Abstract
在本篇論文中我們提出一個設定大型網路傳遞路徑的新方法。這個新方法能夠處理大型網路各組成單元故障率之相互關係。首先整個網路分成許多階層及區域，同一區域內之點並不需要是相連的。每一個區域建立一個虛擬區域網路。再利用一個新的類神經網路模組求出每一虛擬區域網路中二點間的傳遞路徑。這個新的類神經網路模組能夠處理網路組成單元故障率之相互影響。然後就可以同時建立每個區域之多階層最短路徑表。最後由一個新的通訊協定依據多階層最短路徑表上的資料求得傳輸路徑。模擬結果顯示：此方法建立傳輸路徑並不需要花費很多的記憶体來儲存網路資訊。所提出的類神經網路模組在本篇論文中證明穩定解存在。並且計算出此架構設計參數的上下限，這對此架構在實際運作時的參數選擇非常有幫助。另外我們也分析本方法在一階層及多階層分區域網路上的效益。結果顯示：記錄最短路徑表所需的記憶体變小了，所需的儲存空間為Baratz &Jaffe所提出方法的三分之一。網路可任意切割；因為沒有限制同一區域內之所有通訊點都連接在一起。同一層的虛擬區域網路與最短路徑表可同時建立；因為每一通訊點只屬於一個區域。本方法於多階分區域網路架構下所需的儲存空間決定於階層、區域與邊界點等三個數目。削去法則將所需 REQUEST 訊息的數目由正比於邊界點數目的平方降至一次方。A new routing algorithm is proposed in this dissertation forlarge communication networks with dependent unreliablecomponents. The entire network is first partitioned into amulti level environment without assuming that all nodes in thesame cluster be directly connected. An auxiliary networkassociated with each cluster is then constructed. Based on theauxiliary networks, a new neural network model capable ofhandling dependent component failures of communicationnetworks, is used to calculate the shortest path for eachorigin-destination pair. Therefore, hierarchical shortest pathtables can be established in parallel for each cluster of thesame level from bottom up. Finally, by the hierarchicalshortest path table, a new routing protocol finds the resultanttransmitting path. Simulation results indicate that the routingpath can be found not at the expense of the table size and thestorage space for each node. The proposed neural network modelis proven to have a stable solution. Moreover, useful upper andlower bounds for the design parameters are derived tofacilitate their selection during implementations. Thealgorithm's performance for one-level and multi-level clusterednetworks are analyzed. Results indicate that the shortest pathtable size can be reduced and the storage space saving for eachindividual node is approximately one-third of that proposed byBaratz and Jaffe.