| 摘要: | 行動機會網路(Mobile opportunistic network; MON) 是一種新型態的行動及無線網路，網路之內的訊息傳送可藉由節點移動與相遇的過程來轉遞訊息，以此方式逐步將訊息送達目的端節點。在行動機會網路的環境下，網路拓撲是動態且破碎的，並且難以保證任一對訊息來源端至目的端之間擁有持續性的端點到端點(end-to-end)路由路徑。因此，為了傳遞訊息，行動機會網路訴諸於使用耐延遲網路(Delay tolerant network; DTN)架構下的store-carry-and-forward傳遞模式。在這樣不穩定的環境下，訊息傳遞經常採用訊息複製的方式，來提升目的端收到訊息的機會。然而反覆地複製訊息也會消耗大量的網路資源，例如有限的儲存空間以及間斷的傳輸頻寬等。為了減少這類因複製訊息所導致的資源消耗，我們無疑地需要有效的 (1) 中繼節點選擇、(2) 訊息傳輸排程、以及 (3) 儲存空間管理機制。因此，本論文致力於在行動機會網路下儲存空間管理機制及訊息傳輸排程之研究，具體提出一組路由方法及策略，共包括：(1) Enhanced Buffer Management for Message Multicasting; (2) Probabilistic Routing with Contact Periodicity and Regularity; (3) On Exploiting Temporal Periodicity for Message Delivery。
雖然過往的研究曾提出許多儲存空間管理機制及訊息傳輸排程的研究，但是這些研究貢獻主要是基於訊息只有單一目的端的單播環境。如果考慮到訊息需要傳送到多個目的端的群播環境，那些研究所提出的作法由於未考慮到群播訊息傳遞之問題及特性，因而造成他們方法的應用及效能受到限制。因此，本論文研究的第一部份，提出了一個在行動機會網路群播環境下，有效的儲存空間管理機制及訊息傳輸排程方法(E-GBSD)。這個方法首先延伸了先前研究Optimal Knowledge-based Scheduling and Drop Policy，藉以得到一個效用函數來決定訊息傳遞及刪除的排序，使得以最大化群播訊息的傳遞成功率。模擬結果顯示，在訊息群播下，E-GBSD比起其它傳統典型的儲存空間管理機制及訊息傳輸排程機制，更能在群播方面展現出效能的優勢。
針對行動機會網路下訊息複製方法之效能改善，本論文之研究進一步結合機率性路由方法，用以提高訊息傳遞成功率。有鑑於過往的機率性路由方法，其路由決策值delivery predictability的計算主要依靠節點之間的相遇頻率。然而，我們研究發現delivery predictability在一個節點相遇頻率較低的時間區間內會變的相當不可靠。此外，機率性路由配合傳統的如先進先出(First-in, First-out)佇列方法，也會導致Looping Problem使得傳輸成本變得異常高。因此，本論文研究的第二個部份，提出了一個基於相遇周期性及規律性的機率性路由方法(PRCPR)。這個方法考慮當節點處於一個相遇頻率較低的時間區間內時，不使用原來的delivery predictability模式而是改採用剩餘的時間區間來決定訊息路由。因此，這個方法能夠有效地判斷節點此刻是否需要傳送訊息給相遇節點。同時，這個研究也提出一個輕量級的儲存空間管理機制來應對Looping Problem。模擬結果顯示，在訊息流量擁塞的情況下，PRCPR比起傳統的機率性路由，能在傳遞成功率上增加10%，同時還能維持在適度的傳輸成本。
再者，近期的文獻經由分析真實的移動軌跡檔之後，發現人類的移動具有周期性。而行動機會網路下的訊息路由正因為依賴於節點移動特性，節點周期性即變成一個用來強化路由方法的重要因子。因此，為了利用周期性及節點的相遇特性，本論文研究的第三個部份，則提出了一套基於相遇周期性和相遇相似度的訊息路由機制(TPMD)。這個機制首先識別出節點相遇在時間上的周期性，進而提出一個量化周期的方法。模擬結果顯示，在移動具有周期性的環境下，TPMD比起傳統知名的PRoPHETv2路由方法，能夠獲得更高傳遞成功率及較低的傳輸成本。
本論文研究之整合成果能夠促進MON的訊息傳遞以及管理機制之效率。而我們相信，這些在MON架構環境下所研發出來的成果，未來將能夠應用於即將到來的新興技術如：Device-to-Device Communications、Vehicle-to-Everything Networking、Edge Computing、Information-Centric Networking等。;The mobile opportunistic network (MON) is an emerging thread of mobile and wireless networks. In MONs, message distribution takes advantage of node mobility and opportunistic encounters with other nodes over infrastructure-less networks. Network topologies in MONs are fragmented dynamically, and persistent end-to-end paths between any pairs of source and destination nodes cannot be guaranteed. Hence, many studies resort to store-carry-and-forward transfer model originating from the delay/disruption tolerant networks (DTNs) architecture to deliver messages in MONs. In this environment, message delivery often applies the message replication methodology to contend with network dynamics, thus increasing the opportunity that destinations can receive message replicas of an original message from a source node in a network. However, repeatedly replicating messages among mobile nodes consumes significant resources like local buffer space on nodes and limited bandwidth during inter-node communications. To reduce resource cost accompanied with repeatedly replicating messages in a MON system, (1) relay selection, (2) transmission scheduling, and (3) buffer management are three crucial research dimensions that must be resolved certainly. To this purpose, the study in this dissertation addresses chiefly on buffer management, message scheduling, and routing aspects, and results in several contributions for message delivery in MONs: technically, (1) Enhanced Buffer Management for Message Multicasting, (2) Probabilistic Routing with Contact Periodicity and Regularity, and (3) On Exploiting Temporal Periodicity for Message Delivery.
Although previous studies proposed various buffer management and scheduling policies, their efforts mainly contributed to a simple scenario of message unicasting from a source to a singular destination in a network. When message multicasting towards multiple destinations is considered, previous solutions performed inefficiently since the essence of their efforts were not optimized for message multicasting with different performance measures. Therefore, the first part in this dissertation proposes an efficient buffer management and scheduling scheme, called E-GBSD, on the base of a generic replication-based routing model for message multicasting in MONs. The proposed design elegantly extends an optimal knowledge-based scheduling and drop policy, and derives a new utility function to prioritize messages in a buffer for maximizing the successful delivery rate in a network. Simulation results manifest that E-GBSD outperforms not only the original policy but also several buffer management policies under message multicasting in MONs.
Probabilistic routing with message replication is recognized as an important methodology that can attain high performance of successful message delivery. Prior probabilistic routing studies mainly depended on delivery predictability with contact frequencies among nodes. However, our study finds that the accuracy of delivery predictability is sensitive and unreliable during time periods of low contact frequency. Conventional probabilistic routing schemes with flat queuing policies, like First-in and First-out, may result in a looping problem that will cause nonnegligible transmission overhead. Hence, the second part in this dissertation presents the probabilistic routing based on contact periodicity and regularity (PRCPR) to maintain the performance of probabilistic routing in MONs. This scheme considers the residual duration of a current period instead of the original delivery predictability when nodes undergo some periods of low contact frequency. This scheme can thus enable nodes to accurately determine whether or not to hand over messages when two nodes encounter during movement. Besides, this study also proposes a lightweight buffer management policy to cope with the message looping problem. Simulation results manifest that the proposed scheme obtains better delivery performance that could be enhanced at most 10% with moderate transmission overhead than conventional probabilistic routing scheme when message traffic is congested.
Recently, some studies argued that human mobility involves some periodicity as a result of various realistic mobility traces and analyses. Whereas routing in MONs is sensitive to node mobility and network dynamics, periodicity has become an important factor for routing design in MONs. Hence, to exploit the properties of periodicity and contact relationship among nodes, the third part in this dissertation proposes an efficient message forwarding scheme, named Temporal Periodicity for Message Delivery (TPMD), which is based on the contact periodicity and contact similarity. This study effort identifies the characteristics of contact periodicity among nodes in a temporal scale, and formulates a period quantification procedure. Referring to temporal periodicity in MONs, we conduct extensive simulation to examine TPMD and show its efficiency on improving successful delivery rate and lowering message overhead as compared with the famous PRoPHETv2 scheme.
Therefore, the contribution of this dissertation can promote the efficiency of message delivery and management in MONs. We believe that these efforts to MONs can be integrated to the emerging network technologies such as Device-to-Device Communications, Vehicle-to-Everything Networking, Edge Computing, and Information-Centric Networking in the coming future. |
