Real-time news data placement policy considering data access patterns

Continuous media data (Video-On-Demand, News-On-Demand etc.), in characteristic, need not only a real-time processing for the users' requests but also a high disk I/O bandwidth. NOD data are smaller sire than VOD data and the popularity of NOD data is move leaning than that of VOD data. Therefore, we should have a buffer cache as to serve users' requests efficiently. A number of NOD servers will be hierarchical structures with buffer cache a,ld Mle ave going to require using new strategies of disk placement policy for these systems. In order to use efficient disk I/O bandwidth, the previous studies proposed data placement schemes on disks according to ifs popularity. But these policies did not take disk access probability into consideration. As a result of that, If there exists a data in buffer cache, which has a number of requests but loll. probability of disk access, then previous placement policies are not efficient anymore in modern system. Hence. this paper studies the disk placement policy for original and replica data, which considers both data popularity and disk access probability. In this policy we can replicate data that have not best popularity, but disk access probability. By reason that the best popular data is in buffer cache and relatively have low disk access probability. This method can increase user sen ice capacity by repeating data that have second best request popularity, but has higher probability of disk requirement than others.