (2; 2; 3). We prove the prime factorization sequence of an integer is. unique in Lemma 1. Definition 1. The prime factorization sequence of an integer n. > 1, written pfSeq (n), is an. ascending ordered set of all prime factors of n, where the product of all elements in the set is. 14.

(29) equal to n.. pfSeq (n) = (p0 ; p1 ; p2 ; :::; pm ) where pi pj 8i < j and. 1. m Y i=0. pi = n:. Lemma 1. The prime factorization sequence of each integer n > , pfSeq. (n), is unique.. Proof. We prove this lemma by mathematical induction. Basis:. pfSeq (2) = (2), which is unique.. (n) is unique 8n k Induction: n = k + 1, assume pfSeq (n) is not unique.. Hypothesis: assume pfSeq. pfSeq (n) = (p0 ; p1 ; :::; pl ) = (q0 ; q1 ; :::; qm ) We can prove pfSeq. (n) is unique by showing p0 = q0 since pfSeq(n=p0) is unique. according to the hypothesis.. * n = p0 p1 ::: pl = q0 q1 ::: qm. ) ) ) ). p0 j(q0 q1 ::: qm ) p0 jqj for some j p0 = qj (* p0 and qj both are prime numbers) p0 q0 (* qj q0 ). Similarly,. p0 q0. ) p0 = q0 ) pfSeq (n) is unique when n = k + 1:. By mathematical induction, the prime factorization sequence of each integer is unique. Figure 3.3 gives an example of addresses each DHCP proxy can assign. The first node in the network is root proxy, and it configures its address as 1. The root proxy allocates all prime numbers, in ascending order, to new nodes attached to it. For a non-root DHCP proxy with address n and pfSeq. (n) = (p0; p1; :::; pm), it can assign the address equal to its own address. multiplied by a prime number, starting from its largest prime factor pm . Each node can also identify the address of its parent simply by dividing its own address by the largest prime factor. 15.

(30) A. B. F. K. 4 18. G. L. C. 2 H. 30. M. D. 3 I. 6 ... 9. 1 5. E. 7. .... J. 15 ... 25.... 42 .... Figure 3.3: An example of address allocation tree.. 6 for example, the largest prime factor of 6 is 3 and the sequence of addresses node G can assign is, 6 3, 6 5, 6 7, and so on up to the largest address bounded by the address space. Node M in Figure 1 has the address 42, and the address of its parent is 42=7 = 6. of its own address. Take node G with address. Because the prime factorization sequence of each integer or address is unique, we can prove in Theorem 1 that no two proxies can generate the same address by running the proposed PNAA algorithm. Therefore, DAD is not necessary during the process of address resolution. Besides, in Theorem 2, we also prove that each integer or address is assignable, so there is no hollow address and we can utilize all addresses efficiently. Theorem 1. No two nodes will assign the same address in PNAA. Proof. Given an integer n,pfSeq. (n) = (p0; p1 ; p2; :::; pm). According to PNAA, the address assignment path of integer n is. (1; 0j=0pj ; 1j=0pj ; 2j=0pj ; mj=0pj ) * pfSeq (n) is unique (proved in Lemma 1) ) the address assignment path of integer n is also unique, and thus no two nodes can assign the same address.. 16.

(31) Theorem 2. Each address is assignable in PNAA. Proof. We prove this theorem by mathematical induction. Basis: Address 1 is assignable because it is the address of the first node (root) in the network. Hypothesis:For all positive integer n k , n is assignable.. = k + 1, and pfSeq(k + 1) = (p0; p1; :::; pm ) By Lemma 1, pfSeq (k + 1) is unique.. Induction: For n. + 1 is a prime number: k + 1 is assigned by the root. case 2.k + 1 is not a prime number: * (k + 1)=pm is less than k ) (k + 1)=pm is assignable according to the hypothesis k + 1 is assigned by the node with address (k + 1)=pm case 1.k. ) n = k + 1 is also assignable. By mathematical induction, all positive integer n is assignable.. 3.3.2 Address allocation procedures After being sure that each node can assign disjoint addresses by running PNAA, we describe operations of proposed prime-based self-configured addressing mechanism. First, each DHCP proxy maintains a sequence of prime numbers and its own allocation status, a pointer to the prime number multiplied for the last assigned address. When a new node joins a network, Figure 3.4 illustrates the message flows of how the new coming node acquires an address. First of all, the new node issues a DHCP Discover broadcast message as a normal DHCP client does and starts a timer. When the neighbor nodes, DHCP proxies 1 and 2, receive the message, they runs PNAA algorithm to generate an address, and then encapsulates the address in a DHCP Offer message. Before the timer expires, the new node gathers all DHCP Offer messages with available addresses, and it chooses the smallest address to prevent the address allocation tree from growing too fast. The choice is set in DHCP Request broadcast message to inform all its neighbors. Finally, the chosen proxy 1 updates its address allocation status and then sends a DHCP Ack to the new node for confirmation. Consequently, instead of. 17.

(32) DHCP Proxy 1. New Node. DHCP_Discover. Boot. DHCP Proxy 2 DHCP_Discover. DHCP_Offer DHCP_Request. DHCP_Request DHCP_Ack Configured. Broadcast. Unicast. Figure 3.4: Message flows of the address allocation procedure of a new coming node. whole-MANET broadcasts, DHCP Offer and DHCP Request incur just a single-hop broadcast, which is necessary anyway in the wireless environment. We use Figure 3.5 as an example MANET to illustrate the operations of Prime DHCP. It should be noted that Figure 3.3 is just a logical tree of the address allocation. Once a host obtains an address, it can move freely and continuously use the address as long as it is still in this network. Figure 3.5 is a possible physical topology of MANET hosts in Figure 3.3. Without loss of generality, we assume that the address space of this MANET is 128 in the following discussion. Furthermore, the messages can be forwarded by any proactive routing protocols for MANETs. When a new mobile host N joins a MANET, it issues a DHCP Discover broadcast message as a normal DHCP client does. When the neighbor hosts, L, B, and F, receive the message, they start serving as DHCP proxies of the host N. Rather than forwarding the broadcast message further, each DHCP proxy runs PNAA algorithm to generate an address, and then encapsulates the address in a DHCP Offer message. Note that proxy L relays the DHCP Discover to its parent proxy G for help because its minimum assignable address is 150, which exceeds the address space 128. Therefore, host N receives three DHCP Offer messages offering addresses 66, 10, and 8, respectively, from DHCP proxies G, B, and F. Host N chooses the smallest address 8 to prevent the tree from growing too fast, and broadcasts its choice in a DHCP Request. Again, its neighbors do not flood the message further, but proxy L relays the message to its parent proxy G. Finally, the chosen proxy F updates its address alloca-. 18.

(33) DHCPProxies. H(9) E (7). K (18) D (5) DHCP_Release. J(25) M (42). I (15). L (30). B (2). G(6) Assigned: 42 Recycle: {18} C (3). F (4). A (1) 10. 8. 66. (1) DHCP_Discover (3)DHCP_Request (2)DHCP_Offer (4)DHCP_Ack. N. Figure 3.5: An Example of MANET topology with a new coming host. tion status and then sends a DHCP Ack to host N for confirmation. Consequently, instead of whole-MANET broadcasts, DHCP Offer and DHCP Request incur just a single-hop broadcast, which is necessary anyway in the wireless environment, and possibly seldom multi-hop unicasts. Figure 3.6 gives an example of address allocation results for a 3x3 mesh network. We assume the order of nodes joining the network is from node A to node I. In the first step (a), node A is the only node in the network, so it configures its own address as 1. Two nodes B and C then attach to node A and obtain address 2 and 3 respectively as shown in step (b). In step (c), node E receives two addresses 6 and 9 from node B and C respectively, and it chooses the smaller one as its address. Similarly in step (d), node G also chooses the smaller address, 8 among 8 and 18. In steps (e) and (f), nodes H and I choose the smaller addresses 18 and 16 from node E and G, respectively. Figure 3.7 is the address allocation tree of Figure 3.6. The solid lines represent branches of the address allocation tree, whereas the dotted lines are the links not on the address allocation paths. In order to avoid the address leak problem, a host should depart gracefully by informing its parent of leaving, and each host maintains a recycle list to record the allocation statuses for its departed children in its allocation status. For example, proxy K in Figure 3.5 is leaving, so it sends a DHCP Release message to its parent proxy G. When proxy G receives a DHCP Discover from a new node, it will first give out the lowest recycled address and inform the new node the allocation status associated with the offered address. If the root is about to. 19.

(34) A. C. F. A. 1 B. C. 1 E. H. F. 3. B. E. H. G. I. 2 D. G. I. D. (a) A. (b). C. F. 1 B. E. C. 1 H. 2 D. A. 3 B. 6 G. I. D. G. 4. B. F. 3 E. D. A. 9. 6 I. 9. E. 2. H. 6. D. 8. F. 3. B. 18. G. 4. C. 1. H. 2. I. 8 (d). C. 1. H. 6. (c) A. 9. E. 2. 4. F. 3. 18. G. 4. I. 8. (e). 16. (f). Figure 3.6: An example of address allocation result.. A. 1. B. 2 D. 4. C. 3. E. F. 6. 9 H. G. 8. 18. I. 16 Figure 3.7: The address allocation tree.. 20.

(35) leave, it informs its greatest descendent to be the new root proxy. However, it is likely that a host may leave gracelessly, packets may be lost, address may be limited, or MANETs may merge or split. In the next subsection, we will explain how the prime DHCP handles these exceptions.. 3.3.3 Exception handling Most message losses, except DHCP Release that can be treated as a graceless departure, can be recovered by having the sender set a timer and resend the messages when the timer expires. For a host left gracelessly, its address may become not usable. In order to reclaim leaked addresses, the root proxy periodically broadcasts a DHCP Recycle message to ask all hosts report their address allocation statuses, including addresses they have given out or recycled. By gathering statuses from existing hosts, the root proxy can reconfigure the address allocation tree, and inform each proxy of updated allocation status. The DHCP Recycle broadcast message could be piggybacked in the messages of MANET routing protocol, DSDV [43] for example, so no additional overhead will be introduced by the address recycling. Network merging and partition can also be detected by periodically recycle process. When two MANETs merge, a root can receive allocation statuses from nodes in the other network, detect address conflicts, and ask one of the two conflicting nodes to inquire a new address. On the other hand, although network partition will not result in address conflicts, the split MANET needs a new root for address recycle. If a node misses several recycle messages, it may claim to be the new root by sending a DHCP Recycle after a backoff time reversely proportional to its address. This can make the node with the largest address the new root to increase address utilization. When the address space is limited, a node might gain an address exceeding the range of addresses by running PNAA while receiving an address request. In such case, the node relays the request to its parent node to obtain a legal address. Assuming the address space of the network in Figure 3.6 is limited to 20, we illustrate the last two steps of address allocation results in Figure 3.8. If a DHCP proxy generates an address greater than the address space, it relays the request message to its parent node to retrieve an address. As shown in step (e), node H acquires. 21.

(36) A. C. 1. F. 3. B. E. 2 G. 4. C. F. 1. H. 6. D. A. 9 B. 15. 3 E. H. 2. I. D. 8. 6 G. 15 I. 4. (e). 9. 8. 5. (f). Figure 3.8: The last two steps of Figure 3.6 when address space=20. Table 3.1: Qualitative analysis of address allocation mechanisms DHCP MANETconf ODACP Prophet PrimeDHCP Uniqueness. Yes. Yes. Latency. O(4l) O(4td). O(2l) O(2td). Complexity. Low. High. Signal overhead. Yes. No. Low. High. O(2l) O(n=2) O(2td) O(2t). Yes. O(n=2) O(2t) Low. address from its neighbor node E and F. The next address node F can assign is 27, which is larger than the address space, so it relays the request to its parent node C and obtains an address 15. Since 15 is smaller than 18, which is assigned by node E, node H chooses 15 as its address. In the last step, node I similarly obtains address 5 from neighbor node H.. 3.3.4 Performance evaluation Table 3.1 shows the qualitative analysis of the proposed prime DHCP and other address allocation mechanisms. Assume the numbers of hosts and links are. n and l, respectively; the. network diameter is d; and the average one-hop latency is t. First, Prophet is the only method that can not guarantee the uniqueness of addresses. Second, DHCP needs to perform server discovery and DAD, so at least. 4l hosts need to process signal packets and the latency is 4td.. MANETconf and ODACP need to perform DAD and server advertisement respectively, and. 2l and 2 t d. Prophet and Prime DHCP both send requests to neighbors only, so the overhead is the average degree (n=2) of each node in the network and the latency is 2t, assuming that the address space is sufficient. At last, MANETconf and thus the overhead and latency are. Prophet scheme involve more complicated computation in address allocation. Since the behaviors of address assignment in Prime DHCP and Prophet scheme are similar, the latency and overhead of Prime DHCP can be referred to [21]. Here we focus the analysis on. 22.

(37) 100%. 80% c=30 c=100. 60%. c=200. U. c=300 40%. c=400. 20%. 0% 0. 5000. 10000. 15000. 20000 t (sec.). Figure 3.9: The effect of recycle period on the address utilization. 100%. 80%. N = 2 24 60%. N = 2 20. U. N = 2 16. 40%. N = 28. 20%. 0% 30. 200. 500. 800. 2000. 5000. 8000 15000 30000. c (sec.). Figure 3.10: The effect of address space on the address utilization. how recycle period affects the address utilization, i.e., the percentage of the effective addresses that Prime DHCP has given out or can possibly assign currently. Suppose the arrival and departure of hosts follow a Poisson distribution with the mean rates. and , respectively.. The address utilization Uk before the kth recycle can be represented as Equation (1), where. N is the size of address space; n~ (Ck ) is the number of hosts in the MANET at the time of kth recycle; and is the average number of descendants of a host.. Uk = 1. n~ (Ck ) gra eless per entage Nsubtree : N. (3.1). Figure 3.9 shows the address utilization Uk at different time instances for different recycle periods. ( ), where N = 256, = 0:8 and = 0:9. As shown in the figure, when the recycle. period becomes shorter, the recycle process will be performed more frequently but utilization. 23.

(38) will be better. We can also observe from Figure 3.10 that although the utilization decreases as the recycle period increases, the speed of descent is slow if the address space is sufficiently large. Therefore, address recycling introduces only slight overhead if the MANET has an address space greater than class B, and this could be achieved by configuring the MANET as a private network.. 3.4 Summary We apply DHCP directly for address allocation in single-hop networks. In multi-hop networks, we propose two methods to reduce signal overhead and latency of DHCP. The DHCP relay method configures all hosts except DHCP server as DHCP relay, so all DHCP signal message will be relayed to DHCP server directly without be flooded all over the whole network. The DHCP proxy method we proposed is Prime DHCP to make each host as a DHCP proxy. The proposed Prime DHCP method is simple but very effective in terms of both signal overhead and address allocation latency. Each proxy assigns addresses according to the prime number tree, so DAD broadcasting is also unnecessary.. 24.

(39) Chapter 4 Internal Routing Mechanisms for Hierarchical Mobile Networks After solving address allocation problem, we next need to solve the problem of routing in hierarchical mobile networks. We first need to make sure that each mobile host can communicate with other hosts in the same network, which is internal routing.. 4.1 Motivation There are four different wireless networks in our hierarchical mobile networks. Among these four networks, NEMO is single-hop network. Mobile hosts in a NEMO can just send packets to MR and then to other hosts like original local routing in Local Area Networks. MANETs have flexible network topologies, and there are already lots of routing protocols proposed for MANETs. The left two wireless networks, wireless sensor networks and wireless mesh network, are both multi-hop networks with almost fixed network topologies. They do not need complicated mechanisms designed for MANETs since they have fixed topology, but it is also not proper for them to apply routing mechanisms in wired environments due to huge signal overhead caused by multi-hop behaviors. Therefore, we propose a prime-based selfconfigured routing protocol in Section 4.2 to solve internal routing problems for multi-hop wireless networks with fixed topologies.. 4.2 Prime-based self-configured routing protocol In this section, we propose a prime-based self-configured routing protocol, which is a routing protocol applied in a local network to solve internal routing problem. This protocol take advan-. 25.

(40) tage of address results of PNAA described in Chapter 3. By employing PNAA, addresses of all nodes can form a unique addressing allocation tree which can represent location relationship of all nodes. Each node can easily be aware of locations of all nodes by their addresses, and this they can find routing paths by themselves without either periodically exchanging routing information or asking other nodes when necessary. Before describing the proposed routing protocol, we define Sequence GCD of two integers in the Definition 2 first. Nodes can find their routing path by computing the sequence GCD of source and destination addresses. In the following subsections, we will present how nodes find routing paths by just computing sequence GCD of the addresses of itself and the destination node. We proposed stateless and stateful routing protocols. In the stateless routing protocol, all nodes do not need to record anything, but the routing path may be longer. The stateful routing protocol can help nodes to find a shorter path but nodes need to record the addresses of its neighbors.. = (p0; p1 ; p2; :::; pm) and pfSeq(b) = (q0 ; q1; q2; :::; qn), Q we define seqGCD (a; b) as the sequence GCD of integers a and b. seqGCD (a; b) = rr is the product of all elements in the longest common prefix sequence y = (r0 ; r1 ; r2 ; :::; rj ) of pfSeq (a) and pfSeq (b). Which means j is the largest number that makes ri = pi = qi for all 0 i j . If there is no common prefix of pfSeq(a) and pfSeq(b), seqGCD(a; b) is equal Definition 2. Assuming. pfSeq (a). j 0. to 1.. 26.

(41) A. 1 B. F L. 4 O. 8 P. G. 12. 6 M. C. 2. K. 3. H. 10. N. 18. D. J. 5 I. 9. E. 7 J. 15. D. 25 E. 25. O. 3 M. 6. F. P. 8 A. 1. F L. O. 8 P. G. 4. 12. 6 M. K. 18. C. 3. H. 10. 16. (b) A. 2. 18. L. 4. (a). B. 9. G. 2. 12. 16. C. B. 10. 27 H. 1. K. 27. 15. A. 7. N. I. 5. N. 9. D. 1 5 I. E. 15. B. 7 J. F. 25 L. 27. O. 8 P. 16. G. 4. 12. 6 M. K. 18. C. 2 10. 3. H. N. 9. D. 5 I. 15. E. 7 J. 25. 27. 16. (c). (d). Figure 4.1: An example of setting routing path: (a) Address allocation tree; (b) Mesh topology; (c) Routing paths while applying stateless routing protocol; (d) Routing paths while apply stateful routing protocol. E.g.,. seqGCD(16; 12) = 2 2 = 4. * pfSeq (16) = (2; 2; 2; 2) pfSeq (12) = (2; 2; 3) the longest common prefix of pfSeq. (16) and pfSeq(12) is (2; 2):. seqGCD(18; 15) = 1. * pfSeq (18) = (2; 3; 3) pfSeq (15) = (3; 5) the longest common prefix of pfSeq. (18) and pfSeq(5) is ;:. 4.2.1 Stateless prime-based self-configured routing protocol In the stateless routing protocol, all nodes can configure routing paths without either exchanging routing information or recording any information. We provide two alternatives: source. 27.

(42) 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17. x=seqGCD(src, dst) // setting up the routing path from src to x while(src!=x){ parent = src/(the largest prime factor of src) add parent into the routing path src=parent } // setting up the routing path from x to dst int primeFactors[] primeFactors = pfSeq(dst/x) for(int k=0; k< number of elements in primeFactors; k++){ child =x for(int i=0; i<k; i++){ child = child*primeFactor[i] } add child into the routing path }. Figure 4.2: The procedure of setting routing path in the source routing method of stateless routing protocol. routing method and hop-by-hop routing method. If source routing method is applied, the source node is responsible for setting the whole routing path, otherwise each node finds the next hop by itself to send packets. We take Figure 4.1 for example. Figure 4.1 (a) is the address allocation tree of a mesh network in Figure 4.1 (b). In the stateless routing protocol, nodes purely proceed along branches of the address allocation tree to obtain the routing path. Figure 4.1 (c) illustrates two routing paths from node O to node P and from node M to node I. We can observe that the routing path can be divided into two segments: from the source node to the least common ancestor and then to the destination node. We will describe the detail procedures of two alternatives of stateless routing algorithms in the subsections below. Source routing method Figure 4.2 is the procedure of how a source node finds the whole routing path to the destination node, where sr is the address of the source node and dst is the address of the destination node. The source node first finds the least common ancestor by computing seqGCD. (sr ; dst). Then. it builds the first segment of routing path by recursively adding parent nodes until reaching the common ancestor as shown in the line 3 to 7 in Figure 4.2. For the second segment of the routing path, the source node needs to choose right child nodes. We know that the position of each node in the address allocation tree is determined by its address. Assuming. pfSeq (dst) = (p0 ; p1 ; p2 ; :::; pm ), we can guarantee that the address of the common ancestor. 28.

(43) 1 2 3 4 5 6 7 8. x=seqGCD(cur, dst); if(x<cur){ parent = cur/(the largest prime factor of cur) send packet to parent }else if(x==cur){ child=x*(the smallest prime factor of (dst/x) ) send packet to child }. Figure 4.3: The procedure of setting routing path in the hop-by-hop routing method of stateless routing protocol.. Qi=j p. i=0 i for some 0. j m, and addresses of nodes along the path from the common Q Q Q ancestor to the destination are ( i=j +1 p , i=j +2 p , ..., i=m p ). Therefore, the source node is. i=0. i. i=0. i. i=0 i. can recursively add these nodes into the routing path as shown in the line 8 to 17 in Figure 4.2. Taking Figure 4.1 for instance, node M(18) wants to set the routing path to node I(15). Node M(18) first computes seqGCD. (18; 15) and is aware that their common ancestor is node. A(1). It starts to recursively add addresses of parent nodes, 6, 2, and 1, into its routing path. While reaching the common ancestor, node M(18) factorizes the quotient of obtains pfSeq. 15 by 1 and. (15=1) = (3; 5). It then recursively adds 1 3 and 1 3 5 into the routing path.. Consequently, it gets the whole routing path (6, 2, 1, 3, 15) to node I(15). Hop-by-hop routing Besides source routing method, we also propose another alternative, hop-by-hop routing, to enable each node to compute the next hop by itself. Although this method might increase the computation load of each node, it can decrease the header overhead introduced by source routing. Figure 4.3 is the pseudo code of how a current node with address ur finds the next hop for a destination having address dst. First of all, the current node also computes seqGCD. ( ur; dst) to find the least common. ancestor. If the address of the least common ancestor is smaller than its own address, the current node knows it is in the first segment of routing path and sends packet to its parent. On the contrary, if the address of the least common ancestor is its own address, the current node is in the second segment and sends the packet to the correct child node. The address of the child node can be computed by multiplying ur with the smallest prime factor of the quotient of the address of destination by the address of itself.. 29.

(44) 1 2 3 4 5 6 7 8 9 10. selfGcdValue=seqGCD(cur, dst); for (int i=0; i<number of neighbors; i++){ neighborGcdValue[i]=seqGCD(address of neighbor[i], dst); } gNeighborGcdValue = the largest neighborGcdValue; If(gNeighborGcdValue>=selfGcdValue){ send packets to the node with gNeighborGcdValue; }else if(gNeighborGcdValue==selfGcdValue){ send packets to the parent node; }. Figure 4.4: The procedure of setting routing path in the stateful routing protocol. We here also take Figure 4.1 for instance to demonstrate how node O(12) sends packets to node P(16). Node O(12) first computes seqGCD. (12; 16) and is aware that 4 is the address of. the common ancestor. Since 4 is smaller than its own address, it sends packets to its parent. (4; 16) and gets the answer as 4, which is equal to its own address. Therefore, the node F computes the smallest prime factor of (16=4 = 4), node F(4). Node F(4) also computes seqGCD. which is 2, and multiplies this value with its own address 4 to gets the address (2*4=8) of the correct child node. When node L(8) receives the packets from node F(4), it also computes. seqGCD(8; 16) and sends packets to node P(16) according to the same procedure. 4.2.2 Stateful prime-based self-configured routing protocol Although stateless routing protocol can configure routing paths without recording any information, the routing paths unfortunately are usually long because nodes find routing paths only from the branches of the address allocation tree. However, there are still some links, like dotted lines in Figure 4.1, not belong to the address allocation tree. If nodes can also choose these links as part of their routing paths, the length of routing path should be shorter. Therefore, we propose stateful routing protocol to shorten the routing path by making each node recording addresses of its neighbors. In the stateful routing protocol, each node computes the next hop only. When a node with address. ur receives a packet, it follows the procedures in Figure 4.4 to find the next. hop for the destination. dst. First of all, the node computes the seqGCD for itself and all. its neighbors to the destination. If the largest its own. seqGCD value of its neighbors is larger than. seqGCD value, it sets the neighbor with the largest seqGCD value as the next hop. because that neighbor is nearer to the destination. Otherwise, the node sends the packet to its parent node as usual.. 30.

(45) We can observe that in Figure 4.1 (d) the routing path from node M to node I is much shorter if we apply stateful routing protocol. For node M(18), its own seqGCD value to node. (18; 15) = 1, and all neighbors’ seqGCD values are: seqGCD(16; 15) = 1 for node P(16), seqGCD (6; 15) = 1 for node G(6), seqGCD (9; 15) = 3 for node H(9). The. I(15) is seqGCD. largest seqGCD value of neighbors is 3, which is bigger than its own value 1. Therefore, node M(18) sends the packet to the neighbor node H(9). We need to mention that each node does not need to be aware of it is in the first or second segment of routing path as the procedures in the hop-by-hop method. If the destination is its descendant, it will send the packet to its child naturally since its child is also its neighbor and in most cases that child will have the largest seqGCD value as its own address unless there is a link connect to the destination directly. We also take Figure 4.1 (d) for instance. When node F(4) receives a packet destined to node P(16), it computes seqGCD. (4; 16) = 4, and seqGCD. values of its neighbors are 8, 4, 2 of node L, O, and B respectively. We can notice that node P(16) is a descendant of node F(4), so node L(8), a child of node F(4) and an ancestor of node P(16) has the largest seqGCD value. Therefore, node F(4) sets its child node L(8) as next hop naturally for packets destined to node P(16). Consequently, if there is no neighbor having a larger seqGCD value, the node is in the first segment of the routing path and it can just send packets to its parent directly. We have proved in Theorem 3 that the proposed prime-based self-configured routing protocols, including stateless and stateful routing protocols, can guarantee loop-free routing paths. Theorem 3. Prime-based self-configured routing protocol can guarantee loop-free routing path N1 , N2 , N3 , ..., Nk from N1 to Nk . Proof. Case 1: stateless routing protocol: In the stateless routing protocol, all routing paths are configuring based on branches of the address allocation tree. Since there is no loop in the branches of a tree, there is no cyclic routing paths. Case 2: stateful routing protocol: Based on the routing algorithm of stateful prime-based self-configured routing protocol, each node chooses the neighbor with the largest seqGCD of destination as the next hop or sends. 31.

(46) packets to its parent if no neighbor has larger seqGCD than its own seqGCD of destination. Therefore, 8i < j , seqGCD. (Ni; Nk ) seqGCD(Nj ; Nk ).. Assume there exists a loop in the routing path N1 , N2 , N3 , ..., Nk , which means 9Nm. = Nn. where m > n.. . seqGCD(Nn ; Nk ) = seqGCD(Nm ; Nk ) seqGCD(Ni ; Nk ) seqGCD(Nj ; Nk )8j < j ) seqGCD(Nn; Nk ) = seqGCD(Nn+1 ; Nk ) = ::: = seqGCD(Nm ; Nk ). *. According to the routing algorithm, a node sends packets to its parent node only if there is no other neighbors having a larger seqGCD to the destination. If two adjacent nodes, Nn and. Nn+1 for instance, in the routing path have the same seqGCD value to the destination, there is no neighbor of the former one (Nn ) with a larger seqGCD. In such case, the latter one of two adjacent nodes is the parent node of the former one.. ) Nn+1 is the parent of Nn Nn+2 is the parent of Nn+1 ::: Nm is the parent of Nm 1. Since Nm is an ancestor of Nn , Nm. 6= Nn. Therefore, there is no loop in the routing path. N1 , N2 , N3 , ..., Nk from N1 to Nk . 4.2.3 Performance evaluation In this section, we study the performance of proposed addressing and routing protocols. We compare the signal overhead and average length of routing paths of our protocol to those of the OSPF (Open Shortest Path First) [11] protocol. OSPF is a link-state routing protocol, in which each host broadcasts its link status to all other hosts. After gathering information from all hosts, each host runs a shortest path algorithm, Dijkstra algorithm [58] for example, to find the shortest paths to all other hosts.. 32.

(47) (a). (b). Figure 4.5: Two network topologies when network size=5*5: (a) Link=4 and (b) Link=8.. Link 4_s t ateless prime routing Link 4_st at ef ul prime routing Link 4_link state routing. Link 8_s t ateless prime routing Link 8_st at ef ul prime routing Link 8_link state routing. 4000. Signal Overhead. 3500 3000 2500 2000 1500 1000 500 0 3x3. 4x4. 5x5. 6x7 x78 Network size. x89. Figure 4.6: The signal overhead.. 33. x91. 0x10.