內容傳遞網路與內容複製數佳化

Content Delivery Network and Optimal Content Replication

趙圻軒(Chi-Hsuan Chao)

指導教授 ： 張時中

國立臺灣大學/電機資訊學院/電機工程學研究所/碩士(2004年)

https://doi.org/10.6342/NTU.2004.01394

摘要

主從式內容傳遞網路(Client/Server Content Delivery Network)是網際網路裡，目前內容傳遞網路業者(Content Delivery Networker)最普遍採用的網路架構。主從式內容傳遞網路把內容來源伺服器(Origin Server)上的多媒體內容檔案，透過複製分發的方式，預先儲存放置在每一個網路架構中的代理者(Surrogate)裡。當使用者發出內容服務請求之後，透過請求路由系統(Request Routing System)搜尋便可以就近的代理者獲得多媒體內容，減少使用者感受到的延遲時間，也減少了釵h集中於內容來源伺服器(Origin Server)的封包流量，更可提高整體網際網路的運作效能。主從式內容傳遞網路的關鍵議題包括內容管理與網路效能。內容管理的問題之一便是代理者中所儲存多媒體內容複製數，而在網路效能方面，其網路系統架構往往決定了網路整體運作效能表現的優劣。
本論文探討主從式內容傳遞網路架構的改良設計與實現，與代理者中多媒體內容複製數的問題。我們改良了[RFC3466]的架構，以及對每個實體作細部的設計實現。在架構的改良方面我們簡化使用者端的操作步驟，只需向請求路由系統提出內容服務的請求之後，即可由就近的代理者取得該使用者所要求的內容服務。若使用者提出內容服務的請求為冷門的多媒體內容檔案，則在從內容來源伺服器傳遞該檔案到使用者的過程中，我們設計經過分發系統時，在分發系統保留該多媒體內容檔案一份，以利往後使用者取用該多媒體內容。並且進而將比較熱門的多媒體內容檔案，預先存放在各個代理者裡。使用者提出熱門的內容服務請求之後，立即可在鄰近的代理者取用內容服務。如此一來，便可以減少服務使用者取得內容的時間，使網路頻寬資源能被有效地利用，降低整體網路狀態以及內容來源伺服器的負載情形，增加主從式內容傳遞網路的服務效率。
在主從式內容傳遞網路多媒體內容最佳複製數方面，我們把熱門多媒體內容檔案影片排行榜總數的取捨與軟硬體花費管理負擔成本之間的關係，具體描述成為一個凸型最佳化(Convex Optimization)數學模型，並且以數值模擬求解討論內容複製數、代理者總數與使用者所需等待的服務時間(Service Time)三者之間的關係。我們的分析顯示，內容複製數增加時，成本與命中比率亦隨之上升，但內容來源伺服器端的網路負載與服務時間隨之減少，並且代理者總數增加時，成本增加，服務時間有急遽減少的趨勢。若能有效最佳化多媒體內容複製數，使用者可以就近取得所需的內容服務，內容傳遞網路業者花費合理的成本來提高使用者請求內容的回應速度，以便將內容豐富的多媒體內容服務送到用戶端。
本論文根據所改良設計的主從式內容傳遞服務網路系統架構，進行一個具體而微的實驗環境實作，建構一個提供非串流媒體的內容服務簡易的主從式內容傳遞網路，符合基本的內容提供者(Content Provider)期望與使用者需求，其中請求路由系統、分發系統、代理者皆建構在Linux平台上，內容來源伺服器則建構在Window XP平台上，各個元件彼此使用SSH(Secure Shell)安全連線，或藉由FTP(File Transfer Protocol)協定來作資料的傳輸。分發系統、請求路由系統與代理者具有MySQL資料庫，儲存多媒體影片檔案名稱。實驗結果顯示網路系統架構的可行性，並且可以減少使用者感受到的延遲時間

關鍵字

內容複製 ； 內容傳遞網路

並列摘要

Client/Server Content Delivery Network (C/S CDN) is now commonly used by many Internet users. A C/S CDN replicates multimedia contents from an Origin Server to each surrogate scattered over the Internet. A request from a client for content is directed by a Request Routing System to a surrogate, which finds and delivers the requested content, and is located close to the client to reduce the client’s waiting time. Furthermore, the performance of content delivery improves as the traffic congestion at the Origin Server decreases. Issues of C/S CDN include content management and network performance. One aspect of content management is the design of the optimal number of “multimedia content replication” among surrogates. The architecture of network is crucial to the network performance.
In this thesis, we first improve and design in detail the infrastructure of C/S CDN. We then focus on determining the optimal number of multimedia content replication. We improve the infrastructure of CDN in and design delicately each entity in the architecture. We simplify operating steps of clients. Clients can then get the requested content quickly. Unpopular multimedia content, which is does not have a copy at surrogates, is delivered from Origin Server to clients and is replicated in the Distribution System once a client requested for it. This procedure is advantageous to subsequent clients in getting multimedia content. In addition, we design that C/S CDN replicates popular multimedia content among surrogates so that clients can get the requested content immediately from surrogate close to clients after sending popular content requests. In this approach, the time for clients to get multimedia content decreases, and the loads of both internet and Origin Server are reduced so that internet bandwidth can be used effectively, and service efficiency increases.
We present a model describing the tradeoff between delay and storage costs with respect to the number of replicating popular content in C/S CDN multimedia content replication problem. This problem is formulated as a convex optimization problem. Then, we solve the problem by numerical simulation and exhaustive search. Results are used to investigate the relationship between the number of replicating popular content, surrogates and service time for clients. In our analysis, we found that total cost and the ratio that a request gets the content from the surrogates increase as the number of replicating popular content become larger. At the same time, the load of Origin Server and Service Time decrease. With the increase in number of Surrogates, the total cost will increase, and service time will tend to decrease rapidly. If multimedia content replication is optimized, CDN service providers can make it easier and faster for clients to access the content without significant cost.
We have constructed a simple C/S CDN and perform an experimental implementation design according to the architecture, which we improved and designed delicately. This simple C/S CDN, which provides non-streaming content, satisfies clients’ demand and content provides’ expectation. Request Routing System, Distribution System and Surrogates are implemented with Linux operation system. On the other hand, Origin Server is implemented with Window XP operation system. Every entity connects with each other by Secure Shell (SSH) or transfer data by File Transfer Protocol (FTP). Request Routing System, Distribution System and Surrogates have MySQL databases which store multimedia files’ name. Experimental result shows that this network architecture can work and reduce waiting time for clients

並列關鍵字

contetn delivery network ； content replication

參考文獻

[Bou96] Thomas Boutell, CGI Programming in C and Perl, Addison Wesley Press, 1996.

[CaC97] Robert L. Carter and Mark E. Crovella,“Server Selection Using Dynamic Path Characterization in Wide-Area Network”, Proceedings of the INFOCOM ’97 Sixteenth Annual Joint Conference of the IEEE Computer and Communications Societies, 1997.

[CLW02] C. W. Cameron, S. H. Low, D. X. Wei, “High-Density Model for Server Allocation and Placement” Proceedings of ACMSigmetrics, Marina del Rey, CA, June 2002.

[Erc02] O. Ercetin, “Market-Based Resource Allocation for Content Delivery in the Internet,” Ph.D. Dissertation, Dept. of Electrical and Computer Engineering, the University of Maryland, July 2002.

[GMS00] R. Gibbens, R. Mason, R.Steinberg, “Inter Service Classes Under Competition”, IEEE Journal on Selected Areas in Communications, Vol 18, December 2000.