結合訊框與優先權佇列之即時訊務排程方式的設計與分析

Abstract

隨著網際網路的興起，各種軟體的應用與開發，使網路的發展朝向寬頻化 與智慧化。而在現今多媒體網路中許多訊務都有即時性的需求，如網路交 易、電傳視訊、 隨點視訊播放等。所以未來的網路設計除了要在硬體上 提升路由器或交換機的速度外，如何處理即時性的訊務及有效的管理系統 資源將是網路設計者所必須解決的問題。 ATM 交換技術因具有保證服務 品質的優點， 使其被廣泛的應用於即時通訊網路。而為了在ATM網路上提 供即時服務品質的保證，必須加入即時訊務排程的控制模組，才能有效的 管理網路上多樣性的訊務。在本篇論文中我們發展出一種以訊框為基礎的 排程方式-FBPQ (Frame-based priority queue)。它結合了訊框訊流控 制的方法和優先權佇列的觀念來提升系統的排程效率。在實作方面，我們 分別以固定優先權和動態優先權的排程方式來實現 FBPQ 。FSPQ 可比 (Stop and Go, TCRM) 等以訊框為基礎的排程方式容納更多的即時通訊連 線。而 FDPQ 則以訊框的方法控制連線的輸入速率，並以Earlist- deadline-first的服務方式來達成使用者的服務品質要求。這兩種排程方 法都非常的簡單，不僅能有效的排程即時訊務， 更能保留較多的頻寬服 務非即時通訊連線且使用較少的緩衝記憶體空間，非常適用於高速網路的 硬體實現。 With the growing use of the Internet, the network evolution trends to be wide band and intelligent. In the existing multimedia network, many type of servicerequires real-time delivery, such as EDI, teleconference and video on demand.So, besides upgrading the hardware processing speed, the future network must provide real-time communication and efficient resource management. The ATM network was originally developed for use in the telecommunication network, but dueto it's capability of providng QoS guarantee, it has been used for real- time communication in data communication networks. In order to provide real-time QoS guarantee in ATM network, an additional scheduling module must be employed to handle the diverse types of traffic. In this thesis, we develop the frame-based priority queue(FBPQ) which combines the framing strategy and priority queueing concept to develop an efficient cell scheduling scheme for real-time QoS requirements. Considering implementation complexity, we use both the static and dynamic priority scheduling to implement the FBPQ. We show that FSPQ(frame static priority queue) can enlarge the admission region of the original frame based scheme(S&G, TCRM) and with proper buffering scheme, it may outperform the rate-monotonic scheme. FDPQ(frame dynamic priority queue) adopt the framing mechanism as a rate controller and use earlist-deadline-first service schedulingdiscipline to achieve the user's QoS reqirments. Both of our proposed schemesare efficient and simple, and thus are very at tractive for hardware implementation in high-speed network.