A Method for Energy Consumption and Performance Optimization of Many-Core Systems Based on Cooperative Approximate Computing

Abstract

The invention discloses a method for optimizing energy consumption and performance of a many-core system based on cooperative approximation calculation. The method can combine different abstraction layers and adopt multiple approximation techniques on the premise that the output result of the control application program meets a certain error range, including in Reduce the computing workload of the application at the application level, selectively delete data at the network layer to reduce network congestion, and apply approximate computing to different abstraction layers of the many-core system through the optimal regulation of the global controller and the resource allocation of the local controller . The invention measures the importance of discarded data based on the quality model, considers the collaborative management of communication and computing, formulates a multi-objective optimization problem to minimize network congestion, application running time and limit the quality of results, and provides accelerated applications for many-core systems New ways to program runtime, reduce energy consumption, and improve chip-level energy efficiency.

Description

technical field [0001] The invention relates to the technical field of resource management of a many-core system, in particular to a method for optimizing energy consumption and performance of a many-core system based on cooperative approximate calculation. Background technique [0002] Chip multiprocessors have become the mainstream for building high-performance computers, and many-core systems are widely used in cloud computing servers and big data systems to provide sufficient computing power. Transistor scaling has driven a surge in the number of processing elements available in many-core systems' on-chip networks, but many-core chips cannot scale properly with chip area due to overheating issues caused by Dennard scaling breakdown. Under the premise of satisfying the power consumption budget of the many-core system, only a part of the nodes can be turned on, and the rest of the nodes can be turned off to prevent possible overheating and permanent chip damage. Unfortuna...

AI Technical Summary

Problems solved by technology

[0005] Unilateral approximation concepts, while potentially helpful in improving performance at one abstraction layer, do not guarantee optimal power efficiency of the chip, and introducing approximate computation at one abstraction layer may impair the execution of another abstraction layer

Images