Dynamic provisioning of virtual machine using optimized bit matrix load distribution in federated cloud

1.1 Federated cloud

For completing the task of cloud providers, more resource management techniques needed. The QoS becomes conservative factor in such a way that providers delay in resource management strategies by reducing the number of resources allocated to requirements. The providers must be in the position to dynamically increase the virtual resources based on the server request to enable QoS by accepting the workload [3]. To enhance such scenario, establishment of cloud federation can be done for developing coordination between providers.

A cloud federation follows regulations for providers to allot resources which aims for reducing resource that available in the local infrastructure. This leads to rejection of client requests due to acceptance of requests to other members in the federation. More specifically, for cloud computing to properly deliver on its promise, there needs to be the technological means to federate various data centers, including those controlled by other companies. Infrastructure providers can only use their combined capacities to offer a seemingly limitless service computing utility through federation. When a request for a virtual machine (VM) is made on demand and non-availability of idle resources in the data centers, the federated cloud may decide to raise the spot price and stop providing the VM.

For resource provisioning, Openstack software platform [26] is utilized through data center in order to manage hardware pools of computing, storage, and networking resources. It is a cloud operating system that manages sizable computing pools and gives customers the ability to add resources.

The main objective is to allocate virtual resources to fulfil as many on-demand requests as feasible with respect to profit for providers. The proposed algorithm uses different strategies that facilitate decision-making choices hold by providers regarding user’s demand through federated cloud.

The rest of this paper is organized as follows. section 2, summarize the literature review. The system model with Cloud federation using bit matrix concept is described in Section 3. In Section 4, the policies for resource provisioning and bit matrix techniques are explained. Evaluation of proposed algorithm is explained in section 5. In section 6, experiments, implementation, evaluation and inference of the proposed model is given in section 7.

2.1 Fair resource allocation

Based on fair load distribution, the following works has been adopted in cloud environment for sharing resources under contract method in federated cloud. Jinlai Xu et al. [4] produces the method that depends on the resource allocation. cloud service providers are both buyers and sellers. The fair allocation of resources done through contract allocation auction that emphasize on maximizing revenue for a specific provider. The beginning of resource fair allocation is finalized with contract duration at the agreed-upon cost. The vendor is required to share the contractual resources within the allotted time frame. Experimental results proposed that job distribution is effective over the established contract period.

A.B.M Alam et al. [5] describes trust assessment in multi-cloud environments for improving QoS with resource allocation model. The main aim of the proposed approach is to allocate resource in a reasonable amount of time effectively with trust evaluation using genetic algorithm. The proposed service broker policy outlines the user’s SLA expectations. The broker then finds a suitable datacenter that offers the best performance and lowest cost for supporting specified user applications in the cloud. The primary goal of the proposed framework is to allocate requested user applications in the cloud at a lower cost. Evaluated outcome presents in large rise in user duties servicing at cheaper prices.

Jie Wei et al. [6] produces a creative strategy for allocating resources that selects datacenters based on latency. The approach classifies datacenters based on the K-Means for identifying delay using algorithm. The proposed approach measures datacenter latency and then selects appropriate datacenters based on latency. Once chosen, the federated cloud’s VMs are placed there using the Binary Quadratic Programming approach and made available to real-time application providers. This technique has been tested using real latency between datacenters, the price for spending virtual machines, and the expenditure of data transfer, and it has demonstrated a reduction in overall latency as well as financial advantage.

The approach of resource allocation done for IaaS that reduces delay and focused on resource elasticity [13]. The approach ensures system stability by periodical processing of overclocking. The servers balanced with delay dependent optimization that takes over the admission control for overclocking. The resource utilization from data center is occurred from cloud edge even in the demand of high server and volumes. The methodology for requesting routing and resource allocation is systematically made and simulation environment is verified by overclocking at maximum rate.

2.2 Energy and cost consumption

In the current environment’s polluted surroundings, awareness of energy consumption is quite important. Thus, a method that took into account both internal and external datacenter physical resources, as well as the associated power utilization, is described in Fagui Liu et al. [7]. The method that is being presented consolidates smart software by transferring virtual machines between non-green and green datacenters using energy-conscious operations in a federated cloud. The proposed method creates a matrix of destination and granularity to evaluate energy usage. Later it finds a path that is a collection of components, each of which the optimum green location at the time in a given period is identified.

The two metrics i) quickest method to analyze path for selecting a datacenter based on bandwidth availability ii) A bioinspired method for work scheduling and server selection within a datacenter is explained in Siqian et al. [8]. The major objective of the suggested methods aims to lower network costs for a virtual machine allocation on demand. The algorithm does two different types of sorting: a local sort that yields a list of servers that are ranked by their cost derived with an average computation. Second worldwide list of individuals, arranged by level of fitness. If the following server is not a better candidate for allocation, starting with the first server of the first chromosome, virtual machines are allocated.

In Divya et al. [10], a method for smart clouds and framework for parking predictions are suggested. The combination of IoVT based kernel and smart cloud is designed which is economically advantageous in terms of energy and the environment. It is supported to dynamically scale both up and down computer funds to manage a spike in demand. The proposed approach, which i) reduces the datacenter deployment per provider ratio and share, enables numerous providers to make the most use of compute resources. ii) Using aggregators to distribute the available energy. iii) Finally, choose more environmentally friendly, carbon-free engines when appropriate. A relevant issue about the energy consumption and cost utilization are dealt in Savita et al. [18]. In this approach, provider demands a minimum prize per VM and proves minimum utilization of power. The mechanism attempts to turning off idle nodes to save power ensures the providers to increase their profit over the prices to compete in the Cloud market [20].

2.3 Provisioning in intercloud environment

In Rashid et al. [9], describes a method for securing IoT medical data in Health Care Systems in terms of its storage on public cloud systems. proposed system will greatly help in efficient storage of IoT application medical data. The method uses Enhanced Role Based Access Control (ERBAC) provide secure storage of medical data in the cloud systems based on these role based access policies.

The inter cloud service execution requires integrated version of monitoring and verification. In Pooja et al. [11], the strategy of analysis done for minimal metric monitoring service that increases reliability in variety of private cloud infrastructure. The scheduled tasks estimate responsiveness with varying in time and load conditions. The presented results evaluated hardware characteristics, overall load and utilization of actual cloud providers that is in line with SLAs. In such cases, CPUS were under provisioned and it is reflected in the pattern of infrastructure.

InterCloud’s goals, and architectural components are detailed in [16]. The architecture that is being shown allows for application scaling across many cloud providers. There are three modules in architecture. Cloud Broker, Cloud Coordinator, and Exchange. Several cloud service providers and users are linked by Cloud Exchange. It utilizes the resource data that is currently available and supplied by cloud coordinators, it evaluates infrastructure requests received from brokers. Most of the algorithm focuses on maximizing IaaS provider’s profit that analyzes through service composition. In this model, optimal allocation strategy for decision making was proposed. This model includes parameters such as workload, Cost of outsourcing, the quantity of resources outsourced, and server maintenance costs.

2.4 Federated cloud based on SLA

Hariharan et al. [12] show the effects of federated cloud computing on SLA. A precise model for federated cloud SLA assessment has been developed, and this will be compared to non-federation systems. Two models are used to calculate the evaluation: first focuses on deterministic and second evaluation based on probabilistic. In Rejin et al. [17], data integrity for federated cloud architecture is implemented for ensuring benefits of SLA to meet user needs, problems with the federated cloud environment, as well as possible solutions, are examined. It has been argued that the development of federated clouds, which enable a variety of services to be offered and free organizations from uncertainty, is the greatest solution for all organizations. Previously, outsourcing issue with a data centre termination option for spot VMs are provided. Under such circumstances, comparison of two attempts can be identified [21].

2.5 Load balancing

A model that partitions the public cloud to balance the load has been developed [14]. Servers, balancers, and the primary controller are all included in the design. Each cloud’s status is indicated as overload, normal, or idle. The primary controller contacts all of the partition balancers when a task is added, and the work is then scheduled for a partition that is not overcrowded. Each partition’s balancer further determines the load on each server, and tasks are then scheduled on the server with the least load. This architecture increased resource availability, performance, and resource consumption by dynamically allocating work to the server that was least busy.

A new efficient VM load balancing algorithm [15] determines the anticipated response time for each VM. When a request is made, the algorithm notifies the data centre controller for VM with the fastest anticipated response time and schedules the request to that VM. The allocation count for the chosen VM is then increased by the algorithm. The data centre controller [19] reports the algorithm for VM deallocation when the service request is complete. The assessed outcome demonstrates that selecting VMs wisely can decrease average response times while also boosting cloud performance as a whole.

3.1 Cloud federation

Functional diagram of cloud federation processing framework in this work is shown in Fig. 1. The cloud federation is depending on provider who acts autonomously. Federation can help providers to envision overload and balance them with specific focus in on demand services. At the center of this model, there is a single pool supporting three interoperability techniques such as resource provisioning, data storage and combination of complementary resources.

OPEN IN VIEWER

Fig. 1

Federated cloud management.

Cloud coordinator can decide allocation of resources from different cloud service providers (CSP). The cloud coordinator is the component who estimates the amount and range of idling capacity for every supplier who exchanges with other members. The cost estimation done by providers based on demand basis. Federated cloud takes care of distributed data and workflow management. The federation of cloud resources enables a client to select the best cloud services provider to address a unique technological or business necessity inside their organization, taking into account flexibility, cost, and service availability. The agreements between federated members and providers are connected with profit to all its members. The cloud federation allows agreements as Federation Level Agreement (FLA) for allocation of resources. The mechanism deals with private information pertaining to the provider with regards of actual resources used by the providers. Considering the above scenario and predictions, resource allocation policy is proposed in the next section that identifies how various service providers decisions have an impact on revenue among the clients.

The main challenging task when providers allocate resources of their capacity in the form of spot VMs. A federated cloud provider must choose between raising the spot price and terminating spot VMs or outsourcing the request to another federation member. This would be done when it receives an on demand request for VMs when there are no available resources in the data center.

Federated cloud shares the workload between the CSP and determines interoperability between the two clients. Some of the major issues and challenges in processing federated cloud are as follows:

3.3 Resource provisioning

In a cloud management, when all the nodes are operative, provider demands a minimum price per VM and a minimum utilization. The proposed policy focuses on profitable utilization of local resources which is extremely preferred to external resources. However, if the workload cannot be supported locally, later it increases the provider’s profitability. Regarding the resource provision in cloud federation, most of the algorithm focuses on maximizing IaaS provider’s profit that analyzes through service composition. This system model refers a thorough analysis of a problem with practical advantages. The proposed policy is based on VM provisioning to improve the allocation of resources which aids in selecting the most practical choice based on the services that are available. The outsourcings to other providers are implemented for renting free resources or power-saving node shutdowns to idle nodes. This method won’t result in a decrease in the price of buying and hosting IT equipment. This envisions remain profitable when the providers had to raise their pricing, which reduced their ability to compete in the cloud market.

3.4 Optimal allocation of cloud resources

Using Virtualization, resources for networking and computing must be managed and optimized together. Since the data is dynamically stored and within networked clouds, to real substrate resources, this necessitates the provisioning of collections of virtual resources driven by users. The impact of creating need for virtual machines (VMs) that are preloaded with operating systems and application software must be maintained by the cloud consumer.

In this work, allocation of cloud resources is estimated the following aspects

a) This work proposes the idea of bit matrix, which is representation of the environment and current resource availability status generated by each individual cloud member when processing a user request.

b) Design an optimal allocation policy BMNF for load distribution reduced response times for requests, improved energy efficiency among various cloud members, and improved quality.

c) Analysis of BMNF policy’s performance and comparison of another existing algorithm.

4.1 VM allocation algorithm

In this part, the algorithm based on virtual machine allocation is maintained by controller components for optimal allocation. This will support various resource pool and group entities to aim to find hosts with a lot of workloads to boost utilization. Less hosts will be implemented, which will result in higher utilization. OpenStack compute is the element that makes it possible for users to create and control virtual servers using machine images. Large networks of virtual computers are configured and managed by OpenStack computing. In the first stage, based on the request of consumer, initialization of Host and VM was furnished then based on RAM and CPU utilization cost, the VM allocation is done.

In the second stage, when we request for allocation, it checks for reserved request and based on that utilization threshold values were compared. The comparison will result the minimum value which leads to allocation of VM to the reserved request. Either case if request is on demand, then resource provisioning will compare the neighbor cost as well as compute cost to allocate the VM. The general method followed in proposed algorithm has been shown in Fig. 2.

OPEN IN VIEWER

Fig. 2

VM allocation based on request.

In general, the necessary characteristics should be offered by an allocation method included in Section 4. The algorithm was done for each type of virtual machine that can scale without bounds. The providers are working as on-demand type who will process the allocation of Virtual Machine. If the allocation is not reserved then the operation will be handled till it reaches the first provider after they transmit it to the next request. since the allocation will be worked on their own benefits.

The proposed algorithm is an algorithm which aims for an ideal virtual resource distribution in a federated cloud. The algorithm produces cost estimation, load balancing, and enables high availability with higher profit.

Algorithm Optimal Resource Allocation Algorithm

Input: Availability of running VM list and host list

Input: Request of consumer

Input: Initialize host and VM

Output: VM allocation using Thresholdvalue

Globalvariable requestutilization, thresholdvalue;

// optimal resource allocation algorithm

For the proposed algorithm, the rules that have been suggested, each of which differs in how it addresses fresh on-demand demands that can’t be met by nearby resources.

A distinct web hosting application service is available on VM hosts, whose workload used a uniform distribution to construct and whose CPU utilization distribution is considered. Every time the creation of an instance to implement web-hosting service required 150000MIPS or it takes 10 min to complete execution with assumption of 100% utilization.

1) Effect of spot request: The experiment intended for assessing alterations in the ratio of on-demand queries may affect performance indicators.

2) Effect of persistent spot requests: This evaluation shows how adjusting the amount of persistent spot VM requests has an impact on the providers.

3) Effect of providers in the federation: with the ratio, it is ideal that the provider can handle the growing number of providers and also have an outsourced option which have less VMs available on demand. These are refused which make use of serving the outsourced request.

The challenges in cloud federation will accept trading idle resources for virtual resources or outsourcing decisions that are profit-driven. The strategy also considers the providers’ ability to turn off any data centre nodes that are not in use in order to conserve energy. However, they failed to take into account various VM types, or prospective measures like terminating low priority leases on-demand and spot [26]. Recently, the goal of utilizing spot VMs has drawn a great deal of focus. To achieve, an optimal allocation can be done with consideration of QoS as a level by approving more service request with respect to maximize cloud provider’s profit.

4.4 Methodology for estimating BMNF

FCM is the interface via which users interact with the federated cloud. An FCM has several linked clouds. The manager interacts with each cloud broker in response to a request before deciding which cloud is the best one based on a variety of performance criteria. Cloud Broker monitors a certain cloud, is in charge of placing VMs on the proper host, and gives consumers access to computing resources to fulfil their requirements. Users who require virtual infrastructure to fulfil their needs using FCM are the last segment.

The major three working components of FCM are broadly categorize as: i) examining the LDA of each cloud and dispersing the load over many clouds in an even manner. ii) Determining the best by examining the obtained bit matrix pattern in accordance with certain performance standards. iii) Managing the request scheduling and VM deployment in the chosen cloud.

The experiment was carried out by creating more than 1500 jobs of varying lengths, which were continually delivered to the FCM every 5 seconds, ranging from [5–50] requests. At each data centre, VMs are dynamically constructed during the simulation process to handle the incoming requests.

The application prototypes employ a federated cloud architecture, which is representative of a practical deployment configuration. The architecture consists of (i) the Cassandra-X [29] deployment arrangement, which includes two nodes of the Apache Cassandra cluster (stable version 3.11.1), one of which is active and operational, whilst the other node is inert (i.e., dormant). the standby mode; and (ii) the MongoDB-X deployment configuration contains a single node Service MongoDB (stable version) 3.4.9); (iii) the Cassandra-Y deployment configuration includes only one node Apache Cassandra (stable version) service 3.11.1); and (iv) a single node MongoDB service (stable version 3.4.9) is included in the MongoDB-Y deployment scenario.

CloudSim report toolkit [27], which consists of Netbeans IDE 12.5 [28] and CloudSim 3.0.3, has been utilized as the platform for the simulation experiment. To resemble the suggested algorithms, the classes of the CloudSim simulator have been expanded. This working model is developed in Java. A 64-bit Windows 10 computer with an Intel® CoreTMi7-3770 processor and for the test, 8GB of DDR3 SD RAM was used. Results from Round Robin and Distance Based load distribution are compared with those from the proposed BMNF model to show its efficacy and fairness.

In Fig. 3, the proposed BMNF estimates the percentage of load distribution among various cloud members at various time periods. It has been compared with round robin, and distance-based load distribution methodologies. Notably, the suggested BMNF model shows compared to the fair load allocation among all cloud members, avoiding unfair load distribution caused by the other two existing solutions for each time slot. The suggested BMNF model nevertheless upholds equity among various cloud members when load increases over time, according to a general trend that may be observed here. This is because, while choosing the cloud and allocating requests to it, FCM takes into account both the Bit Matrix status that it receives from the cloud members as well as the LDA at each cloud.

OPEN IN VIEWER

Fig. 3

Effect of load distribution at different time slots.

Figure 4 shows the average response times derived by three approaches for requests served at various clouds. The average response time for all three clouds in the proposed BMNF model is decreased by 9.4 seconds when compared to the Round Robin technique and by 9.6 seconds when compared to the Distance Based approach. Here broker works allotment of virtual machine with load balancing based on request and results in Predicted Completion Time (PCT). Figure 4 shows that, as compared to current methods, each cloud’s average response time to queries is considerably low. This is so that managers may choose the best cloud. Every cloud broker creates a Bit Matrix to show if requests can be fulfilled within a specified time frame either by launching a new VM or using an already-existing VM. Additionally, the broker chooses VMs within each cloud that offer the bare minimum PCT.

OPEN IN VIEWER

Fig. 4

Response time of requests.

The average request waiting time is illustrated in Fig. 5. Compared to the other two methods that are currently in use, the proposed BMNF model has a shorter waiting time. Due to bit matrix, FCM can find a cloud that fulfills the request on time. Additionally, a cloud’s broker assigns requests to virtual machines (VMs) with the lowest PCT.

OPEN IN VIEWER

Fig. 5

Waiting time of requests.

A random matrix with the same dimensions to study the effectiveness of the dynamic cost-based optimization strategy has incurred. The skew values illustrate different plan with similar executions. The extensive computation costs and its waiting time will be generated using random matrix computation.

In comparison to the other two load distribution approaches, the suggested model’s reduced rate of SLA violation is shown in Fig. 6. Because each request is processed immediately after receipt, there are fewer SLA violations now. FCM determines a cloud that focuses and fulfil the workload on an active VM by the deadline given or alternatively builds a VM on an ideal cloud. As a result, the majority of requests are fulfilled on time and in accordance with SLA.

OPEN IN VIEWER

Fig. 6

Percentage of SLA violaitions.