Over the past decade, Cloud computing (CC) is emerging as a prominent paradigm and its usage had witnessed considerable development [1]. Small scale users and largescale commercial businesses and scientific applications get benefitted from using cloud. By the use of minimum involvement, clients can benefit service from CC since it allows universal and on demand access to a common pool of CC resources which can be hardware, software, and applications are communal resources. Fault might take place on each of these layers; nonetheless, software enabled algorithm is recognized and employed to recuperate from faults. Fault tolerance is defined as a capability of a system to keep on running its planned operation against faults or errors [2,3]. Even a well-developed system with the highest services and components could not be named reliable with non-fault tolerant abilities [4]. For the reason that a great amount of delay aware applications have to be performed, reliability is a crucial feature of CC. Moreover, service reliability is crucial to cloud wide recognition. Consequently, fault tolerant approaches have gained considerable attention from the researcher. There are different fault tolerance models— checkpointing, Reconfiguration, replication, Safety-Bag Checks, Task Migration, Retry, Task Resubmission, Self-Healing, Masking, and so on. [5–8]—to address the fault at different levels whether reactively or proactively. CC includes the dynamic resource allocation and the usage of data center that is often distributed geographically. The splitting of the server accessible resource into virtual machine (VM) take place by VM monitor (VMM). Multiple or individual VM are allocated for running the provided applications based on client requests. The advantage of using a VM is that it enables us to perform applications on different IDEs, software environments, and operating systems. Fault tolerance based on hardware or time redundancy to cover component or task failure. Time redundancy includes the re-execution of failed task afterward the malfunction was recognized. It is enhanced by the application of checkpoint technique however, still it imposes a significant latency. These two technique presents negative and downside aspects, replication needs additional resources and re-execution need additional time, especially energy. This makes a tradeoff between hardware and time redundancy, in IaaS-CC system replication is preferred mainly since the response time is usually very vital [9,10]. This study focuses on the design of Gorilla Troops Optimizer Based Fault Tolerant Aware Scheduling Scheme (GTO-FTASS) in CC environment in order to schedule the tasks and allocate resources by considering fault tolerance into account. The GTO-FTASS algorithm is based on the social intelligence nature of gorilla troops. Besides, the GTO-FTASS model derives a fitness function involving two parameters such as expected time of completion (ETC) and failure probability of executing a task. In addition, the presented fault detector can trace the failed tasks or VMs and later schedule healing submodules in sequence with a remedial or retrieval scheduling model. The experimental validation of the GTO-FTASS model has been performed and the results are inspected under several aspects.

This section reviews the recently developed models for task scheduling with fault tolerance in CC. Kasu et al. [11] presented a bloom filter enabled data-aware probability based fault tolerant (DAFT) method that might deal with these faults. Also, presented a data and layout aware method for fault tolerance (DLFT) to efficiently deal with the false positives match. Kanwal et al. [12] presented genetic algorithm (GA) based multi-phase fault tolerance (MFTGA) method for scheduling the task through the VM for multiple users. The presented method effectively maps optimum VM with users based on the service level agreement (SLA). The proposed technique encompasses four stages such as local phase, individual phase, fault tolerance phase, and global phase. Sathiyamoorthi et al. [13] present an adaptive and effective fault tolerant scheduling method to help error-free scheduling. The presented technique considered the highly effective variables namely present workload and failure rate of the resource for optimum Quality of Service (QoS). The presented technique can be validated by utilizing the CloudSim toolkit. Jing et al. [14] proposed a QoS scheduling algorithm by integrating the cloud features, later, we developed a task scheduling object to guarantee fault can be tolerated at the time of the task execution. At last, we presented a QoS-aware scheduling approach, QoS-DPSO, for satisfying the QoS requirement in CC system. Alaei et al. [15] presented an adaptive fault detector approach based Improved Differential Evolution (IDE) method in CC that could reduce the energy utilization, the make-span, overall costs, and, simultaneously, tolerate up fault while scheduling scientific tasks. This presented method employs an ANFIS predictive method for proactively controlling resource load fluctuation which improves the fault predictive performance beforehand of failure or fault existence. Karthikeyan et al. [16] established a SALDEFT model for improving the communication overhead (that is, overall energy utilization and the network resource). Now, choosing an optimum Physical Machine (PM) is taken into account as an optimization issue and an enhanced DE approach is applied to resolve the issue. In [17], proposed an approach to resolve allocating tasks through Ant Colony Optimization (ACO) by adopting fault tolerance and Reinforcement learning (RL) for making fault-tolerant scheduling process, and to accomplish the goal of minimal makespan. Arora et al. [18] presented the method to handle various requests from multi-user to the cloud. It contains two parts: initially, each VM is monitored and the optimum VM with adequate resources to manage the request is carefully chosen and next, the task is assigned to the essential resource. Also, check the load on every VM by continuously observing, and once the load is improved, relocation can be performed to the following VM. In the next section 3, discusses proposed model, followed in section 4 experimental evaluation section, the final section discusses conclusion with future findings.

This study has designed a new GTO-FTASS technique for scheduling tasks and allocating resources in the CC environment. The GTO-FTASS model has computed a fitness function comprising two parameters such as ETC and failure probability of executing a task. In addition, the presented fault detector can trace the failed tasks or VMs and later schedule healing sub-modules in sequence with a remedial or retrieval scheduling model.

In this section, the experimental validation of the GTO-FTASS model is performed using two scenarios. In the first scenario, a set of 5 cloud users with 5 brokers and 2 data centers are involved. Next, in the second scenario, a set of 10 cloud users with 10 brokers and 5 data centers are involved. The results are investigated under varying numbers of tasks. Tab. 1 provides a comprehensive comparative examination of the GTO-FTASS model with existing models [20] on scenario 1 in terms of makespan (MSN), failure ratio (FRR), and failure slowdown (FSD). The experimental values indicated that the GTO-FTASS model has resulted in an effectual outcome on scenario 1.

Fig. 3 illustrates a brief MSN examination of the GTO-FTASS model with recent models under varying tasks in scenario 1. The figure indicated that GTO-FTASS model has resulted in effectual outcomes with minimal values of MSN over the other methods. For instance, with 10 tasks, the GTO-FTASS model has offered lower MSN of 409 whereas the MTCT, MAXMIN, ACO, NSGA-II, and DCLCA models have obtained higher MSN of 636, 622, 594, 580, and 537 respectively. Besides, with 50 tasks, the GTO-FTASS model has offered lower MSN of 466 whereas the MTCT, MAXMIN, ACO, NSGA-II, and DCLCA models have obtained higher MSN of 1359, 1118, 948, 806, and 622 respectively. Along with that, with 100 tasks, the GTO-FTASS model has provided reduced MSN of 721 whereas the MTCT, MAXMIN, ACO, NSGA-II, and DCLCA models have resulted in increased MSN of 3075, 2508, 1955, 1530, and 991 respectively.

Fig. 4 exemplifies a detailed FRR inspection of the GTO-FTASS model with recent models under variable tasks in scenario 1. The figure designated that GTO-FTASS model has accomplished better results with least values of FRR over the other methods. For instance, with 10 tasks, the GTO-FTASS model has provided minimal FRR of 43.42 whereas the MTCT, MAXMIN, ACO, NSGA-II, and DCLCA models have gained maximum FRR of 63.34, 53.38, 55.51, 50.53, and 50.53 respectively. Meanwhile, with 100 tasks, the GTO-FTASS model has accomplished decreased FRR of 11.41 whereas the MTCT, MAXMIN, ACO, NSGA-II, and DCLCA models have depicted increased FRR of 24.92, 22.08, 23.50, 18.17, and 17.46 respectively. Additionally, with 50 tasks, the GTO-FTASS model has provided lesser FRR of 22.43 whereas the MTCT, MAXMIN, ACO, NSGA-II, and DCLCA models have attained increased FRR of 38.44, 34.53, 34.88, 33.46, and 30.26 respectively.

Fig. 5 demonstrates a comprehensive FSD investigation of the GTO-FTASS model with recent models under changeable tasks in scenario 1. The figure reported that GTO-FTASS model has led to superior performance with lower values of FSD over the other methods. For instance, with 10 tasks, the GTO-FTASS model has accomplished reduced FSD of 0.54 whereas the MTCT, MAXMIN, ACO, NSGA-II, and DCLCA models have demonstrated increased FSD of 1.91, 1.47, 1.50, 1.50, and 1.08 respectively. Besides, with 50 tasks, the GTO-FTASS model has reached minimal FSD of 1.76 whereas the MTCT, MAXMIN, ACO, NSGA-II, and DCLCA models have obtained higher FSD of 3.04, 2.96, 2.87, 2.62, and 2.57 respectively. Along with that, with 100 tasks, the GTO-FTASS model has provided reduced FSD of 2.74 whereas the MTCT, MAXMIN, ACO, NSGA-II, and DCLCA models have resulted in increased FSD of 6.19, 4.55, 4.14, 3.84, and 3.53 respectively.

Tab. 2 offers a detailed comparative study of the GTO-FTASS model with existing models on scenario 2. Fig. 6 illustrates a brief MSN examination of the GTO-FTASS model with recent models under varying tasks in scenario 2. The figure indicated that GTO-FTASS model has resulted in effectual outcomes with minimal values of MSN over the other methods. For instance, with 10 tasks, the GTO-FTASS model has offered lower MSN of 1772 whereas the MTCT, MAXMIN, ACO, NSGA-II, and DCLCA models have obtained higher MSN of 2842, 3043, 2742, 2909, and 2341 respectively. In line with this, with 100 tasks, the GTO-FTASS model has provided reduced MSN of 721 whereas the MTCT, MAXMIN, ACO, NSGA-II, and DCLCA models have resulted in increased MSN of 8828, 6353, 5852, 6253, and 4347 respectively.

Fig. 7 depicts a clear FRR analysis of the GTO-FTASS model with existing approaches under dissimilar tasks in scenario 2. The results portrayed that GTO-FTASS model has reached improved performance with reduced values of FRR over the other methods. For instance, with 10 tasks, the GTO-FTASS model has reached inferior FRR of 43.82 whereas the MTCT, MAXMIN, ACO, NSGA-II, and DCLCA models have displayed enhanced FRR of 69.69, 65.89, 63.82, 51.75, and 53.82 respectively. Eventually, with 100 tasks, the GTO-FTASS model has provided reduced FRR of 5.19 whereas the MTCT, MAXMIN, ACO, NSGA-II, and DCLCA models have resulted in increased FRR of 35.54, 30.02, 26.23, 22.44, and 11.75 respectively.

Fig. 8 explains a brief FSD assessment of the GTO-FTASS model with existing approaches under varying tasks in scenario 2. The experimental outcome revealed that GTO-FTASS model has accomplished better results than the other methods with minimal values of FSD over the other methods. For instance, with 10 tasks, the GTO-FTASS model has provided least FSD of 0.65 whereas the MTCT, MAXMIN, ACO, NSGA-II, and DCLCA models have gained increased FSD of 3.58, 2.51, 2.30, 1.77, and 1.48 respectively. Similarly, with 100 tasks, the GTO-FTASS model has provided reduced FSD of 6.05 whereas the MTCT, MAXMIN, ACO, NSGA-II, and DCLCA models have resulted in increased FSD of 9.59, 9.22, 9.01, 8.11, and 7.78 respectively.

After examining the above mentioned tables and figures, it is apparent that the GTO-FTASS model has resulted in effectual scheduling performance in the CC environment.

This study has designed a new GTO-FTASS technique for scheduling tasks and allocating resources in the CC environment. The GTO-FTASS model has computed a fitness function comprising two parameters such as ETC and failure probability of executing a task. In addition, the presented fault detector can trace the failed tasks or VMs and later schedule the healing sub-module in sequence with a remedial or retrieval scheduling model. The experimental validation of the GTO-FTASS model has been performed and the results are inspected under several aspects. The extensive comparative analysis reported the better outcomes of the GTO-FTASS model over the recent approaches. As a part of future scope, the GTO-FTASS technique can be applied as an effective tool for accomplishing effective scheduling and fault tolerance in CC environment.