HEACT: Hybrid Evolutionary Algorithm for the Multi-region Multi-objective Cloud Task Scheduling Problem. A Study of Workflow Scheduling in AWS EC2

Autor(es)

de Carvalho, Vinicius Renan ; Sichman, Jaime Simao

• Tipo de documento: Artigo Científico
• Fonte: INTELLIGENT SYSTEMS, BRACIS 2024, PT II; v. 15413, p. 16-pg., 2025-01-01.

Resumo

Cloud computing has revolutionized the provisioning and access of computing resources, offering scalable and flexible alternatives to traditional infrastructure. However, defining how to use these computational resources may be challenging. This paper addresses the challenge of workflow scheduling in cloud environments, focusing on Amazon Web Services (AWS) Elastic Compute Cloud (EC2). We present HEACT, a novel approach that integrates a multi-objective evolutionary algorithm with a specialist scheduling heuristic. The evolutionary algorithm is responsible for generating an initial set of machines (with their performance capability and cost information). The set is sent to the specialist scheduling heuristic for efficient task assignment in these machines. Our approach considers fourteen AWS regions, accurate pricing information from AWS, and employs SimGrid to simulate task execution. The proposed method was benchmarked considering established heuristics (HEFT, PEFT, HSIP, MPEFT) and meta-heuristics (NSGA-II, AGEMOEA2). Results demonstrated that the combinations of AGEMOEA2 with MPEFT and AGEMOEA2 with HEFT yield the best performance, indicating AGEMOEA2's efficacy as a state-of-the-art meta-heuristic for workflow scheduling. (AU)

• Processo FAPESP: 23/15949-8 - Otimização de recursos de nuvem baseada em IA visando a redução de emissões de carbono
• Beneficiário: Vinicius Renan de Carvalho
• Modalidade de apoio: Bolsas no Brasil - Pós-Doutorado