L1/L2 Apparatus and Method for detecting Working set Size of Application Based on L1/L2 Prefetch Characteristic in Many-core

摘要

The present invention provides a new method and device to identify a working set based on L1/L2 prefetch characteristics through various techniques such as a working set model and a stack model for dynamically identifying a working set of applications. The present invention comprises: a hardware performance counting part monitoring performance characteristics including the number of prefetches and the consumption of L1, L2 resources of cores; a hardware prefetching part predicting the next data to be approached by a core based on a monitoring result of the hardware performance counting part, and loading the predicted data from a memory into a CPU cache before the data approaches the memory; an extracting part extracting a memory address approached by each workload by using a PIN tool which identifies an address approach for confirming the working set based on the predicted data loaded by the hardware prefetching part; and a verification part allocating memory by verifying the size of the working set by analyzing a memory reference pattern of application based on the memory address extracted by the extracting part.