Workflow scheduling and reliability improvement by hybrid intelligence optimization approach with task ranking

Workflow scheduling is one of the most challenging tasks in cloud computing. It uses different workflows and quality of service requirements based on the deadline and cost of the tasks. The main goal of workflow scheduling algorithm is to optimize the time and cost by using virtual machine migration. This algorithm computes the subset problem and decision problem in NP time. It works on the decision-making process to reduce the time and cost of computation on the server side. This paper proposes hybrid optimization to optimize the virtual machine locally and globally. The PEFT algorithm is used for initialization and worked as a heuristic algorithm. This algorithm reduces the error of random initialization of optimization. The proposed algorithm based upon Flower pollination with Grey Wolf Optimization using hybrid approach shows significant end effective results than flower pollination with genetic algorithm. The proposed approach also considered the reliability parameter on different workflows.


Introduction
Cloud computing is a platform which provides the virtualized access to the services and application to the clients.The physical location of the client it doesn't matter in the cloud computing services; the user can access the services and resources from anywhere and anytime.Sometimes our system is hanged up due to a large number of processes in the queue in which resources are acquired by some process.This process is also similar to the cloud and enhances the load on the cloud [1].The load balancing on the cloud is important issues in cloud computing which affects the performance, reliability and scalability of the cloud.Load balancing is basically a process in which load is distributed among all nodes so that each node works efficiently and the performance of the cloud does not degrade.Load Balancing can be done by scheduling task, propose resource allocation and task migration.Load balancing is done by using the device called Load Balancer [26].It increases the capacity and reliability of the applications.It also improves the performance of the system by dividing the burden equally.Load Balancer works on the two layers that are network layer and transport layer.It assigns the data to the servers according to the data found in the application layer [2].Load balancing in the cloud is mainly used in the cloud to the proper allocation of resources, reduce the resources consumption and maximize the throughput with minimum response time.On the basis of the systemstate, load balancing is divided into two types that are Static load balancing and Dynamic load balancing [3].Static load balancing is needed when there is no variation in load.Task allocation or load distribution depends on the load at the time of selection of a node or based on the average load of the server.Performance of the server is taken into consideration for the selection of the server.This work is assigned on the basis of the incoming time of the job, execution time and inter-processes communication.
Dynamic load balancing is performed at the runtime.Distribution decisions are based on the current load information.In this balancing scheme, any advance information is not available.It works on the current situation and load balancing.It also works on the heterogeneous system's communication plays an important part to improve the performance of the system.It is also divided into two parts distributed and Non-distributed [4].

Motivation
The concept workflow scheduling is considered as the projecting issues in the mechanism of workflow scheduling which attempts to outline work process undertakings to the VMs dependent on various useful/functional and nonpractical/non-functional necessities.Despite the fact that few planning concerns have been broadly examined, for example, the vulnerabilities delivered by framework failure, PC heterogeneity, asset versatility, cutoff time requirements, and budget restrictions among others, still there are some different uncertainties that have not pulled in the consideration they justify.These workflow or jobs will register with less time and less assets.Separately, it computes the process in a right manner and do not overlap with one another.This paper addressed a new meta-heuristic hybrid algorithm namely Flower Pollination with Grey Wolf Optimization (FPA_GWO).This algorithm has been implemented using different Scientific Workflows datasets like Sight, Ligo, and Genome and in most of the cases it gives better results.

Literature Survey
For solving the issue of load balancing we need some effective algorithm which effectively balances the load.The ant colony optimization algorithm is used for the load balancing of the cloud.This is a Metaheuristic algorithm and provides the optimal solution to the problem [26].The response time of the resources is optimized by the ACO by distributing the load dynamically [5].To distribute the load efficiently on the cloud fuzzy row penalty method is presented in [6] for cloud computing.The fuzzy approach is used to solve the problem of uncertainty in the response time in cloud resources.The fuzzy row penalty approach is used for both balanced and unbalanced fuzzy load on the cloud.The response time and space complexity are two parameters on which performance is measured.The main goal of load balancing is to distribute the equal amount of load to each processor on the cloud so that the performance and scalability of the cloud don'taffect by more load.The hybrid optimization algorithm is used for the optimization of a complex network of the cloud.The hybrid Ant Colony Optimization (ACO) algorithm is used for an optimal solution [7].The load balancing is also based on the dynamic threshold in cloud computing.This is done by organized the virtual resources of data centers efficiently.
The proposed approach reduces the makespan and enhances the resource utilization with minimum usage of energy [8].To find the best virtual machine on the cloud by considering the minimum makespan and power resource utilization [33].The effective virtual machine is selected by using a chaotic Spider algorithm and it tackles the problem of task scheduling.The overall makespan during the scheduling is minimized by using a weight-based random selection method.The best virtual machine is finding out by using global intelligent searching [9].S. Chenthara et al., proposed an ensure privacy and security of electronic health records (EHRs) in the cloud.The main storage for the data whether it is related to healthcare, computational based data, scientific calculation based data etc is cloud servers [27] [30].In this paper author highlights privacy aspects of data and address the cryptography and non-cryptography security techniques.Hua Wang et al., proposed Role-base access control(RBAC) algorithm which implement access control and its policies, authorization, grant and revoke permission to access the outsource data with security in cloud [29].Md Enamul Kabir et al., proposed micro aggregation method to secure the microdata of the cloud computing [28].Kang, Seungmin et al. addressed the issue of scheduling in multi-cloud systems.Prediction technique is used to deal with the issue of uncertainty of node.Dynamic scheduling strategy is proposed for multi-cloud systems.The total processing time of loads is minimized by using scheduling techniques.It considers the availability and heterogeneity of computing nodes.By using the proposed DSS method, divisible load theory and node availability give high performance [10].Ajay et al., presented five different met heuristic algorithms and compare their performance and express the performance using different benchmark functions [32].The author also proposed meta-heuristic optimized GACO algorithm for load balancing in cloud environment.In this paper author compare the performance using 17 different benchmarks functions.The results of proposed algorithm show better result than existing GA, PSO, ABC algorithms [34].

Research Problem
The logical work process scheduler details the planning issues as a weighted optimized issue of two targets (runtime and monetary cost).It allocates tasks to VMs to limit execution time and money related cost dependent on client necessities.To figure this issue, the considered undertakings or tasks are disseminated among the queues of VM, .A queue (line) is characterized as disintegration of a set into incoherent (disjointed) subsets whose association is the primary set [31].In light of this model, the problem of scheduling is characterized as finding the relating components of each VM queue to expand the accompanying augmented (amplified) objective function represented as follows: Since the owners of the application do not have devices to precisely measure complete execution time or on the other monetary cost related expense for their work processes, our methodology offers a unique and adaptable approach to pick a specific configuration of scheduling driven by and presenting the time of execution and the monetary-based cost optimization, where the weights reply over the following condition: Thusly, the owner of the work process provides a weightbased percentage to constraints dependent on the need.

Proposed Approach
Hybrid Intelligence Optimization Approach is used for scheduling of the tasks and reliability.In this paper Flower pollination algorithm (FPA) and Grey wolf optimization (GWO) algorithms are used as hybrid using PEFT algorithm for global and local optimization.The details of algorithm are following:-FPA: Flower Pollination Algorithm is based on the concept of the transferring of pollen from one plant to other.The pollens transfer agents are mainly insects, bats and birds.Pollination process is divided into two parts that are Biotic Pollination and abiotic pollination.The most common pollination is biotic pollination and it occurred 90% in the flowers [11].The abiotic pollination occurs 10%.The biotic pollination needs a pollinator agent to complete the pollination process but abiotic pollination does not need an agent to complete the pollination process.In pollination, process insects go to different flowers and insects bypass some species of flower and this process is called as Flower consistency [12].Flower pollination process is classified into two parts that are cross-pollination and Self-pollination.In crosspollination, process pollens are transferred to different plants by the pollen agents.In the self-pollination process pollens of the same flower is responsible for fertilization.The cross and biotic pollination are called as global pollination and follow the Levy Distribution.The self and abiotic pollination are called local pollination [14,15].
The reproduction ration can be computed by using the consistency of flower and it is proportional to the degree of similarity between two flowers.

Due to physical conditions like wind local pollination has an advantage over global pollination
The General steps followed by the Flower Pollination Algorithm are the following: I. Minimize or maximize the objective function and then create the random population with a specific size.
II. Identify the best solution and then select the pollination process among cross and self-pollination.
III.After this apply Levy Distribution and global pollination approach and calculate the fitness function.
IV. Replace old solution if the new solution is better and finds the current best position for the next generation.
PEFT: Predict Earliest Finish Time (PEFT) is a scheduling algorithm which is based on the list and worked on the bounded on the number of heterogeneous processors.This algorithm works in two phases in which the first related to Task Prioritization (for computing task prioritization) and the second phase is Processor selection in which best processor is selected for executing the current task [16].
GWO: Grey Wolf optimization algorithm is a bio-inspired algorithm which is based on the leadership and hunting behavior of the wolves in the pack.The grey wolves prefer to live in the pack which is a group of approximate 5-12 wolves.In the pack, each member has social dominant and consisting according to four different levels.The belowgiven figure shows the social hierarchy of the wolves which plays an important role in hunting [17].
1.The wolves on the first level are called alpha wolves (α) and they are leaders in the hierarchy.Wolves at this level are the guides to the hunting process in which wolves seek follow and hunt and work as a team.Decision making is the main task that is performed by the alpha wolves and order by the alpha wolves is followed by all members of the pack [18].2. Second level wolves are called beta (β).These wolves are called subordinates and advisors of alpha nodes.The beta wolf council helps in decision making.Beta wolves transmit alpha control to the entire packet and transmit the return to alpha.3. The wolves of the third level are called Delta wolves (δ) and called scouts.Scout wolves at this level are responsible for monitoring boundaries and territory.The sentinel wolves are responsible for protecting the pack and the guards are responsible for the care of the wounded and injured [19].4. The last and fourth level of the hierarchy is called Omega (ɷ).They are also called scapegoats and they must submit to all the other dominant wolves.These wolves follow the other three wolves.

Results
This section presents the results on the different workflows of the cloud that are LIGO, Genome, and Sight.The results analysis is done on the two parameters that are Cost and Time.

 The decision of Balancing Time of computation and Cost
This point is very challenging in this research because when the cost is increased the number of VM is also increased for fewnumbers of tasks and workflow at the time of computation.The waiting time is reduced but the cost is enhanced.So, the decision of optimization playsa vital role to balance the VM and time of computation and needs the global and local optimization.For both local and global optimization algorithms like particle swarm optimization (PSO) [22], Ant colony optimization (ACO) [24] is used but they increase the time because of two optimizations.In this work, hybrid optimization is used in which one is used for local and meantime other for global according to local optimization output.

Reliability
In general, the concept explained above can be represented verbally by an eminent term 'reliability', which usually refers person's quality or entity that is trust worthy.Trust in the informational society is generally built on several distinct grounds on the basis of knowledge, calculus, or on the basis of social reasons [25].In an organization, the notion of trust could be defined as the certainty of the customer such that the organization is able to provide the needed services infallibly and accurately.The principle of certainty that also helps in expressing the customer's faith in soundness of its operation, in its expertise, its moral-based integrity, in the mechanism of security-based effectiveness, and in its abidance by the overall laws and regulations, on simultaneous basis, it also carries the affirmation of a minimized risk factor, by the party-based relying.

Conclusion
The proposed work is based on the effective hybrid optimization approach that is a combination of FPA and GWO algorithm.In this work, two parameters are considered for scheduling that is cost and time.In this algorithm FPA is used because it is a local optimization algorithm and GWO is a global optimization algorithm which optimized the VM globally.On the other hand for comparison FPA_GA algorithm is used which is also a swarm-based algorithm.In this work, when the cost is increased the number of VM is also increased for a few numbers of tasks and workflow at the time of computation.The waiting time is reduced but the cost is enhanced.So, the decision of optimization plays a vital role to balance the VM and time of computation and needs the global and local optimization.For both local and global optimization algorithms like particle swarm optimization (PSO), Ant colony optimization (ACO) is used but they increase the time because of two optimizations.

EAI
Endorsed Transactions on Scalable Information Systems 10 2019 -01 2020 | Volume 7 | Issue 24 | e7The variables , , , and are updated continuously at the time of workflow scheduling procedure mirroring the worst and the best configurations of VM's.These values generally represent the minimum and maximum values of monetary cost and runtime.Runtime is characterized as the most extreme time taken by the least or slowest incredible VM to implement the present queues of occupations, communicated as: Here, represents the execution time to execute on Monetary or cost is generally defined as the summation of runtimes of each and every VM multiplied with its corresponding cost, represented as:

Algorithm FPA_GWO Step 1 :Step 2 :Step 3 :Step 4 :Step 5 :CF=
Input Population, maxiteration, the transmission coefficient Initialize by PEFT Ranking.Initialize the Flower Pollination algorithm for the best solution and global Check the convergence.If converged the scheduling threshold otherwise initialize the Grey Wolf Optimization Algorithm.Calculate fitness function for every search agent best search agent  second beat search agent  Third best search agent While (T<Max iterations) For ( in every pack) Update current position of wolf by Eq. (1) Update x, X and Y Calculate the fitness function for all search Reliability Improvement by Hybrid Intelligence Optimization Approach with Task Ranking EAI Endorsed Transactions on Scalable Information Systems 10 2019 -01 2020 | Volume 7 | Issue 24 | e7

Figure 3 .Figure 3
Figure 3. Cost on LIGO workflow for FPA_GWO and FPA_GA Figure 3 depicts the cost of LIGO workflow for FPA_GWO and FPA_GA.The green curve shows the cost of FPA_GA and the red curve shows the cost of FPA_GWO.The cost of proposed FPA_GWO is less than FPA_GA.In Figure 3 comparison of flower pollination algorithm with genetic algorithm (FPGA) and flower pollination algorithm with grey wolf optimization (FPAGWO) on Cost parameter.The cost analysis is done

Figure 4
Figure 4 depicts the time of LIGO workflow for FPA_GWO and FPA_GA.The green curve shows the cost of FPA_GA and the red curve shows the cost of FPA_GWO.The time of proposed FPA_GWO is less than FPA_GA.

Figure 4 .Figure 5 .Figure 5
Figure 4. Time on LIGO workflow for FPA_GWO and FPA_GA Selecting Proposed algorithmIn this paper, the hybrid optimization is used because of the reasons mentioned in the above-given section but to decide which algorithm will hybrids discussed in this section under this point.First of all, two optimization algorithms that are based on bio-inspired and swarm intelligence are considered in which FPA is hybrid with GWO.In FPA_GWO algorithm, FPA is used because it is a Local optimizer and locally optimizes the task in VM and GWO global optimizer which take the decision of VM migration.In other hand use, FPA_GA is also swarm based and bio-inspired method.Results with Genome

Figure 6 .Figure 6 Figure 7 .
Figure 6.Cost on Genome workflow for FPA_GWO and FPA_GA Figure 6 depicts the cost of Genome workflow for FPA_GWO and FPA_GA.The green curve shows the cost of FPA_GA and the red curve shows the cost of FPA_GWO.The cost of proposed FPA_GWO is less than FPA_GA.Results with Sight

Figure 7
Figure 7 depicts the cost of Sight workflow for FPA_GWO and FPA_GA.The green curve shows the cost of FPA_GA and the red curve shows the cost of FPA_GWO.The cost of proposed FPA_GWO is less than FPA_GA.

Figure 8 .Figure 8
Figure 8.Cost of Sight workflow for FPA_GWO and FPA_GA Figure 8 depicts the Sight of Genome workflow for FPA_GWO and FPA_GA.The green curve shows the cost of FPA_GA and the red curve shows the cost of FPA_GWO.The time of proposed FPA_GWO is less than FPA_GA.

Figure 9 .
Figure 9.Comparison of Reliability in different workflows