Reliable Distribution of Data Using Replicated Web Servers

In this paper we describe an algorithm for accessing files over replicated Web servers. The algorithm breaks the request for a document W, which is assumed to be replicated among N different servers, into N requests such that any K replies are sufficient to reconstruct the whole page. In this way, the document is downloaded from the K fastest servers. The algorithm is feedback-free and does not require any computation to be performed on client or server sides. Thus, it may be implemented on devices with limited computing capabilities. Analytical evaluation of the proposed algorithm shows that, under the model's assumptions, a correctly tuned value for K significantly improves the probability of completing earlier the transfer of W