The cache partition mechanism is introduced to solve the problem of contention and pollution of caches in Chip Multiprocessor. One of the key questions is to analyze online the utility characteristics and efficiency of cache requests originating from different sources. Cache request is differentiated by hardware thread context id. Based on the working set model, within a certain sampling time frame, a number of specific cache lines are preserved by modifying the cache address mapping mechanism and its number is the same as that of the hardware threads of the processor, performing statistics on the frequency of Stack Distance under LRU respectively. Based on the characteristic of the frequency distribution of Stack Distance, We propose a metric, L*D, to measure online the recognition of the cache utility efficiency of the working set and providing information for the decision making of cache allocation and management. Experiments performed in multiple Benchmarks in Spec Cpu2000 show that the online monitoring mechanism and quantitative metric discussed in this article have achieved fine-grained online monitoring of the efficiency of cache utility based on the working set. Not only do they allow the online recognition of cache space utility characteristic, but they do also enable the identification of any changes in the efficiency of cache utility, achieving an accuracy of over 75%.