Self-adaptive help thread pre-fetch quality control method

Abstract

The invention discloses a self-adaptive help thread pre-fetch quality control method, and belongs to the technical field of multi-core cache performance optimization. The method can be used for improve the quality of help thread pre-fetch in irregular data intensive applications. Real-time online self-adaptive regulation of a help thread pre-fetch control parameter value can be achieved through the techniques such dynamic sampling, program operation stage detection, pre-fetch gain feedback, and self-adaptive regulation, and then the control quality of help thread pre-fetch can be improved. The method mainly includes the following steps: 1) dynamic information acquisition and feedback statistics; 2) index calculation; and 3) self-adaptive pre-fetch control. The method fully utilizes hardware information of a multi-core platform to achieve real-time pre-fetch control of a help thread, can be widely applied to access performance optimization, help thread pre-fetch quality control and sharing cache pollution control of the irregular data intensive applications in a current big data environment.

Description

technical field [0001] The invention relates to the field of multi-core computer shared cache performance optimization, in particular to an adaptive adjustment method for controlling and helping thread prefetching quality, which can be used to improve the performance of irregular data-intensive application programs. Background technique [0002] Current data-intensive applications involve big data analysis, social network, scientific computing and other fields. Such applications often present irregular data-intensive access characteristics, which makes traditional data prefetching technology based on spatio-temporal locality invalid. In order to improve the program performance of such applications, the helper thread prefetching technology has received extensive attention. The essence of the helper thread prefetching technology is a Leader / Follower structure, which is a non-traditional parallel technology. On multi-core computing platforms, the destination for helper thread ...

AI Technical Summary

Problems solved by technology

When the environment in which the application executes or the input set of the program changes, the original control parameter value will become invalid, and it is necessary to re-optimize the control parameter value through static enumeration, which will be a complicated and time-consuming process.

Therefore, there are two main problems in the method of static selection to help threads control parameter values: on the one hand, the time spent on parameter value selection and optimization is too high, and on the other hand, the method of setting parameter values ​​offline cannot adapt to the dynamics of the program. Changes to Stage Fetch Behavior

These problems have brought great limitations to the application and promotion of thread prefetching technology

[0005] In response to these problems, the present invention proposes an adaptive adjustment method to help thread prefetch control to solve the problem of traditional manual enumeration selection parameter values

Images