標題: 呼叫層級及晶格層級爆發性下非同步傳輸網路中速率式流量控制的模擬研究
A Simulation Study of ATM Rate-based Flow Control with Call- level and Cell-level Burstiness
作者: 吳中雄
Wu, Chung-Shong
林盈達
Ying-Dar Lin
資訊科學與工程研究所
關鍵字: 非同步傳輸網路;速率式流量控制;回饋控制;爆發性;呼叫層級的爆發性;晶格層級的爆發性;ATM;rate-based flow control;feedback control;burstiness;call-level burstiness;cell-level burstiness
公開日期: 1995
摘要: 非同步傳輸網路的出現為那些需要較多頻寬、較低異常性以及有限度晶格 流失和延遲的應用程式帶來解決之道。除此之外,它亦提供了區域網路模 擬及非同步傳輸網路上的網際網路協定,使得現存之應用程式也可以在非 同步傳輸網路上執行。這些應用程式使用非同步傳輸網路所提供可獲得的 位元速率服務而被交付於流量的控制。為了避免造成非同步傳輸網路的效 能劣於現今較慢的網路,我們應該謹慎地檢視這些具爆發性的資料源所帶 來的影響。 在這篇論文中,我們發展出一個叫作ERMC (Explicit Rate based on Multipl Condition)的機制來模擬呼叫層級及晶格層級爆 發性下下的速率式流量控制。在此機制中,根據回饋 節點的壅塞狀態將 會調整資料源的晶格傳輸速率,而不是晶格產生速率。利用此一資料流 及網路的模式,我們可以得到晶格流失機率及等待時間的平均值。 從這些模擬結果中,我們發現在晶格層級爆發性下的晶格流失機率會比在 呼叫層級爆發性下來的敏感。而且我們也發覺用固定來源個數和機率所形 成的函式來作變動來源個數的逼近並不是一個理想的方法。有關控制機制 方面,多重的暫存區門檻條件將會比單一的暫存區門檻條件有較好的效率 。除此之外,流量控制比較適合於區域網路,因此,廣域網路上的流量控 制也將會是個值得研究的主題。 The appearance of ATM(Asynchronous Transfer Mode) networks brings the possible solution for those applications demanding more bandwidth, low jitters, and limited cell loss and delay. It also supports LAN emulation and IP-over-ATM for the existing applications to run on ATM networks. These applications use ATM ABR (Available Bit Rate) services and are subject to flow control. Effects of flow controlling these bursty traffic sources should be carefully examined to avoid inferior performance in ATM than in slower networks. In this work, we simulate a rate-based flow control scheme called ERMC (Explicit Rate based on Multiple Conditions) under burstiness at call and cell levels. Rather than fixing the number of sources, the number of sources is variable. In addition, the cell transmission rate, instead of arrival rate, of a source is adjusted based on the congestion state of the feedback point. Moreover, the control mechanism is based on ER-type feedback. Using the traffic and network models, we obtain the average cell loss probability and average waiting time. With these results, we find that the cell loss probability under burstiness at the cell-level is more sensitive than under burstiness at the call-level. Also, the performance measures of using the function, based on fixed number of sources and probability, is not a good approximation to the performance of variable number of sources. About the control mechanism, the multiple-threshold mechanism has better performance than single- threshold mechanism. Moreover, the flow control is suitable on LAN rather than on WAN. Hence, the flow control on WAN environment would also be an interesting topic.
URI: http://140.113.39.130/cdrfb3/record/nctu/#NT840394008
http://hdl.handle.net/11536/60449
顯示於類別:畢業論文