Optimizing Kernel Block Memory Operations

This paper investigates the performance of block memory operations in the operating system, including memory copies, page zeroing, interprocess communication, and networking. The performance of these common operating system operations is highly dependent on the cache state and future use pattern of the data, and no single routine maximizes performance in all situations. Current systems use a statically selected algorithm to perform block memory operations. This paper introduces a method to dynamically predict the optimal algorithm for each block memory operation. This dynamic selection involves the prediction of the current state of the cache as well as whether or not the target data will be reused before it is evicted. By using the results of these predictions to select the optimal software algorithm, the performance of kernel copy operations can be improved.