寬頻網際網路服務品質保證(III)─子計畫三:寬頻網際網路中以量測為基礎的訊務控制與動態資源分配之研究與實作

Abstract

目前網際網路所提供的「盡力服 務」，由於無法保證服務品質，所以並不 能給予即時應用程式有效的支援。為了解 決這問題，目前已經發展出一些即時媒體 應用能夠視封包遭遇到的延遲而動態地 調整資料的傳送。這些應用並不需要絕對 可靠的封包延遲上限，但必須盡可能避免 封包在傳輸過程中被丟棄。本研究即針對 這類應用程式提供『無壅塞服務』，我們 將其定義為『保證聚合資料流的最大封包 遺失率為ε』，希望將來可以應用於IETF 整合服務架構下的負載控制服務。本研究 首先對台大校園骨幹網路做實際流量的 量測及趨勢分析，以瞭解網路資料流的特 性。初步結果顯示網路負載具有近似常態 分配的特性，可進一步用於網路流量的預 測以及網路資源的規劃與控制。此外，我 們針對『無壅塞服務』的提供，提出一個 動態頻寬與緩衝區分配的方法。實驗結果 也顯示避免頻寬使用率太高及動態分配 頻寬可以有效抑制壅塞的發生。為了進行 更精確的量測以及效能評估，我們發展了 一套量測工具。藉著這套工具的使用，我 們可以獲得網路中流動封包對應到每一 層通訊協定的相關資料；量測的基本單位 可以細分至每個資料流。不但讓預測更精 準，也便於off-line 效能分析。

Many real-time applications are capable of adapting their transmission to the network state and can as well tolerate occasional delay bound violations in the presence of transient network congestion. For this type of applications, an absolutely reliable bound on packet delivery times is not required. Moreover, many non-real-time applications such as on-line transaction processing and distributed simulation would desire a congestion-free packet delivery service from the network. This new type of service is called “Congestion-Free Service” with the guarantee of a maximum packet loss rate. This work considers resource allocation in the support of “Congestion- Free Service” in Broadband Internet. First, we studied traffic characterization under different measurement models via analyzing traffic traces collected from National Taiwan University campus network. The preliminary results show that traffic load can be approximated by the Normal distribution. In the second part of the work, we proposed a dynamic bandwidth and queue management scheme to support “Congestion-Free Service.” To better understand the characteristics of the broadband Internet, we have developed a tool called AppMeasure which provides three sets of service: IPFlow, AnyTrace, and AppFlow. IPFlow provides on-line tracking and reporting of individual IP flows. In AnyTrace, user can specify the number of bytes of a packet to trace. It serves as a measurement base to enable all-layer traffic accounting and analysis, and differentiated charging based on Quality of Service. Appflow tracks and reports any type of traffic specified by the user . It aims to integrate with policy-based QoS service.