IEEE Consumer Electronics Magazine - January/February 2022 - 74

Special Section on Secure and High-Speed Electronic Systems
evaluation of the scheduling algorithm proposed.
The evaluation results show that our
solution can significantly improve system
performance.
BACKGROUND
Reducing the load unbalance on heterogeFigure
1. CPU-GPU heterogeneous computing
platform.
has reached dozens. It is difficult for a processor
to support such a huge amount of calculation.4,5
Heterogeneous computing systems have so far
been developed to incorporate various computing
units such as CPUs, DSPs, GPUs, and FPGAs.
Such heterogeneous systems combine computing
units for various types of instruction sets,
allowing different cores to cooperate to deal
with multiple computing scenarios and improve
system performance effectively. While heterogeneous
systems provide a novel platform for
emerging computing paradigms, such as big
data and cloud computing, they also face many
challenges. Figure 1 shows a CPU-GPU heterogeneous
computing platform architecture. One of
the challenges is the inability to match the
appropriate processor cores during the execution
of multitasking, which seriously restricts
the performance of heterogeneous computing
systems. An efficient task scheduling strategy is
of great importance to solve this problem to
improve system performance.
In this article, we focus on improving the performance
of heterogeneous multicore processors
and propose a swarm intelligence task
scheduling strategy based on a genetic algorithm
(GA). First, we propose a swarm intelligence
task scheduling strategy for highperformance
heterogeneous multicore processors
based on a GA to improve the performance.
Second, we introduce a mechanism to detect the
computing capacities of different cores by characterizing
specified tasks beforehand, and we
propose to use an adaptive mutation strategy
and injection strategy to avoid premature convergence
and fall into a local optimal solution.
Third, we perform an extensive experimental
74
neous computing systems and fully utilizing the
computing capabilities of heterogeneous processing
cores have become an urgent challenge
to be tackled in the field of high-performance
computing.
To solve the task scheduling problem in multitask
computing, linear programming-based
methods6 have been explored by Baruah et al.
There are other prior work focusing on the
assignment problem of multitasks.7-9 A combination
of GA and particle swarm optimization
(PSO) algorithm10 was proposed by Li et al.An
efficient hybrid GA and PSO algorithm for molecular
dynamics simulation on heterogeneous
supercomputers load balancing was further
developed. To improve the efficiency of multitasks
execution, a bypass shared cache management
method11 was developed by Juan et al.A
free and open-source software toolkit based on
GUN radio12 was developed by Perea et al. Techniques
have also been proposed to utilize the
support vector machine classification method to
classify computing tasks into CPU type and GPU
type.13,14 Machine learning-based techniques
have also been proposed to improve consumer
electronics performance.15,16
SWARM INTELLIGENCE TASK
SCHEDULING STRATEGY
The scheduling strategy designed in this article
aims to improve the overall operating efficiency
of the system. Binding the execution
relationship between various types of tasks and
heterogeneous processor cores appropriately,
we employ an improved genetic algorithm
(IM_GA) to reduce the additional overhead
caused by general task scheduling schemes.
Aiming at the problem of falling into local optimal
solutions and premature convergence in a
GA, we first do prescheduling by detecting the
computing capacities of different cores by processing
specified tasks beforehand, and then we
IEEE Consumer Electronics Magazine
IEEE Consumer Electronics Magazine - January/February 2022

Table of Contents for the Digital Edition of IEEE Consumer Electronics Magazine - January/February 2022
