Source-linked AI summary
Hybrid Genetic Algorithm for Cloud Computing Applications
Saeed Javanmardi, Mohammad Shojafar, Danilo Amendola, Nicola Cordeschi, Hongbo Liu, Ajith Abraham
TL;DR
Cloud scheduling must assign jobs to resources while balancing load and reducing execution time and cost. This paper combines a modified genetic algorithm with fuzzy theory to schedule jobs using job lengths and resource capacities, reporting about 45% lower execution cost and about 50% lower total execution time.
Problem
Cloud scheduling needs to assign jobs across resources while balancing system load and reducing execution time and cost.
Method
The approach modifies a genetic algorithm with fuzzy theory to assign jobs according to job length and resource capacities.
Results
About 45% lower execution cost and about 50% lower total execution time were reported for the fuzzy-modified genetic algorithm.
Takeaways & Limitations
The paper concludes that fuzzy modification of the genetic algorithm improves cloud-scheduling performance in execution cost and total execution time.
Abstract
from arXiv · showhide
In this paper with the aid of genetic algorithm and fuzzy theory, we present a hybrid job scheduling approach, which considers the load balancing of the system and reduces total execution time and execution cost. We try to modify the standard Genetic algorithm and to reduce the iteration of creating population with the aid of fuzzy theory. The main goal of this research is to assign the jobs to the resources with considering the VM MIPS and length of jobs. The new algorithm assigns the jobs to the resources with considering the job length and resources capacities. We evaluate the performance of our approach with some famous cloud scheduling models. The results of the experiments show the efficiency of the proposed approach in term of execution time, execution cost and average Degree of Imbalance (DI).
1 Introduction
Cloud computing provides internet-accessible, shared services through an infrastructure whose location is hidden from users. Scheduling algorithms assign resources to jobs while seeking to reduce execution time and cost and maintain system load balance.
- 1 Introduction: Users access cloud services through the infrastructure without knowing where those services are located.Users pay for the services they request.
- 1 Introduction: Scheduling algorithms assign resources to received jobs while targeting lower execution time and cost and better system load balancing.A good scheduler should consider both load balancing and total execution time.
- 1 Introduction: Cloud infrastructure provides on-demand access to shared resources and services over a network.These services can be accessed through the internet.
2 Related Works
Prior cloud-scheduling research emphasizes efficient load balancing across distributed resources and applies artificial-intelligence methods such as genetic algorithms and fuzzy theory. Related models rank jobs, analyze classical scheduling algorithms, use premigration across multiple load dimensions, and combine genetic algorithms with knapsack-based multiobjective fitness.
- 2 Related Works: Cloud load balancing aims to distribute jobs efficiently across many distributed computing resources, motivating artificial-intelligence approaches such as genetic algorithms and fuzzy theory.
- 2 Related Works: A job-oriented scheduling model assigns cloud resources according to job rank and analyzes Round Robin, Preemptive Priority, and Shortest Remaining Time First algorithms.
- 2 Related Works: A premigration-based model addresses scheduling using CPU utilization, network throughput, and disk I/O rate as three load dimensions.
- 2 Related Works: A hybrid genetic algorithm combined with a multiple-fitness knapsack problem is proposed to improve resource utilization and reduce energy consumption.The reported approach claims both lower energy consumption and increased resource utilization.
3 Proposed Approach
The proposed approach represents jobs as genes in chromosomes and uses two chromosome-construction criteria with fuzzy theory to guide crossover, resource assignment, and fitness evaluation. Fuzzy membership sets and Mamdani rules incorporate job and VM characteristics to select suitable assignments.
- Chromosome Representation: Jobs are represented as genes assigned to computational resources, while sets of genes form chromosomes constructed using job length, CPU speed, and resource RAM criteria.A second chromosome type is also created using another criterion, but the supplied passage truncates its details.
- Fuzzy Crossover: Fuzzy crossover replaces single-point and two-point crossover, using job length, VM bandwidth, VM MIPS, and assigned-resource RAM to guide chromosome and resource selection.The fuzzy system targets both parent-chromosome selection and job-to-resource assignment.
- Fuzzy Fitness Evaluation: Fitness evaluation maps input parameters to overlapping fuzzy sets through membership functions that assign degrees between 0 and 1.The approach creates three overlapping fuzzy sets and uses membership functions to determine each input’s degree of belonging.
- Fuzzy-Set Implementation: Fuzzy sets for job length and VM MIPS are created with MATLAB’s fuzzy logic toolbox for the first and second performance experiments.The supplied passages identify these sets as inputs to the performance-evaluation experiments.
- Fuzzy-Rule Reasoning: For VM MIPS of 1000, membership degrees are 0.3 for low, 0.7 for medium, and 0 for high, supporting Mamdani fuzzy-rule reasoning.The rules are defined according to the cloud environment and its administering policy.
4 Performance evaluation
CloudSim experiments show that the proposed hybrid scheduler outperforms ACO and MACO in makespan growth and degree of imbalance. Its performance is attributed to assigning longer jobs to more powerful resources while considering job length and resource capacity.
- Experimental setup: The experiments use CloudSim with 10 data centers, 50 VMs, and 100–1000 jobs whose lengths range from 1000 to 20000 MI.The first two experiments compare the proposed approach with ACO and MACO using makespan and degree of imbalance.
- First experiment: The proposed approach achieves lower makespan growth than ACO and MACO as the number of jobs increases.Its increment ratio is lower because it combines local and global search.
- Second experiment: The proposed approach achieves better degree of imbalance than ACO and MACO.The degree of imbalance is dimensionless and reflects the maximum, minimum, and average total execution times across VMs.
- Second experiment: The scheduler assigns larger jobs to more powerful resources by considering both job length and resource ability.This assignment reduces each VM’s total execution time and explains the improved degree of imbalance.
5 Conclusion
The paper modifies a genetic algorithm with fuzzy theory to reduce population-generation iterations, using two chromosome types with different QoS parameters. The approach improves system performance by reducing execution cost by about 45% and total execution time by about 50%.
- 5 Conclusion: Fuzzy theory modifies the standard genetic algorithm by reducing population-generation iterations and evaluating chromosome fitness.The approach defines two chromosome types with different QoS parameters.
- 5 Conclusion: About 45% lower execution cost and about 50% lower total execution time are achieved with the fuzzy-enhanced genetic algorithm.These improvements target the study’s main goals of reducing execution cost and total execution time.