由於傳統無線網路架構受到間歇性的傳輸連結、有限的傳輸距離以及低節點密度等因素，會使得行動及無線網路應用間的通訊中斷，造成無法確保端點至端點存在連線。因此，耐延遲網路技術被用來解決上述的訊息傳遞問題。耐延遲網路不同於傳統電腦網路和行動隨意網路，其訊息傳遞至目的端是採用「儲存、攜帶和轉送」的方法，只有當節點與其他節點相遇時，才會傳遞訊息。在耐延遲網路的路由協定方法中，透過部署靜態中繼節點(又稱暫存盒)的策略被認為可以有效地增加訊息的傳遞成功率或減少訊息的傳遞成本。然而，目前在這類於耐延遲網路使用暫存盒的研究中，只有少數的研究提供完整的分析模型和路由方法。 本研究提出的演算法將有效的減少的訊息傳遞成本於節點是異質性的耐延遲網路。本論文的內容可分為下列三部份：首先，本研究認為暫存盒的部署能幫助耐延遲網路的訊息傳遞，並提出一個基於效用值的演算法來決策兩節點相遇時，訊息該如何傳遞。接著，本研究介紹如何用Absorbing馬可夫鏈設計一個分析架構以量化和預測本論文提出的演算法之效能。特別地，這個架構能夠將最佳化的問題轉化成馬卡夫鍵問題，當兩個節點相遇時，即可決定轉移機率，進一步解出最佳解。最後，本研究透過模擬來衡量所提出的演算法在訊息傳遞成功率和訊息傳遞成本之效能。結果顯示，本研究提出的方法相較幾個有名的演算法有著較好的效能;Many data and network applications for mobile and wireless networks cause disruptions in communications where traditional Internet architectures fail to ensure end-to-end connections due to intermittent connectivity, limited transmission range and low nodal density. The delay-tolerant networking (DTN) technology is introduced to deal with these problems. Contrary to traditional computer networks and MANETs, the message forwarding in delay-tolerant networks adopts the store-carry-and-forward approach to send messages to destinations when nodes have any opportunistic contacts with others in a network. Among the approaches for routing protocols in DTNs, the strategy of deploying the stationary relay nodes called “thrown boxes” is considered for increasing message delivery probability and reducing the message overhead efficiently. Till now, few studies have provided comprehensive analysis models and routing solutions for thrown-box-based network paradigms in the DTN research field.
This thesis addresses the efficiency on message forwarding over heterogeneous nodes in DTNs. The technical content of this thesis includes three parts. Firstly, this study considers the deployment of thrown boxes that can assist in message delivery in delay-tolerant networks. A utility-based algorithm is applied to make the message forwarding decision between two nodes whenever they are in contact. Secondly, this study introduces the adaptation of absorbing Markov chain to develop an analytical framework to quantify and predict the performance of the proposed algorithm. Specifically, this framework is able to map an optimization problem over the presence of thrown box into a Markov chain over the relevant solution space. The contact between two nodes can dictate the chain transition probability and be used to advance the finding of an optimal solution by following neighboring solutions in chain. Finally, this study runs simulations to evaluate the proposed scheme in terms of the message overhead ratio and message delivery probability. Performance results show that the proposed algorithm performs better than other popular algorithms.
