Meta heuristic-based task deployment mechanism for load balancing in IaaS cloud

Abstract

An Infrastructure-as-a-service (IaaS) cloud is a new paradigm which offers computing and storage services in form of virtual machines to the cloud users. Due to increasing demand and diversity of the applications, most of the servers are become overloaded or imbalanced and affects the performance of the cloud system. Most of the existing strategies have developed a series of algorithms to pick out the optimal target server to achieve the immediate load balancing. However, it is proved in the literature that immediate load balancing does not meet the high execution efficiency and resource utilization of the servers. In this paper, we propose a new load balancing mechanism for a long-term process referred as LB-RC (load balancing resource clustering). The meta-heuristic Bat-algorithm is applied to obtain optimal resource clustering and their cluster centers for faster convergence. We also propose a new dynamic task assignment policy to achieve the minimum makespan and execution cost within the given constraints. The proposed algorithm is tested and compared with the existing algorithms over various synthetic datasets and performance matrices to prove its superiority.

Introduction

Cloud computing is a new distributed environment has gained popularity for providing computing resources over the Internet (Garg et al., 2012; Suresh and Sakthivel, 2017; Banerjee et al., 2014). The cloud service provider runs data centers consist of thousands of servers which can host multiple virtual machines (Adhikari and Amgoth, 2018). The cloud provides three types of services such as Infrastructure-as-a-service (IaaS), Platform-as-a-service (PaaS), and Software-as-a-service (SaaS). An IaaS cloud offers computing and storage services in form of virtual machines to the cloud users. The infrastructure of the cloud deals with dynamic load changes, ranging from peak performance to utilization gaps. This brings up two points: firstly, pre-establishment of cloud infrastructure to handle the dynamism of resource consumption and secondly, it has to utilize the resources efficiently and reduce the resource wastage (Garg et al., 2014; Teylo et al., 2017). For example, if the remaining resource capacity of the server is smaller than the resource amount requested by the task, then the server cannot handle the task within the given time constraints. On the other hand, most of the servers remain relatively ideal which cause the resource wastage of the cloud data center.

Most of the recent works (Kumar Bishwkarma and Vyas, 2016; Basker et al., 2014; Elzeki et al., 2012; Song et al., 2014; Nurmi et al., 2009; Jia, Yang, Wei, Ding, Hu, Xu; Maguluri and Srikant, 2013; Xu et al., 2013; Pham et al., 2016; Wang, Tao, He, Kuang; Dhinesh Babu and Venkata Krishna, 2013; Hu et al., 1998; Cao et al., 2014; Chitra Devi and Rhymend Uthariaraj, 2016; Gill et al., 2017a; Gill et al., 2017b; Kumar and Sharma, 2017a; Kumar and Sharma, 2017b; Wang et al., 2011; Google. Google Cluster Da; ZhangJoseph and Boutaba, 2011; Mishra et al., 2010; Moreno et al., 2014) have focused on immediate load balancing on the servers and deployed the tasks to the optimal loaded server. This may affect the performance and efficiency of the data center in the long-term and may increase the waiting time of the users. Due to increasing demands and diversity of the tasks, the resources requested by the tasks have rarely possible to meet the remaining resources available on the server. Based on the above statement, it does not ensure that the immediate load balancing mechanism is successful in achieving their objectives. Resource clustering is a procedure of organizing the resources into several homogeneous groups whose members are similar in nature (Moreno et al., 2013). Clustering algorithms are normally classified in two ways: hierarchical and partitional clustering (Sarkar et al., 1998; Frigui and Krishnapuram, 1999). In hierarchical clustering, resources are grouped like a tree structure. The partitional clustering is a procedure that attempts to divide the resources into some non-overlapping subsets. Implementing partitional clustering can group the servers into some homogeneous groups according to their load and a high degree of similarity. Some of the others clustering methods are (Zahn, 1971; Paterlini and Minerva, 2003; Omran et al., 2005). Most of the clustering methods achieve the local optimum solutions but suffer from achieving the global optimum.

In this work, we propose a new load balancing approach for finding the optimal set of servers for task assignment to balance the load of the servers in a long-term process. The proposed algorithm referred as LB-RC. The algorithm is divided into four phases, resource clustering, merging of clusters, resource optimization, and task assignment policy. In the first phase, servers are grouped into homogeneous clusters based on their loads and cluster centers. Next phase, the null cluster and small cluster problem are solved by adopting a rule-based merging technique. This phase merges the null cluster and the small cluster with the similar types of the clusters to reduce the number of clusters. In resource optimization phase, the population size is optimized using Bat algorithm to obtain the optimal clustering set of the servers to overcome the global optimum problems. The Bat algorithm is a nature-inspired technique and it is selected for the optimization of cluster centers because this procedure is equipped with quick convergence capability for detecting multiple objects. Finally, the tasks are deployed to the optimal servers with the best-fit VM instances based on the QoS constraints such as deadline and budget constraints. The performance of the proposed algorithm is evaluated using synthetic datasets and compared with the state-of-arts existing algorithms. The main contributions of the work are summarized as follows.

• A comparative review of existing load balancing strategies along with their merits and demerits.

• Devised a new resource clustering strategy to cluster the servers based on their current loads and cluster centers.

• Proposed a new rule-based merging technique to merge the null and small clusters to minimize the number of clusters.

• Applied the Bat optimization technique to combine both the local and global search capabilities for finding the optimal cluster centers with faster convergence rate.

• Developed a new task assignment policy to minimize the makespan and execution cost.

• Finally, a comprehensive validation of the proposed algorithm via simulation runs using synthetic datasets.

The rest of the paper is organized as follows. Section 2 presents the related work. The overview of the Bat optimization technique is discussed in Section 3. The system model along with the problem statement are discussed in Section 4. The proposed algorithm described in Section 5. Performance evaluation of the proposed algorithm is discussed in Section 6 and concluding remarks are given in Section 7.