Patching has been shown to be cost efficient for video-on-demand systems. Unlike conventional multicast, patching is a dynamic multicast scheme which enables a new request to join an ongoing multicast. Since a multicast can now grow dynamically to serve new users, this approach is more efficient than traditional multicast. In addition, since a new request can be serviced immediately without having to wait for the next multicast, true video-on-demand can be achieved. In this paper, we introduce the notion of patching window, and present a generalized patching method. We show that existing schemes are special cases with a specific patching window size. We derive a mathematical formula to help determine the optimal size for the patching window. This formula allows us to design the best patching scheme given a workload. The proposed technique is validated using simulations. They show that the analytical results are very accurate. We also provide performance results to demonstrate that the optimal...

Abstract—In an on-demand video system, the video repository generally has limited streaming capacities and may be far from the users. In order to achieve higher user capacity and lower network transmission cost, distributed servers architecture can be used, in which multiple local servers are placed close to user pools and, according to their local demands, dynamically cache the contents streamed from the repository. We study in this paper a number of caching schemes as applied in the local servers depending on whether the repository is able to multicast movie contents to the local servers or not, and whether the local servers can exchange their cached contents among themselves or not. Our caching schemes keep a circular buffer of data for the movie requested, and hence movies are partially cached. By adjusting the size of the buffer, such caching is able to achieve better tradeoff between network channels and local storage as compared to the traditional caching in which a movie is treated as an entity. For each caching scheme, we study the tradeoff between the local storage and the network channels, and address how the total cost of the system can be minimized by appropriately sizing the buffer. As compared to a number of traditional operations (request batching and multicasting, true-VOD, etc.), we show that distributed servers architecture is able to achieve much lower system cost to offer on-demand video services. Index Terms—Caching schemes, distributed servers architecture, network channels and local storage tradeoff, unicast and multicast, video-on-demand. I.

In an on-demand video system, the repository servers storing all the video contents generally have limited streaming capacities and may not be co-located with the users. To achieve higher user capacity and lower network transmission cost, distributed servers architecture can be used. In such a system, multiple local servers are placed close to the user pools and, according to their local demands, the servers dynamically cache the contents streamed from the repository. We study in this paper a number of caching schemes by considering whether the local servers can exchange their cached contents among themselves or not. All the caching schemes keep a sliding window worth of data for each video being displayed; hence a video can be partially stored. We study the trade-off between the storage capacity required and the network channel used in each scheme, and address how the system cost can be minimized by appropriately sizing the window. We also show the cost advantage in using such a syste...

The proliferation of the Internet is leading to high expectation on the fast turnaround time. Clients abandoning their connections due to excessive downloading delays translates directly to profit losses. Hence, minimizing the latency perceived by end-users has become the primary performance objective compared to more traditional issues, such as server utilization. The two promising techniques to improve the Internet responsiveness are caching and replication. In this paper we present an overview of recent research in replication. We begin by arguing on the important role of replication in decreasing client perceived response time and proceed by illustrating the main topics that affect its successful deployment on the Internet. We analyze and characterize existing research, providing taxonomies and classifications whenever possible. Our discussion reveals several open problems and research directions. 1

The Replica Placement Problem (RPP) aims at creating a set of duplicated data objects across the nodes of a distributed system in order to optimize certain criteria. Typically, RPP formulations fall into two categories: static and dynamic. The first assumes that access statistics are estimated in advance and remain static, and, therefore, a one-time replica distribution is sufficient (1RPP). In contrast, dynamic methods change the replicas in the network potentially upon every request. This paper proposes an alternative technique, named Continuous Replica Placement Problem (CRPP), which falls between the two extreme approaches. CRPP can be defined as: Given an already implemented replication scheme and estimated access statistics for the next time period, define a new replication scheme, subject to optimization criteria and constraints. As we show in the problem formulation, CRPP is different in that the existing heuristics in the literature cannot be used either statically or dynamically to solve the problem. In fact, even with the most careful design, their performance will be inferior since CRPP embeds a scheduling problem to facilitate the proposed mechanism. We provide insight on the intricacies of CRPP and propose various heuristics.

This paper describes the design and analysis of RobustWeb, a scalable and fault-tolerant web server cluster that is based on HTTP redirection. The system consists of a set of N back-end document servers and one or more redirection servers which receive the HTTP requests and redirect them to the document servers. A load distribution algorithm is used for initial distribution of the documents on the servers. Given a specific degree of replication k, the distribution algorithm guarantees that at least k replicas of each document are present after document distribution is complete. The redirection servers redirect requests to one of the replicas with a pre-computed redirection probability. When a server fails, the redirection probabilities are recomputed using a novel algorithm based on network flow. This enables the load to be approximately balanced among the remaining servers, allowing for graceful degradation of the service in the event of failures. A preliminary prototype of RobustWeb ...

this paper is structured as follows. The following section discusses the load balancing techniques used in the system. The next two sections discuss the distribution of the hierarchy and its formation. The use of the previously described capabilities in the modelling and control of the hierarchy is then discussed. The section after that describes the design of the individual storage nodes, and the last section concludes, summarising the achievements. TECHNIQUES FOR LOAD BALANCING Two methods of balancing load across the server instances are used in the architecture: network striping and file replication. Network file striping is the most efficient approach. It is similar in concept to RAID at the level of the network. With this technique, individual files are split into pieces and each piece is stored separately on a different storage instance. Retrieval of a stored file involves the cooperation of all the storage instances to retrieve the full file. In this case rather than the full stream load being imposed on one server (a load of 1), it is shared across the instances (a load of on each). The total number of streams that server instances can support is thus or striped stream replays. The total disk capacity in the system also increases linearly as n increases, this is because each instance stores exactly one unreplicated piece of a file. This non replication, however, has its disadvantages - each storage instance is dependant on all the others for replay of data. If one or more instances are unavailable then the data on the other instances is rendered useless.

The access patterns of most information systems follow the 80/20 rules. That is, 80 % of the requests are for 20 % of the data. A video server can take advantage of this property by waiting for requests and serving them together in one multicast. This simple strategy, however, incurs service delay. We address this drawback in this paper by allowing clients to receive the leading portion of a video on demand, and the rest of the video from an ongoing multicast. Since clients do not have to wait for the next multicast, the service latency is essentially zero. Furthermore, since most services require the server to deliver only a small leading portion of the video, the server can serve many more clients per time units. We analyze the performance of this approach, and determine the optimal condition for when to use this strategy. We compare its performance to a hardware solution called Piggybacking. The results indicate that more than 200 % improvement is achievable.

Recent advances of computer storage architectures, high speed networking technologies and compression techniques support a proliferation of multimedia information dissemination via computer networks. For example, applications like video-ondemand (VOD) and home shopping services are now possible. A major requirement of VOD application is that the VOD storage server has to store and retrieve a huge amount of video data in a continuous manner. Therefore, near-line storage devices, such as robotic tape library, have to be employed in VOD storage servers which can contain thousands of large video objects. In this paper, we study in detail the design issues of a hierarchical multimedia storage server that consists of a disk array for caching and a robotic tape library for the storage of all the video objects. We propose and study in detail the scheduling and replacement policies of the hierarchical multimedia storage server. We illustrate the file layout policy in the parallel disks system, ...

Abstract. We propose and analyze an architecture for storage servers in large Video on Demand (VoD) systems. We describe a method for distributing the collection of titles among the levels of the storage hierarchy, based on estimates of the mean demand for each title. The resulting distribution minimizes cost for a given level of performance. Since high availability is desirable in VoD systems, we consider the use of mirroring or parity-based redundancy (a laRAID) and estimate the e ect on the system's cost and availability. In the very-large-scale storage systems needed for VoD, the placement of disk arrays on the pool of computers mustbechosen carefully to provide high availability for the least cost. We propose a strategy for arranging disk arrays on a pool of PCs � our strategy is inspired by Holland and Gibson's work on parity declustering for RAID. 1