| 摘要 |
In a data processing system, one or more cache management technique are implemented in a distributed file system of a computer network to enforce equitable use of the cache among the file data in the cache. One technique is a timestamp handicapping routine that functions to keep small-to-medium files in the cache. Another technique implements a cache quota, which limits the percentage of the cache which the data from a single file may consume when there is other data in the cache. When caching of a single file approaches the cache quota, the file data is made to look older than it really is so that upon a subsequent file I/O operation, portions of such data are recycled from the cache earlier than they would have been otherwise. When caching of a single file reaches the cache quota, the file must begin reusing cache from itself or from handicapped chunks from other files. The cache quota technique has the effect of causing cached data towards the end of large files to get flushed or recycled from the cache first. A third technique helps to detect file I/O that is not conducive to caching, such as sequential I/O on a file that is larger that the entire cache. A cache policy prevents the large file from stealing cache space by establishing a small, but dedicated area of cache in which portions of such large file may be stored and recycled without requiring a least recently used (LRU) evaluation process. <IMAGE> |
