Thread transfer distribution method capable of preventing local overheat of multi-core processor

Abstract

The invention belongs to the processor design field, and relates to a method capable of preventing overheat of one or several cores of a processor by thread transfer distribution among the cores of the processor. The multiple processor cores in the multi-core processor are divided into multiple groups, and the multiple adjacent processor cores share the memory to form a group. Each group has a thread controller which dynamically transfers and distributes the thread to the processor cores in the group to make the temperature of the over-heat processor core become lower than a preset threshold and most close to the preset threshold with the minimum thread transfer number, and transfers the thread to the processor core with the minimum adjacent processor core number in operation so as to reduce the probability of the occurrence of the over-heat processor core in the future, the thread transfer of the multiple groups can be carried out simultaneously, so the temperature of the over-heat processor core in each group can be reduced to a temperature value which is lower than the preset threshold rapidly.

Description

Technical field [0001] The invention belongs to the field of processor design, and specifically relates to a thread transfer allocation method for avoiding local overheating of a multi-core processor. Background technique [0002] With the rapid development of integrated circuit manufacturing technology, the feature size of integrated circuit chips continues to decrease, and multi-core processors are widely used. However, because multi-core processors are embedded with multiple processor cores, the power consumption density is high, and it is easy to cause overheating of single or part of the processor cores (ie, local overheating), resulting in a decrease in the overall performance and reliability of the multi-core processor. The traditional voltage and frequency scaling method, represented by the scheme (the reference) [2]-[3], reduces the temperature of the overheated processor core by reducing the overheated processor core voltage and operating frequency, but this method will...

AI Technical Summary

Problems solved by technology

However, because multi-core processors are embedded with multiple processor cores, the power consumption density is high, which can easily lead to overheating of a single or part of the processor cores (that is, local overheating), resulting in a decrease in the overall performance and reliability of the multi-core processor.

The traditional voltage frequency scaling method, represented by the scheme (the reference) [2]-[3], reduces the core temperature of the overheated processor by reducing the core voltage and operating frequency of the overheated processor, but this method will be more reduced processor performance

The DTBM scheme (the reference) [4] and the PDTM scheme (the reference) [5] measure and calculate the temperature of each processor core by software based on the workload prediction model, and then transfer the overheated processor core. Threads are sent to cooler processor cores to reduce the temperature of overheated processor cores, but practice shows that the DTBM and PDTM cannot quickly and accurately calculate the actual temperature of the processor cores, and their thread transfer strategies are relatively simple and do not adapt to the workload The working environment of multi-core processors with frequent changes and many running threads

Images