Multi-objective scheduling technique based on hybrid hitchcock bird algorithm and fuzzy signature in cloud computing

Abstract

As the number of requests in the cloud increases, providers face more problems (e.g., task scheduling and resource management). The exhaustive search for determining the optimal solutions of scheduling problem is usually impractical and hence metaheuristic algorithms have been widely used to evolve solutions for task scheduling. In this paper, we firstly propose a hybrid meta-heuristic algorithm called Hybrid Fuzzy Hitchcock Bird (HFHB) and second, a multi-objective form of HFHB (MOHFHB) is introduced to solve multi-objective problems (e.g., task scheduling). The HFHB algorithm consists of three main enhancements: (a) The first random population of birds is improved, (b) The attack regulator parameter is set with a fuzzy Sugeno-signature, and (c) The dead birds are replaced with new birds. In multi-objective form (MOHFHB), two concepts (i.e., crowding distance and ranking non-dominated solution) are added to determine the optimal Pareto front. In the first part, HFHB and MOHFHB are evaluated as two global optimizers. In the second part, HFHB and MOHFHB are evaluated for a task scheduling problem against Moth Search Algorithm with DE (MSDE), Enhanced Multi-Verse Optimizer (EMVO), Fuzzy Modified Particle Swarm Optimization (FMPSO), and Simulated-annealing-based Bees Algorithm (SBA). The results indicate that HFHB improves makespan by 12.57%, 38.61%, and 35.75%, and resource utilization by 1.14%, 7.30%, and 5.25% compared to FMPSO, MSDE, and EMVO, respectively.

Introduction

Cloud computing is a popular computational environment that allows users to scale their resources quickly. Cloud user sends tasks while the provider of cloud presents resources for the execution of tasks. Both participants have various goals: the provider tries to obtain as many profits as possible and enhance the resource utilization while the user attempt to execute tasks with minimum execution time and cost (Mansouri and Javidi, 2020b, Ghahramani et al., 2017, Mansouri and Javidi, 2018, Mansouri et al., 2020b). In the performance of cloud computing, task scheduling plays a vital role in which resources are allocated and adjusted between different tasks. From the booming economy of information processing, designing an efficient task scheduling algorithm has attracted worldwide attention to enhance the quality of service (QoS) (Mansouri and Javidi, 2020a, Alboaneen et al., 2021). The task scheduling process in the cloud is explained in Fig. 1. Firstly, user tasks are submitted to a waiting queue and then the scheduler (i.e., data center broker) gives the necessary information (i.e., available cloud resources) from cloud information services (CIS). Then, it allocates tasks to suitable virtual machines based on the scheduling algorithm. Finally, the selected machines execute the assigned tasks by obtaining the necessary data. Generally, task scheduling algorithms are divided into two categories (i.e., static and dynamic). The static scheduling algorithm needs advanced information about tasks and environments. They are suitable for a system with less workload variation. While dynamic scheduling algorithm monitors the system continuously and can balance the workload. In general, the task scheduling problem is modeled as follows:

Assigning n tasks of users ($T=\{T_1,T_2,\dots ,T_n\}$) to m heterogeneous virtual machines $VM=\{V{M}_1,V{M}_2,\dots ,V{M}_m\}$ where n $>$ m subject to some constraints to optimize some objective functions.

Due to huge computational demands, energy consumption is one of the critical challenges in cloud data centers, which affects the benefit of service providers and meets the standard set in green computing. In Amazon, more than 50% of the budget for data center management is related to powering and refrigeration of servers. Among different ways of energy-efficient, task scheduling is at the heart of successful power reduction and resource management in cloud computing. A lot of studies have already been done to propose efficient task scheduling algorithms in recent years. In Pradhan et al. (2021), the most recent relevant literature is surveyed and the results indicated that service providers’ objectives (i.e., throughput, energy consumption, load balancing, and resource utilization) have not been considered broadly during task scheduling simultaneously. Most of them focused on execution time and did not achieve an acceptable trade-off between QoS factors and energy consumption. For example, Elaziz et al. (2019) introduced a new task scheduling strategy based on the MSA and DE algorithms to reduce makespan. Nevertheless, they did not analyze resource utilization and energy consumption. Chaudhary and Kumar (2019) presented a new scheduling technique based on Hybrid Genetic-GSA (HG-GSA) to reduce execution cost and transfer cost. But, they did not mention makespan and energy consumption.

However, heterogeneous and dynamic behavior, various Quality of Service (QoS) parameters, and unified load balancing are the difficulties of task scheduling design in cloud computing. Moreover, the task scheduling problem is one of the most common combinatorial optimization problems to optimize the key performance factors. Many previous task scheduling algorithms are limited by the fact that they assume only a small number of performance factors. In this work, we tackle the problem of developing a task scheduling strategy to optimize makespan, resource utilization, energy consumption, latency, and degree of load balance. Since low makespan and delay values show that tasks are completed within the desired time and the high degree of load balance ensures that the flexibility and scalability of data centers are well. In addition, improving the utilization of resources can solve the challenge of over-provisioning and under-provisioning and energy efficiency provides environmental and economic implications.

In this regard, different algorithms have been introduced for addressing scheduling problems (e.g., Minimum Completion Time (Madni et al., 2017), First Come First Serve (Kaur and Kinger, 2014), and Min–min (Kaur and Kaur, 2015) and Max–min (Konjaang et al., 2016)). Although these algorithms have been adopted on many cloud computing systems, some research works have verified that these strategies are inappropriate for large-scale scheduling problems (Kaur and Kaur, 2015). Since task scheduling is an NP-hard problem due to its complexity and the necessary time for finding the solution varies by the problem size (Cui et al., 2017). We know that the meta-heuristic techniques can handle a large search space of scheduling problems and determine an optimal solution in polynomial time (Ebrahimzade et al., 2020). In 2021, Houssein et al. (2021) provided a review of task scheduling based on meta-heuristics and the results represented that PSO, GA, and ACO techniques have been widely applied by researchers. Currently, most scheduling algorithms that are used to allocate tasks are not accurate and unable to satisfy the needs of cloud computing development (Alsaidy et al., 2020, Samuel Raj and Rachel, 2013) especially when the number of objectives is increased. In addition, each meta-heuristic technique has some drawbacks. For example, PSO has a slow convergence rate and weak local search. GA has a problem with complex functions in the selection and crossover operators. ACO shows complex computation and complexity in the coding scheme. The nature of most meta-heuristic techniques is such that they easily trap in local optima and hence it is necessary to combine them with suitable techniques and enhance the qualities of solutions. For instance, Madni et al. (2019) introduced a novel scheduling algorithm based on CSA to reduce execution time. But all agents of CSA search the problem space in the same way and hence premature convergence may occur. Moreover, resource utilization and energy consumption are not included.

In the same context, a new meta-heuristic technique called Hitchcock Bird Inspired Algorithm (HBIA) (Morais et al., 2019) has been presented. It is a global optimization inspired by the aggressive bird behavior portrayed by Alfred Hitchcock in the 1963 thriller “The Birds”. The HBIA algorithm is based on the attack pattern of birds and has the stages of lurking, attack, and reorganization. It is successful in solving various cost functions for a variety of domains (Morais et al., 2019, Morais et al., 2020). Therefore, we select HBIA for solving task scheduling problems and finding more satisfactory solutions in large search spaces.

Nevertheless, it suffers from several limitations including the inability to explore the search space effectively and provide a balance between exploration and exploitation activities. These limitations can be observed in its behavior during the updating process (i.e., attacking phase) and generating new solutions at the end of each iteration (i.e., reorganization phase). We know that HBIA exploration depends on replacing the dead birds (i.e., the solution with the lower quality compared to the average quality of population) with new birds. In this regard, the worst solutions obtained are replaced with randomly generated new solutions. However, this process is not efficient to handle this issue and HBIA still needs more improvements. In other words, it must consider an appropriate process which blowing fresh blood to the HBIAs’ population. As a result, we firstly propose the improved version of HFHB based on the Crow Search Algorithm (CSA) (Askarzadeh, 2016), fuzzy theory (Gupta and Saini, 2017), and Levy flight concept (Elaziz et al., 2019) to overcome the mentioned challenges. Then, the improved strategy is adapted to solve the task scheduling problem in a cloud environment. The proposed algorithm (HFHB) is similar to HBIA in its starting stage, which generates a set of solutions. But the difference is in the main loop, which consists of the modified HBIA’s attack phase and HFHB uses the CSO algorithm to improve the exploration activities. Moreover, it has a new parameter called “Attack Regulator (AR)” to balance exploration and exploitation activities with fuzzy theory. Finally, it determines the worst solutions at the end of each iteration and replaces them with the Levy flight technique.

As a brief, the main contributions of this paper can be explained as follows:

• Introduce HFHB as a modified version of the Hitchcock Bird Inspired Algorithm (HBIA) with the combination of the CSO algorithm to improve the exploration phase of HBIA.

• Apply Sugeno-signature fuzzy to provide a balancing between exploration and exploitation.

• Use two techniques (i.e., Levy flight and fuzzy theory) to improve the rate of convergence, the diversity of population, and the quality of solutions.

• Add two methods (i.e., crowding distance and ranking non-dominated solution) into HFHB and introduce the multi-objective form of HFHB that is named MOHFHB.

• Model task scheduling as a multi-objective optimization problem and solve it based on fuzzy theory to optimize the key performance indicator parameters.

• Try to balance the trade-off among the five optimization objectives (i.e., makespan, resource utilization, energy consumption, latency, and degree load balance) during task scheduling.

• Present the comparative and statistical analysis of HFHB by 20 test functions to ensure its capability in solving the global optimization problems compared with Multi-Verse Optimizer (MVO), Crow Search Algorithm (CSA), Hitchcock Bird inspired Algorithm (HBIA), Moth–Flame Optimizer (MFO), Archimedes optimization algorithm (AOA), Arithmetic Optimization Algorithm (ARO), and Grey Wolf Optimizer (GWO).

• Provide the comparative and statistical analysis of MOHFHB by ZDT test functions to ensure its capability in solving the multi-objective problems compared with Multi-objective Multi-Verse Optimizer (MOMVO), Multi-objective Salp Swarm Algorithm (MSSA), and Non-dominated Sorting Genetic Algorithm (NSGA-II).

• Evaluate the performance of HFHB in terms of makespan, resource utilization, energy consumption, latency, degree imbalance, throughput as a task scheduler with the different number of tasks. In addition, evaluate MOHFHB in terms of energy consumption, degree imbalance, and makespan.

The rest of paper is divided into the following sections: Section 2 explains a brief discussion of the related works on task scheduling. Section 3 illustrates the necessary backgrounds about the fuzzy signature, Crow Search Algorithm (CSA), multi-objective problems, and Hitchcock Bird-Inspired Algorithm (HBIA). Section 4 brings our proposed approach. Section 5 models the task scheduling problem and explains how to solve it with the proposed algorithm Section 6 discusses the evaluation results and the statistical test. Section 7 presents our conclusion and future works.