2 Additional Related Work

The recent work on content distribution can be largely divided into three categories: (a) infrastructure based content distribution, (b) overlay network based distribution [7,9,12,13,17,10,6,14], and (c) peer-to-peer content distribution [11,16,1,24]. Our work is directly related to the infrastructure based content distribution network (CDN) (e.g. Akamai), which employs a dedicated set of machines to reliably and efficiently distribute content to clients on behalf of the server. While the entire collection of nodes in a CDN setting may be varying, we assume that the set of currently active nodes is known. The sites supported by multiple mirror servers are referred to the same category. Existing research on CDNs and server replication has primarily focused on either techniques for efficient redirection of user requests to appropriate servers or content/server placement strategies for reducing the latency of end-users. A more recent idea is to access multiple servers in parallel to reduce downloading time or to achieve fault tolerance. Several research papers in this direction exploited the benefits of path diversity between the clients and the site's servers with replicated content. Authors in [20], demonstrate the improved response time observed by the client for a large file download through the dynamic parallel access schema to replicated content at mirror servers. Digital Fountain [4] applies Tornado codes to achieve a reliable data download. Their subsequent work [5] reduces the download times by having a client receive a Tornado encoded file from multiple mirror servers. The target application of their approach is bulk data transfer. While CDNs were originally intended for static web content, they have been applied for delivery of streaming media as well. Delivering streaming media over the Internet is challenging due to a number of factors such as high bit rates, delay and loss sensitivity. Most of the current work in this direction concentrates on how to improve the media delivery from the edge servers (or mirror servers) to the end clients. In order to improve streaming media quality, the latest work in this direction [3,15] proposes streaming video from multiple edge servers (or mirror sites), and in particular, by combining the benefits of multiple description coding (MDC) [2] with Internet path diversity. MDC codes a media stream into multiple complementary descriptions. These descriptions have the property that if either description is received it can be used to decode the baseline quality video, and multiple descriptions can be used to decode improved quality video. One of the basic assumptions in the research papers referred to above is that the original content is already replicated across the edge (mirror) servers. The goal of our paper is to address the content distribution within this infrastructure (and not to the clients of this infrastructure). In this work, we propose a method to efficiently replicate the content (represented by large files) from a single source to a large number of servers in a scalable and reliable way. We exploit ideas of partitioning the original file and using diverse Internet paths between the recipient nodes to speedup the distribution of an original large file over Internet. In the paper, we partition the original file in $n$ equal subsequent subfiles and apply FastReplica to replicate them. This part of the algorithm can be modified accordingly to the nature of the file. For example, for a media file encoded with MDC, different descriptions can be treated as subfiles, and FastReplica can be applied to replicate them. Taking into account the nature of MDC (i.e. that either description received by the recipient node can be used to decode the baseline quality video), the part of the FastReplica algorithm dealing with nodes failure can be simplified.