ATM網路上以遲滯及漏失為考量之服務品質控制

Abstract

非同步傳輸模式（Asynchronous Transfer Mode - ATM）已經成為在寬 頻-ISDN網路上傳送各種媒體（包括資料、聲音、和影像）的重要技術， 這些媒體各有不同的服務品質（Quality of Service - QoS）要求，因此 提供QoS保證是ATM網路必需具備的能力。為了充分利用網路資源，同時維 持令人滿意的QoS，這篇論文從兩個大方向來探討如何建立一個同時考量 遲滯（delay）和漏失（loss）的QoS控制機構：一是估計式通訊允許控制 （Call Admission Control - CAC）、一是優先權式排序法則（ scheduling discipline）。在第一個方向上，這篇論文首先提出一個類 線性雙類別關聯（Quasi-Linear Dual-class Correlation - QLDC）的估 計式CAC方法，根據各個種類媒體的使用者數目，QLDC可藉由簡單的向量 乘法立刻估算出各個種類媒體的封包遲滯（Cell Delay - CD）和封包漏 失率（Cell Loss Ratio - CLR），這些向量可事先從一個雙類別媒體式 的排隊模式（queueing model）得到。接著這篇論文提出一個類神經網路 的估計式CAC方法，根據各個種類媒體的使用者數目，這個方法可藉由事 先已經訓練過的類神經網路，立刻估算出各個種類媒體的CD和CLR。這兩 種方法都只需要很低的時間和空間複雜度就可以在提供QoS保證的考量下 ，決定是否要允許建立新的通訊。在第二個方向上，這篇論文針對ATM交 換機提出一個提供雙遲滯優先權和雙漏失優先權的排序法則—超前且部份 置換（Precedence with Partial Push-out - PPP）。當儲列未滿時， PPP允許一個遲滯優先權較高的新來封包，最多可以超前L個遲滯優先權較 低的封包，當儲列已滿時，PPP允許一個漏失優先權較高的新來封包，擠 掉位於儲列第R個位置之後的最後一個漏失優先權較低的封包。為了精確 地決定L和R來維持該有的QoS，這篇論文提出一個排隊分析及一個代數分 析，可分析出各個種類媒體的CD和CLR，透過這些分析，L和R將可被動態 及有效的調整，以提供適當的遲滯和漏失給優先權較高的封包，同時讓優 先權較低封包的服務品質之降低程度儘量最小。此外，PPP排序法則和ABR 服務模式都具有"儘量努力" （best-effort）的特性，因此這篇論文也以 ABR和VBR這兩種服務模式為例子，設計了一個PPP的實作架構，來展現PPP 的優勢與可行性。 Asynchronous Transfer Mode (ATM) has been widely accepted as a key technology for supporting all conceivable media including data, voice, and video in Broadband-ISDNs. For transporting such a diverse mix of traffic sources requiring various Quality of Services (QoSs), ATM networks are demanded to offer QoS guarantees for various classes of traffic sources. To fully utilize network resources while retaining satisfactory QoSs for each traffic source in ATM networks, this thesis aims at providing delay-and-loss-based QoS control mechanisms by means of two approaches: estimation-based Call Admission Control (CAC) and priority-based scheduling discipline. In the first approach, the thesis first proposes a CAC algorithm based on a novel estimation method, called Quasi-Linear Dual-class Correlation (QLDC). All heterogeneous traffic sources are initially categorized into various classes. According to the number of calls in each traffic class, QLDC conservatively and precisely estimates the Cell Delay (CD) and Cell Loss Ratio (CLR) for each traffic class in real time via simple vector multiplication. These vectors are computed in advance from the results of dual arrival queueing models. The thesis further presents an efficient neural-network-based CAC (NNCAC) mechanism with heterogeneous arrivals. Based on the number of calls in each class, NNCAC efficiently and accurately estimates the CD and CLR of each class in real time by means of a pre-trained neural network. Both the two estimation-based CACs yield low time and space complexity to make call acceptance decisions offering QoS guarantees. In the second approach, the thesis provides a versatile scheduling discipline, called Precedence with Partial Push-out (PPP), in ATM switches supporting two delay and two loss priorities. By employing a threshold L, the PPP discipline provides delay guarantee by allowing a newly-arriving high- delay-priority cell to precede a maximum of L low-delay-priority cells. Through the use of another threshold R, the discipline offers loss guarantee by permitting a newly-arriving high-loss- priority cell to push out the last low-loss-priority cell located beyond the Rth location in a full queue. By setting L and R properly, PPP versatilely performs as any one of the four widely-accepted disciplines, namely the FCFS, head-of-line, push-out, or head-of-line with push-out disciplines. For precisely determining L and R retaining demanded QoSs, the thesis presents an in-depth queueing analysis for the CD and CLR of high-delay-priority, low-loss-priority cells. The thesis further proposes a simple, algebra-based analysis for the CD and CLR of low-delay-priority, high-loss-priority cells. On the basis of these analyses, L and R can be dynamically and effectively adjusted to provide adequate delay and loss guarantees for high-priority cells while incurring only minimal performance degradation for other classes of cells. Furthermore, PPP is a best-effort discipline and the ABR service category is envisioned as a best-effort service. To justify the viability of the PPP discipline, the thesis then provides a feasible implementation architecture realizing the PPP discipline for ABR and VBR service categories in ATM networks.