| 摘要 |
A method for avoiding a cache-full condition in an LRU organized, data set partitioned, page oriented cache of a defined capacity intercoupling a processor to an external storage subsystem is provided. The processor is capable of storing pages of a data set into a plurality of cache partitions. The method comprises the steps of periodically determining the target size of each partition for optimal partitioning, and assigning a reclaim vector to each partition specifying the number of frames to be stolen from each of the other partitions necessary to achieve the corresponding target size. If the target size of a particular partition is less than the partition's virtual minimum size, the method adjusts that partition's reclaim vector to enable the partition to steal sufficient frames from other partitions to achieve the virtual minimum size. In response to a cache miss, frames are stolen from other partitions, as directed by the reclaim vector of the partition storing the data set to which the missing page reference is a member, by removing the lowest page in LRU order and adding the resulting frame to the storing partition and modifying the storing partition's reclaim vector.
|
