The Algorithm And System Development Of The Detection Of Performance Conflict Based On CPI

With the rapid development of cloud computing and virtualization technology, performance conflicts caused by resource sharing have been brought to our attention. In the cloud, resources, such as three levels of cache and memory bandwidth are shared by multiple process, when a process takes up a large number of shared resources, other process won’t get needed resources, which affects performance. That is the performance conflict caused by resource sharing.The study found that CPI(cycles-per-instruction, cycles per instruction operation) provided by the hardware performance counter can reflect performance conflicts caused by resource sharing between the programs. when a process is limited to get the shared resource, it’s CPI rise anomaly and every instruction execution will slow down. As a result, application side will be delayed. Through the process monitoring by CPI, when performance conflict occurs, we can identified it in time. Closing or limiting related processes will release shared resource and restore the interference process performance.This paper focuses on application performance conflict detection techniques and algorithms based on the CPI, and try to develop application system. The main content and innovation as follow :1. The relationship between CPI and performance conflict is studied. The main result of performance conflict is delay on the business end. The study found that the CPI presents positive correlations with delay. When conflicts of performance caused by resource sharing occur, the CPI will be unusually high and business side will be delayed. The real-time collection of CPI data can effectively monitor the performance of the conflict.2. A detection algorithm of performance conflict based on the CPI is put forward. First of all, testing different dimensional modeling by using the collection of CPI data without interference, the prediction interval of specific nodes CPI is presented. And then monitoring the CPI data, when the CPI value beyond the scope of prediction intervals, it shows that process was disturbed, and computer performance conflict exists.3. Based on the CPI performance conflict detection algorithm, builds a CPI monitoring distributed system of performance conflict based on Docker.The experimental results show that: the detection algorithm of performance conflict based on the CPI can effectively detect the computer performance conflicts due to resource sharing. The CPI monitoring system based on the algorithm can work normally.