3. PREAMBLE TO THE DESCRIPTION

The following specification particularly describes the invention and the manner in which it is to be performed

Complete Specification

I. Title of Invention

A high efficiency Video-On-Demand (VoD) server with prefetching of Video Programs using Adaptive Resonance approach based on interest patterns of the clients.
A VoD server product with high efficiency user management with adaptive resonance approach I and cache management. High end CPU with suitable O.S like Linux or Windows servers. High end disk array of terabytes range large memory catering to the cache 20Gbytes. High speed dual mode cache RAM's directly managed under CPU with less buffering delays. Support protocols communication system works as HTTP server or any direct VoD protocol X.23. An Adaptive resonance approach is used for handling consumer request for VoD. This approach is an on-line learning system, allowing both continuous learning (plasticity) and guaranteeing a stable internal representation. Method of managing the cache. Upper segments of clusters automatically filter through adaptive resonance approach. A proportional video programs from each cluster is allowed such that total videos at any time is 1000 programs.

II. Field of Invention

This product, works with clustering using Adaptive Resonance Approach which helps in prefetching the multimedia objects into the proxy server's cache, from the disk and prepares the system to serve the clients more efficiently before the user's arrival of the request. This product is invented for Video-on-Demand server by Adaptive Resonance Approach applied to request- service operation effectively.

III. Background of Invention with Regard to the Drawback Associated with Known Art

In a Video-on-Demand server some of the challenges which have emerged over a period of time are related to data storage, management of processing continuous arrival of multiple requests. It results in potentially unbounded streams from memory to output port that is rapid and time varying. It is generally not feasible to store the request arrival pattern in a traditional database management system in order to perform delivery operation of a video stream at a later time. Instead, our invention deals with handling the request arrival through adaptive resonance approach processed in an online manner and manage accordingly. This also holds the predicatively prefetched video streams and the proposed system assures that the results can be delivered with a small start up delay for the videos accessed for the first-time.

IV. Objective of Invention

The main objective of the invention is to prefetch and store the video data from the server and store in the prefetch cache and to start the streaming of the video at client side with a minimum waiting time.

To cluster users according to the user's request patterns based on adaptive resonance approach of neural network that offers an unsupervised clustering. To decide how two video streams fall into one cluster.

To maximize the total revenue the service provider can make which is limited by the capacity of the server and number of active videos that are currently present in the cache.

V. Statement of Invention

In this invention users are grouped according to the Video-on-Demand user request pattern and it is clustered based on Adaptive Resonance Approach of neural network algorithm. The knowledge extracted from the cluster is used to prefetch the multimedia object from each cluster before the user's request. An algorithm to cluster users according to the user's request patterns based on Adaptive Resonance Approach of neural network algorithm that offers an unsupervised clustering. This approach adapts to changes in user request patterns over a period without losing previous information. Each cluster is represented as prototype vector by generalizing the most frequently used video blocks that are accessed by dl the cluster members.

VI. A Summary of Invention

Adaptive Resonance approach adapts to the change in users request access patterns overtime without losing information about their previous access pattern. In the Adaptive Resonance approach the popularity frequency value is normalized between [0,1].Hence the input values for the vector varies along with the frequency count. It prefetches the request with an accuracy as high as 96% as compared with conventional system The reference count value is highly variable over short time scales, and is much smoother over long time scales. This property allows the popularity value to deal with the long term measurement of request frequency.

VII. Brief Description of the Accompanying Drawings

FIG. 1: Overview of the Invention clustering and prefetching using Adaptive Resonance approach of neural network model.

FIG. 2: The process of clustering and Prefetching is presented. FIG. 3: Converting the popularity value to an analog value. FIG 4 : The working model of our Invention
We the applicant(s) in the convention country hereby declare that our right to apply for a patent in India is by way of assignment from the true and first inventor(s)

VIII. Detailed description of the invention with reference to drawings Multimedia streaming servers are designed to provide continuous services to clients on demand. A typical Video-on-Demand service allows the remote users to play any video from a large collection of videos stored on one or more servers. The server delivers the videos to the clients, in response to the request.

Multimedia Streaming Servers, specifically customized for HTTP, RTSP based streaming are ideally suited for developing quick LP. based streaming systems.
Multimedia streaming setup, as shown in FIG 1, includes two types of interactions. The streaming server processes the real-time multimedia data and sends it to the clients, through the various possible types of sources. The request arrival 100 can be Poisson distribution .The request arrived at 101 is forwarded to 101. 101 classify the request pattern into different group of clusters. This information is used to prefetch the VoD file and into the buffer, before request arrival. 102 represent the clustering of request. 103 predicts and prefetches the video files into the buffer.

Unlike the download and-play mechanism, the multimedia streaming client starts playing the media packets as soon as they arrive, without holding back to receive the entire file. While this technology reduces the client's storage requirements and startup time for the media to be played, it introduces a strict timing relationship between the server and the client. The clustering of video streams concerns the concept of distance or, alternatively, similarity between streams. It is required to decide how two video streams fall into one cluster. Here, we are interested in the time-dependent evolution of a request generated to a video stream. That is to say, two different requests to video streams are considered similar if their evolution over time shows similar characteristics. The customer behavior change over time the accurate predictions are rather difficult. A successful video streaming server is the one which offers customer with large selection of videos. While prefetching the videos from the server's disk and clustering them in one cluster the different factors that are considered are, the time at which the requests are generated and the popularity of the video .

The each request arrived is assigned with a popularity value which is the frequency count of that video S201 .In turn these values are converted to get an analog value relative to the corresponding popularity S202. This will be the input to the adaptive resonance model S203.S204 the vigilance factor controls the number of clusters that can be framed at a given time. The videos are prefetched from the clusters and stored in the cache S205 so that the service can be provided before the request arrival.

The popularity value is the most important parameter to get the effective prefetch operation. This popularity value may consider the long term measurement of request-frequency, which is neglected in the other algorithms. In this work the reference count value is used to get the popularity value S300. The reference count value is highly variable over short time scales, but this is much smoother over long time scales. This property makes the popularity value to deal with the long term measurement of request frequency.S302 converts the frequency count to an analog value, which considers relative value of maximum and minimum value of the popularity value at any given time.

Architecture of Adaptive Resonance approach is shown in FIG 4. It is designed for processing analog as well as binary input patterns. Proposed network module includes two main parts: attentional subsystem and orienting subsystem. Attentional subsystem preprocess analog input pattern, and then choose the best matching pattern under competitive selection rule from the input pattern prototypes. Orienting subsystem carry out similarity vigilance-testing of the selective pattern prototype and trigger resonance learning and adjusting weight vectors when vigilance-testing passed, otherwise get rid of the current active node and search the other new ones. If there is no pattern prototype matching the input pattern, create a new output node to represent it. It's memory capacity can be increased with the increase of learning patterns. The network allows not only off-line learning but also in an on-line learning and applying way simultaneously, that is, the learning and applying states are inseparable.

Clustering is the process of unsupervised classification of patterns (observations, data items, or feature vectors) into groups (clusters). Clustering can be defined as the process of separating a set of objects into several subsets on the basis of their similarity. The aim is generally to define clusters that minimize intracluster variability while maximizing intercluster distances, i.e.finding clusters, which members are similar to each other, but distant to members of other clusters. A proportional video programs from each cluster is selected so that at any time a total of 1000 videos are present in the cache.

1. The overload of network is reduced by clustering and prefetching a community of users.

2. The invention improves the current client management of VoD servers with efficient adaptive resonance approach.

3. Adaptive resonance approach subsequently enables the VoD sever to place most appropriate video programs relevant to different interest groups which were formed currently in the client domain.

4. Adaptive resonance filter, classify the request into different groups based on program interest,since the popularity value is normalized.

5. Start up delay is reduced since most popular videos are stored in the cache.

6. It prefetches request with an accuracy as high as 96% as compared with conventional system.