Heuristic and fair-queuing based VM load balancing strategy for cloud data centers: A hybrid approach

Abstract

Cloud computing is an integrated service, which provides a new versatile of programming paradigm with a good number of features to dynamically migrate virtualized computing services between physical servers of various cloud data centers. This technology made the users and cloud service providers (CSPs) to manage the loads as per the demand. The effective management of loads in the cloud environment gives a rise to introduce new management methods, which are responsible for equally distributing the incoming workloads among all the available VMs in an effective way for a given period of time. To achieve this requirement various schemes have been proposed. However, many of them are not suitable to find the optimal information about required resources to fulfil the user demands. To accommodate this, in this paper, an approach named heuristic and fair-queuing based VM load balancing (HFQ-LB) has been introduced. With the help of fair-queuing, an efficient strategy has been designed. This configures the number of incoming loads for finding an appropriate VM for the assignment to satisfy the goal, i.e., it maximizes the utilization of resources according to the user demand. With the help of CloudSim, the proposed algorithm is validated in terms of makespan, waiting time, CPU utilization, and VM utilization by creating a multiple number of data centers. The proposed algorithm is also compared with other existing eight algorithms. Experimental results show that the proposed algorithm outperforms these existing algorithms in terms of makespan, waiting time, CPU utilization, and VM utilization.

Keywords

Resource utilization resource management load balancing heuristic rules fair-queuing

1. Introduction

Cloud computing is a new kind of epoch for the computing techniques, which emanated as a new stage for the distributed computing system. It provides the required resources to the end users as per their demand using pay-per-use model [1, 2, 3, 4]. In this manner, the cloud can be described as a type of distributed computing system in a virtualized environment, where the resources are provisioned dynamically [5]. According to Berkeley, the cloud can be stated as: “Cloud manages all of the essential applications which are mainly delivered in the form of infrastructure and platform at the data center through the internet that provide services to the end users. The services are itself known as Software-as-a-Service (SaaS)” [6]. Usually, cloud acts as a repository where the data and the information are stored, accessed, and successfully delivered to the end users. Foster presented a scenario in the form of a grid environment entity with cloud [7]. Cloud computing basically termed in three types of service models, namely: Software as a service (SaaS), Platform as a service (PaaS) and Infrastructure as a service (IaaS). The main objective of using IaaS is to increase the resource utilization by following SLAs, which ensures QoS to the end users. On the other hand, to fulfil the requirement of loads (i.e., number of jobs), cloud uses the services of IaaS, which provides a platform in a virtualized manner by creating an infinite number of VMs that fulfills the requirements of loads successfully within a reasonable amount of time [8, 9]. So that, the load of the end users can be processed by the cloud data center efficiently. The requirements for efficiently managing the loads are; resources can be managed efficiently and scheduled for the incoming loads in an intelligent manner so that all the requirements are successfully achieved by the dynamic provisioning of these computing services through the fair allocation of VMs. To assure the on time execution of the computational resources, the cloud service providers (CSPs) integrate the various related data centers that must fulfil the user demand in an efficient way. In order to maintain the SLAs without any violation, the CSPs require efficient load balancing strategies, where the load(s) can be further divided and scheduled by suitable VMs [10]. In order to scale the utilization power of VMs, the American society calculated the infrastructural cost as well as the load balancing costs, which are significantly decreasing from 75% to 25% of the overall operating cost of a VM [11, 12]. In order to handle this problem, resources must be utilized in an efficient manner. To accommodate these instances, in this paper, our focus is to handle the balancing approach in an efficient manner. The main objectives of load balancing approaches are: (a) to maximize the CPU utilization, (b) to minimize the execution time, and (c) to minimize the overall completion time of loads which are assigned to the appropriate VMs.

However, an improper assignment of loads may lead to imbalance of VMs among the various kinds of homogeneous as well as heterogeneous data centers. Hence, creating an efficient load balancing scheme becomes very difficult from CSP side to improve the system performance, CPU as well as resource utilization. Thus, it becomes crucial for developing such approach, which can efficiently improve the overall utilization of resources by effectively balancing the loads [13, 14, 15]. In the recent time, many algorithms such as Two Level scheduling Algorithm (TLSM) [50], LBACO Algorithm [51], HBB-LB Algorithm [52], TS-QoS Algorithm [53], LCFP and SCFP Algorithm [54], LB-BC Algorithm [55], Hyper-Heuristic Scheduling Algorithm [56] and ACO [57] etc., have been designed to solve the above mentioned issues. However, these algorithms did not find the potential information about the loads to manage the resources efficiently. In order to handle the above consequences, in our proposed load balancing algorithm, the number of incoming loads and the size of those respective loads both play an important role in balancing the loads. Another important key factor of our proposed methodology is that it fairly allocates the VMs for achieving an efficient resource utilization mechanism.

In this paper, a load-balancing approach is proposed for cloud data centers based on heuristic rules and fair queuing algorithm for fairly allocating the VMs. The proposed algorithm has two phases. In the first phase, VMs are allocated with the help of heuristic rules and fair queuing algorithm and in the second phase loads are mapped in between the incoming loads and the VMs. On the other hand, we also plan to design an efficient strategy for fairly configuring the VMs based on the availability of incoming loads. By following this strategy, the efficient utilization of resources as well as the CPU has been achieved. The objectives of this paper are stated as follows;

•
A load balancing approach is designed for the data centers in the cloud environment which is more convenient and suitable for balancing the loads.
•
Such an efficient scheme is implemented, which can successfully configure the entire system based on the incoming loads and their respective sizes. This strategy is really helpful inefficiently utilizing the computing resources as well as the CPU.
•
A new Load-VM mapping approach has been developed for minimizing the waiting time by efficiently scheduling the incoming as well as waiting or remaining loads efficiently.

The roadmap of this paper is as follows. Section 2 depicts the related works. Section 3 explores the problem formulation, notations and definitions. Section 4 presents the proposed work along with the system model and algorithms. Section 5 presents the result analysis part of the proposed approach and Section 6 concludes the methodology with future scope.
2. Literature review

This section describes some works proposed and designed for load balancing in a cloud environment. A methodology named host load prediction is designed based on the concept of Bayes model [16]. It was first invented by Zhao et al., where few of them were dealt with VM assignment and SLA violations [43]. According to the strategy developed in [17, 18], “dynamic allocation of the VMs by using the fair queuing algorithm while it meets the SLA requirements creates a problem”. So, these two processes are together, finding a set of the VMs. This algorithm also includes a few constraints about all the VMs, which can achieve the balancing deficiencies. Xu et al. invented a kind of cloud partitioning load balancing model in a public cloud environment [19]. This algorithm uses switching model to successfully choose the various mechanisms for various upcoming conditions. By this, it can improve the efficiency of the entire system in a remarkable way. In the similar fashion, Ma et al. have also investigated about the automated resource allocation and capacity planning [20]. In this, they invented three time independent factors, which are not suitable for IaaS. With the incoming loads for IaaS that exhibited for the patterns of resource utilization. On the other hand, some works related to capacity balancing [21] and behaviour pattern analysis [22] for resources have been compiled to make the resource mechanisms executable.

Maguluri and Srikant presents a stochastic model of load balancing approach [23]. This approach works based on all non-pre-emptive resources, where the resources are partitioned to the entire system and each of the load chooses a set of resources. In this way, the system performance has been improved. Pham et al. stated an approach for the heterogeneous systems about how efficiently balance the incoming loads in a cloud environment [24]. This approach reduces the energy consumption of the allotted VMs as well as the data centers by addressing the VM consolidation. According to Borhani et al., most of the data centers within the system are often underutilized ( $<$ 80%) due to SLA violations which leads to the increment of physical server utilization [25]. However, this strategy typically leads to the higher utilization of resources, but isn’t providing the reasonable access of resources. So, the solution of this problem is to enable a very high level of resource utilization (80% or above). Wand et al. [26] and Doverspike et al. [27] introduced a method based on the SDN technology [28] that shares the bandwidth in an efficient manner. This technology is useful for improving the flexibility of the current incoming loads present in the system. An author named Krishna P. V invented a honeybee algorithmic concept based on a load balancing algorithm for the non-pre-emptive independent incoming loads [52]. The objective of this approach is to balance the incoming loads. So that it can maximize the CPU Utilization as well as the resource utilization and can also dynamically balance the incoming loads. Hu et al. invented an algorithm with the help of Lagrange multiplier of the Euclidean. An optimal data migration algorithm is used to diffusive the load balancing algorithms from the transferred weight [29]. This algorithm is implemented for the minimization of the data in the cloud environment. Cao et al. also proposes an approach for the calculation of an optimized performance of cloud environment [30]. Devi and Uthariaraj proposed an enhanced weighted load balancing algorithm which deals with the heavy loads in the cloud environment [31].

Bandar et al. invented a brokerage system in CSP along with the vulnerabilities and weaknesses related to the system in the multi tire cloud environment [32]. Faraji et al. proposed an identity access methodology for multi-tier cloud models [33]. Castro et al. invented a dynamic consolidation problem of VMs, which executes the required resources with deadline constrains [34]. This approach provides a structured plan for finding and investigating the minimum number of required services to successfully fulfil the requirements of the end users. Kumar et al. has described an efficient strategy for the dynamic consolidation based on the VMs [35]. This strategy estimates the “stability” and reallocates the VMs. Some of these proposed methodologies have been simulated and demonstrated by using the cloud simulation toolkit to maximize the characteristics of VMs [36, 37, 38]. At the higher utilization of resources, the performance is based on the incoming loads which are shared among various VMs. This is done through the access of physical resources with the help of Linux Control Groups with CPU shares by following the Para RMS Algorithm [45]. This ensures that the heavy loads which have the conformation of the access to the availability of the physical resources. This is not applicable for IaaS in cloud environment when revealed with various kinds of loads under the given period of time. In 2010, Fang et al. [50] described an algorithm based on load balancing and scheduling algorithm, namely: ‘Two-level Scheduling Algorithm (TLSA)’, where the requirements are successfully fulfilled by the greater utilization of resources. In 2013, Dinesh et al. proposed an algorithm, namely TS-QoS [53] which considers the priorities of loads. The problem which has occurred in this method was eliminated by the methodologies proposed in [39, 46]. Thus, this load balancing methodology can successfully implemented on different configuration of systems in a cloud environment. But the disadvantage of these algorithms again pinpointed based on the parameters involved in QoS. In 2013, Wu et al. invented a load scheduling algorithm based on the QoS-driven in cloud computing [40]. It includes various QoS parameters such as load length, availability of VMs, and remaining time, etc. In this, loads are scheduled into the services and takes minimum completion time by following the parameters included in the QoS. In 2013, U. Bhoi proposed an Enhanced max-min task scheduling algorithm [41] which enhances the computational resources in the cloud environment as well as the performance of the system based on protocols which allows the shared computation and storage in cloud computing.

In 2013, Chen et al. described an algorithm which is generally begun with any set of tasks is called ‘Min-Min Scheduling Algorithm’ [42]. It starts after finding the availability of resources within the system which take the minimum completion time for all these jobs. When the minimum completion time was successfully found, then the smaller size task is chosen for the assignment of that respective resource. The main drawback of this algorithm is that properly balancing the available loads are not possible because the existing load on the available resource is not considered before assigning the task.

In 2011, Kun Li et al. describes a task scheduling policy in the cloud which is based on the Load Balancing Ant Colony Optimization (LB-ACO) algorithm [51]. The main objective of this algorithm is to reduce the given task sets of makespan by balancing the entire available loads of the cloud computing system. In 2016, Zhao et al. described a methodology based on optimizing the candidate target host. It is mainly used for achieving the immediate effect of load balancing by picking up the optimal target host [43]. In 2011, Sindhu and Mukherjee described a combination of algorithms, namely, Longest Cloudlet Fastest Processing elements (LCFP) and Shortest Cloudlet Fastest Processing elements (SCFP) [54]. The target of both of the scheduling algorithms is to reduce the waiting time and improve the CPU utilization. In 2018, Midya et al. described a multi objective optimization technique for the resource utilization and successfully balancing the loads in a cloud environment [44]. In 2015, Zuo et al. described a biodiversity based multi objective optimization technique for the resource utilization and successfully balancing the loads in a cloud environment. In this, the demand of loads and the availability of resources is defined by the resource cost and user resources cost respectively [47]. Multi-objective optimization method performs approximately 56.6% better as compared to FCFS, Min-Min methods [40].

3. Problem formulation

In this section, first, we illustrate the problem formulation. The formulation is to make an efficient load balancing algorithm by integrating the Heuristic rules and Fair-Queuing algorithm. The objective of this algorithm is to achieve an optimal capacity of VMs and the total numbers of incoming loads present in the cloud data center. For this, we define the notations which are used in various definitions related to the proposed methodology. Notations and their meanings are illustrated in Table 1.

Table 1
notations

Notation	Meaning
$N$	Total number of available VMs
$S$	Number of available systems
$T$	Total number of loads arrived in a system at a given timeinterval
Num(Pe ${}_{i}$ )	Number of processing elements
Pe ${}_{\textit{MIPS}}$	The size of each PEs in terms of MIPS
$T_{i}$	The number of loads waiting in the $i^{\rm th}$ iteration
$C_{ki}$	The completion time of $k^{\rm th}$ load assignment to the $i^{\rm th}$ VMs
$S_{k}$	The size of the $k^{\rm th}$ loads
$C_{i}$	Capacity of each VM
$W_{i}(t)$	Total incoming loads of a VM
${sz}_{i}$	Total size of the incoming loads
$E_{ki}$	Execution time of each $k^{\rm th}$ incoming load
${Ut}_{i}$	CPU Utilization at any instance ‘ $i$ ’
$X_{i}(t)$	Total usage of time of active VMs during its observation time ‘ $t$ ’
$Y_{i}(t)$	VMs’s monitored time
${WT}_{ki}$	Waiting time of each $k^{\rm th}$ incoming load
${BW}$	Bandwidth
${BW}_{m}$	Bandwidth of any ‘ $m$ ’ intermediate modes
${wgt}_{k}$	Weight of ‘ $n$ ’ children of ‘ $m$ ’ nodes
$v_{p}(t)$	Virtual time of thread ‘ $p$ ’
$v_{q}(t)$	Virtual time of thread ‘ $q$ ’
${qt}$	Time quanta
${qt}_{p}$	Time quanta of thread ‘ $p$ ’
${qt}_{q}$	Time quanta of thread ‘ $q$ ’
${wt}_{p}$	Weight of thread ‘ $p$ ’
${wt}_{q}$	Weight of thread ‘ $q$ ’
$S^{i}_{p}$	Starting virtual time of thread ‘ $p$ ’
$S^{i}_{q}$	Starting virtual time of thread ‘ $q$ ’
$\eta^{j}$	Real time of $j^{\rm th}$ invocation of the scheduler
$a$	Thread
$S^{i}_{a}(t)$	Virtual start-up time of thread ‘ $a$ ’
$S^{0}_{a}$	Initial virtual start-up time of thread ‘ $a$ ’
$F^{i}_{a}(t)$	Virtual finish time of thread ‘ $a$ ’
$V(t)$	Virtual time

Definition 1. Calculation capacity of each VM: Let us assume that VM is the set of all ‘ $m$ ’ number of VMs which is represented as; ${\rm VM=\{VM_{1},VM_{2},\ldots,VM_{m}\}}$ . Therefore, the capacity of the i ${}^{\rm th}$ VM is defined as follows:

$\displaystyle C_{i}=Num(Pe_{i})\ast Pe_{\textit{MIPS}}$ (1)

Definition 2. Total capacity of the available VMs: The total capacity of the available VMs is calculated by taking the sum of the capacity of the entire incoming loads. It is represented as follows:

$\displaystyle C_{sum}=\sum\limits_{i=1}^{N}{C_{i}}$ (2) $\displaystyle C_{sum}\leqslant Bw_{j},1\leqslant j\leqslant S$ (3)

Definition 3. Total incoming load of a VM: The incoming loads of $i^{\rm th}$ VM is calculated as follows.

$\displaystyle W_{i}(t)=\frac{sz_{i}}{C_{i}},{\rm where}∼{}sz_{i}=\sum\limits_{% k=1}^{T_{i}}{S_{k}}$ (4)

where $sz_{i}$ represents the summation of the total size of the incoming loads which are waiting in the queue.

Definition 4. Executing time of a load: Executing time of each $k^{\rm th}$ incoming load in the $i^{\rm th}$ VM is defined with the below functionality.

$\displaystyle E_{ki}=\frac{S_{k}}{C_{i}}$ (5)

Definition 5. Waiting time of an incoming load: The waiting time for the $k^{\rm th}$ incoming load is defined as the completion of $(k-1)^{\rm th}$ time interval for the incoming load in the waiting queue of the $i^{\rm th}$ VM. It is formulated as follows:

$\displaystyle WT_{ki}=C_{(k-1)i}$ (6)

Finally, the problem is formulated as; $C_{sum}\leqslant Bw_{j},1\leqslant j\leqslant S$

3.1 VM consolidation

Flow of VM consolidation is depicted in Algorithm 1. Firstly, hosts are created and then VMs are taken for the input. After taking the inputs, VMs are assigned to the respective hosts and then cloudlets are assigned to the corresponding VMs. The status is monitored for every scheduled time interval to detect the under load hosts. These hosts are kept on sleeping mode by transferring all the available VMs to other active VMs. Then overload detection takes place by following the same technique. After completion of these steps, active VMs are selected for migration. A 7 minute iteration is considered to detect the under loaded and overloaded VMs by creating the relevant log files [10, 48, 49, 58].

Algorithm 1: VM consolidation
INPUT:
Number of hosts
min_sup $=$ Minimum Support Threshold
1. VMs are created and assigned to the respective hosts
2. Cloudlets are created and assigned to the VMs
3. for $\{$
4. Execute under load detection algorithm
5. Detect the overloaded host
6. VM is selected among the pool of available VMs for the overloaded host
7. $\}$ end for
8. end

3.2 System architecture

The data center consists of a set of identical hosts which are able to store a multiple number of VMs simultaneously. The related system architecture is depicted in Fig. 1. The role of the ‘administration controller element’ is to find whether a set of loads can be available for the system or not. The main role of the ‘VM scheduler element’ is to find out the best possible configuration of the data center as well as the configuration of VMs. The ‘load scheduler’ will perfectly manage the remaining loads which are in the waiting queue.

Figure 1.

System architecture.

4. Proposed methodology

In this section, we describe the SFQ based VM allocation methodology which is more suitable and convenient to run the proposed scheme named Heuristic Fair Queuing based Load Balancing (HFQ-LB). However, before moving to the proposed methodology, a brief overview of VM consolidation is stated with the algorithm, which describes the VM consolidation approach. The same has been validated with CloudSim Simulator.

4.1 Heuristic rules

Algorithm ${}_{2}$ describes about the heuristic rules applied in our methodology. These heuristic rules are mainly based on CPU Utilization and the amount of incoming loads arrived in the system.

4.1.1 Heuristic rules based on CPU utilization

The CPU can be efficiently utilized with the help of a ParaRMS algorithm. This algorithm effectively applies to the multicore processes. So that it can schedule many loads at a given time. All the processes which are in ready state will be granted for the execution. The algorithm efficiently allocates the deadline to each of the core based on the given time periods. Ones the VMs are available, then assigning them to the incoming loads and finally balance the work, where only one load is assigned to a particular VM at a given time period. For every time interval, loads have been monitored to check execution time based on the deadline constraints. Based on this strategy, the load is assigned to an appropriate VM. On the other hand, if any higher priority load is present in the ready state, then the load which is in execution mode will be pre-empted and goes under the waiting state. Then the higher priority load will get the chance for execution, but if the execution of the lower priority load is completed or goes to ready state then its execution remains same or left. The algorithmic formulation for CPU utilization is depicted in Algorithm 2. Based on the CPU utilization, the rules have been framed in the below manner.

Algorithm 2: CPU utilization
1. Check the schedulability of the loads
if (schedulable)
process goes on
else
exit;
2. Sort the loads according to the given time periods
3. Calculate the remaining time period of these loads
4. In the beginning, all the loads are in the ready state
5. Perform the below steps for running each of the PEs
i. Select the load of having the highest priority among all the incoming loads and changes its
state from inactive to active state and process it only for a time unit.
ii. After the execution the remaining time is reduced by a unit and this reduction of time is a
crucial job because it is should only be done by a particular core.
iii. Previously executed load may have to put into the waiting state before its completion.
6. Increase the count
7. Loads arrived in the state
8. Go to step 5 and repeat the steps until the time reaches to the remaining time period

Rule 1. Rule 1.
When CPU utilization $<$ 100%, then the remaining incoming loads are needed to wait in the waiting queue, i.e., if the CPU utilization is 80%, then the available incoming loads are assigned to the appropriate VMs and the remaining are in waiting queue.
Rule 2.
When CPU utilization $=$ 100%, then the remaining incoming loads can be assigned to the appropriate VMs.
Rule 3.
If the CPU utilization $>$ 100%, it is an invalid case.

4.1.2 Heuristic rules based on incoming loads

The load balancing approach can be achieved by mapping the incoming loads to the VM and mapping all the VMs present in the cloud environment to the hosts. This mapping is done by transferring the remaining loads from the overloaded data center. It follows the rules which are stated as follows:

Rule 1. Rule 4.
VM_Load $=$ Bw $+$ RAM $+$ MIPS
Rule 5.
Lowloaded VM $=$ Maximum Capacity $-$ (load/capacity)
Rule 6.
Overloaded VM $=$ (load/capacity) $-$ Maximum Capacity

4.2 VM configuration based on start-time fair queuing (SFQ)

In order to balance the incoming loads efficiently a fair scheduling approach is required. In this work, to allocate the VMs fairly, Start-time Fair Queuing (SFQ) algorithm is used. This algorithm is based on the concept of organizing the end users of the CPU bandwidth in a tree structure. The root node of the structure is the ‘processor’ of the system and the leaves are the ‘threads’ of each application. Here, the load balancer acts at each level of hierarchy. So, the fraction of the processor (i.e., the root node) bandwidth ‘bw’ is allocated to any intermediate node $m$ is defined as follows.

$\displaystyle\frac{bw_{m}}{bw}=\frac{wgt_{m}}{\sum\limits_{k=1}^{n}{wgt_{k}}}$ (7)

where $wgt_{k}$ , $1\leqslant k\leqslant n$ , the weight of $n$ children of node $m$ . The procedure of allocating resources in VMs is shown in Fig. 2. The working strategy of SFQ algorithm is described in the following manner.

Figure 2.

Structure of SFQ tree for scheduling VMs.

When a VM is inactive then its bandwidth is reallocated to the other active VMs at the same time. When one of the applications of VM is inactive then, its allocation is transferred to the other active applications which are running on the same VM. Similarly, if one of the threads of the system is not working properly, then its allocation is also transferred to the other threads of the applications. Call the virtual time $v_{p}(t)$ and $v_{q}(t)$ of two threads $p$ , $q$ respectively at time $t$ . At time $t$ , the virtual time of the scheduler is denoted by $v(t)$ . Now, call the time quantum $q_{t}$ in milliseconds of the scheduler. The threads $p$ and $q$ have their own time quanta namely qt ${}_{p}$ and qt ${}_{q}$ respectively. The weights of the threads $p$ and $q$ are wt ${}_{p}$ and wt ${}_{q}$ respectively. Therefore, the time quanta of these two threads are $\frac{qt}{wt_{p}}$ and $\frac{qt}{wt_{q}}$ respectively. For any $i^{\rm th}$ activation of any thread namely $p$ will start at the virtual time $S_{p}^{i}$ and finished at the virtual time $F_{p}^{i}$ . Finally, calling $\eta^{j}$ , the real time of $j^{\rm th}$ invocation of the scheduler. To manage the timing sequences, SFQ scheduler follows the rules as;

Rule 1. Rule 7.

Threads exist in the system according to their start-up time and ties can be broken arbitrarily.

Rule 8.

The virtual start-up time of the $i^{\rm th}$ activation for the thread ‘ $a$ ’ is as follows:

$\displaystyle S_{a}^{i}\left(t\right)=\max\left[{v\left({\eta^{j}}\right),F_{a% }^{(i-1)}\left(t\right)}\right]∼{}{\rm and}∼{}S_{a}^{0}=0$ (8)

The condition for thread ‘ $a$ ’ is that ‘ $i$ ’ is to be started as well as $(i-1)$ has finished and that the scheduler is active.

Rule 9. Rule 9.

The virtual finish time of the $i^{\rm th}$ activation for the thread ‘ $a$ ’ is as follows:

$F_{a}^{i}\left(t\right)=S_{a}^{i}(t)+\frac{qt}{wt_{a}}$ (9)

The thread will be stopped when its time quantum is expired and its time quantum is the time quantum of the scheduler that can be divided by the weight of the thread.

Rule 10. Rule 10.

Initially, all the threads having the same virtual time, i.e., $v_{a}^{0}=0$ . The virtual time $v(t)$ at real time $t$ is calculated as;

V(t) $=$ Virtual start time of the thread in service at time $t$ , if the CPU is busy,

V(t) $=$ Maximum finish virtual time of any thread, if the CPU is idle.

Algorithm 3: SFQ based VM allocation
Initialization:
begin
1. for( $j=0;j<N;j=j+1$ ) $\{$
2. $S_{a}^{i}=0$
3. $F_{a}^{i}=0\}$
4. end for
5. $k=$ 0;
6. $q=$ MAX
end
Received Incoming Load: (On the arrival of i ${}^{\rm th}$ load in ready queue)
begin
1. Receive the i ${}^{\rm th}$ load in queue
2. $v_{a}^{i}(t)=$ virtual time corresponding to real arrival time $t$
3. $S_{a}^{i}=\max[F_{a-1}^{i}$ , $v_{a}^{i}(t)]$
4. $F_{a}^{i}=S_{a}^{i}+L_{i}$ ; ( $L_{i}$ is the length of the incoming loads)
5. if (load is not present in the ActiveList) then
6. Insert it into the ActiveList;
End

Updation:
begin
1. while (TRUE)
2. $\{$
3. $p[m]=1;1\leqslant m\leqslant N$ ; ( $p[m]$ is the load at the head of queue $m$ )
4. $k=k+$ 1
5. $q=k\ast MAX$
6. for (every queue in the ActiveList) $\{$
7. while( $F_{j[i]}^{i}\leqslant q$ ) $\{$
8. dequeue $p[i]^{\rm th}$ load at the head of the queue $i$
9. $p[i]=p[i]+1\}$
10. if(queue $i$ is empty) $\{$
11. remove queue i from the ActiveList
12. $\}$
13. $\}$
14. $\}$
end

The algorithm allocates the VM fairly when the available bandwidth varies in time and provides throughput as well as delay guarantees. The algorithm schedules the threads according to their virtual start-up time and the shortest one will be chosen as first, where the length of the time quantum is not required when a thread is scheduled. The corresponding pseudo code of SFQ based VM allocation is presented in Algorithm 3.

4.3 Load-VM mapping

Whenever the system is configured successfully, then based on its performance the datacentre fixes the maximum queue length for the VMs. Therefore, the incoming loads can be distributed among the available variable queue length VMs of the systems to increase the CPU utilization. This functionality is stated as follows:

$\displaystyle Ut_{i}=\frac{X_{i}(t)}{Y_{i}(t)}$ (10)

where $Ut_{i}$ is the CPU utilization at any instance ‘ $i$ ’, $X_{i}(t)$ is the total usage of time, where the VMs are in active state during its observation time ‘ $t$ ’ and $Y_{i}(t)$ is the amount of time, where the VMs are monitored.

By integrating the heuristic and fair queuing strategies, the proposed methodology has been designed. The integration of fair queuing and mapping in terms of integrated heuristic and fair-queuing based VM load balancing algorithm is stated in Algorithm 4. On the other hand, the methodology of the proposed load balancing approach is validated with an example shown in Section 5.1.

Algorithm 4: Integrated Heuristic and Fair-Queuing based VM Load Balancing
Input: Incoming loads
Output: best possible VMs
Phase 1: VM Configuration based on SFQ Algorithm
begin
1. assign the incoming loads to the appropriate VMs based on the heuristic rules
2. VMs are fairly allocated by the SFQ algorithm
3. sort the loads according to their CPU utilization by following Para RMS algorithm
4. for(each load is waiting in the waiting queue)
5. $\{$
6. if(VMs are properly allotted to the loads with the help of SFQ)
7. $\{$
8. assign the loads to the appropriate VM having the maximum number of hosts
9. else if(VMs are not properly allotted to the loads with the help of SFQ)
10. $\{$
11. loads are need to wait in the waiting queue for getting the appropriate VM
12. else $\{$
13. New VM is configured by following SFQ algorithm and assigned it to the load
14. $\}$
15. $\}$
16. $\}$
17. $\}$
Phase 2: Load and VM Mapping
1. compute the capacity, total loads, execution time, waiting time and CPU Utilization
2. repeat step 1 to step 17 for new incoming loads in the cloud environment
end

Finally, all the reaming loads are assigned to the appropriate VMs based on SFQ algorithm. Figure 3 depicts the flowchart of the proposed methodology. With this, the system calculates the required capacity of each load and also calculates the total incoming loads, execution time, and waiting time of the loads as per the computations depicted in Eqs (2), (3), (5)–(7) respectively.

Figure 3.

Flowchart of the proposed approach.

5. Result analysis

The proposed algorithm is implemented in java and simulation has been performed with CloudSim toolkit 3.0.3. We create cloudlets, VMs, Datacentres, Hosts with the specific configuration which satisfy the load balancing policies. We design and deploy the proposed algorithm in the environment of cloud simulator. After the submission of cloudlet requests for the participating VMs, the resource allocation policy has been applied. It shows how the loads are distributed to the VMs of the data centres. Figures 5–7 depict the comparison of waiting time, CPU Utilization and VM Utilization of 10, 15, 20 and 25 sizes respectively. The proposed algorithm is compared with the other existing algorithms named; TLSM [50], LB-ACO [51], HBB-LB [52], TS-QoS [53], LCFP and SCFP [54], LB-BC [55], HHSA [56], and ACO [57].

Figure 4.

Makespan.

Figure 5.

Waiting time.

5.1 Working example

In this section, an example is presented to illustrate our proposed methodology. First, the less number of incoming loads are considered and the system efficiently hosts the VM. Initially, we consider the number of VMs by increasing the size as 10, 15, 20, and 25 respectively, where the considered capacity of each CPU is 1000 MIPS. In this example, we discuss about three loads namely; L ${}_{1}$ , L ${}_{2}$ and L ${}_{3}$ with their corresponding sizes 3,00,000, 1,50,000 and 70,000 respectively, where the total capacity of the system is 60,000 MIPS (i.e., consists 60 CPUs). Configuration of the VMs is given as: (3, 0, 2, 0), (0, 0, 6, 0). We obtain the best possible optimal configuration of VM for the incoming loads by following the below mentioned rules:

•
Assign the initial incoming load L ${}_{1}$ : Initially, the CPU is idle and the utilization $<$ 100%, then a new type of VM instance is created by VM size A and finally load L ${}_{1}$ is assigned. Therefore, the completion time of the initial load L ${}_{1}$ is calculated as 3,00,000/30,000 $=$ 10.
•
Assign the incoming load L ${}_{2}$ : When the CPU is not idle, but there are some available resources in the system and utilization $=$ 100%, then a new VM instance is created of size B and finally load L ${}_{2}$ is assigned. Therefore, the completion time of the initial load L ${}_{2}$ is calculated as 1,50,000/25,000 $=$ 6.
•
Assign the incoming load L ${}_{3}$ : When the CPU is not idle, but there are no available resources in the system, then the new VM instances are created of size C and D respectively. Therefore, the waiting time as well as the execution time of the incoming load L ${}_{3}$ is calculated as Min (A, B) $+$ 5 $=$ 10 and Min (10, 20) $=$ 10 respectively. The completion time of the load L ${}_{3}$ is calculated as 70,000/15,000 $=$ 7.

5.2 Performance comparison

In this section, the comparative results of the proposed algorithm have been described with other existing load balancing algorithms in terms of Makespan, Waiting Time, CPU Utilization, and VM Utilization. The comparison scenario of Makespan, Waiting Time, CPU Utilization, and VM Utilization are stated in the following manner.

Makespan: It is defined as the difference of time in between the starting and ending sequence of incoming loads in the cloud environment. In this, we calculate the makespan for every incoming load after execution of the type of VM instances. Figure 4a–d depicts the performance of the load balancing algorithms for 10, 15, 20 and 25 distinct types of VMs. With this observation, we found that our proposed approach outperforms with the above mentioned load balancing approaches.

Waiting Time: It is defined as how long an incoming load can wait in the waiting queue for getting a response from an appropriate VM. In other words, it is the difference in between the load’s arrival time and the execution time of load. Here, we calculate the waiting time by measuring the different set of incoming loads after its execution on the different types of VM instances. Figure 5a–d depicts the performance of the load balancing algorithms for 10, 15, 20 and 25 distinct types of VMs. With this experiment, it is found that our proposed approach obtains less waiting time as compared with the existing load balancing algorithms.

Table 2
Statistical comparison between load balancing approaches

Algorithm used	Exec. time	Waiting time	Resource utilization	Load balancing	Throughput	Performance	Fairness
(HFQ-LB) Proposed	Fast	Low	Best	Very good	Good	Very good	Good
TLSM [50]	Good	High	Good	Good	Moderate	Moderate	Good
LBACO [51]	Moderate	High	Good	Moderate	Moderate	Poor	Poor
HBB-LB [52]	Good	Low	Good	Good	Good	Good	Moderate
TS-QoS [53]	Moderate	High	Moderate	Good	Poor	Moderate	Poor
LCFP and SCFP [54]	Good	Moderate	Good	Moderate	Good	Good	Moderate
LB-BC [55]	Good	Low	Good	Good	Moderate	Moderate	Good
HHSA [56]	Moderate	Moderate	Good	Good	Good	Good	Moderate
ACO [57]	Low	High	Moderate	Good	Good	Moderate	Good

Figure 6.

CPU utilization.

CPU Utilization: This parameter shows that the maximum CPU utilization does not depend on any of the VM instances while efficiently executing the load balancing approach. It depends on the maximum number of active VMs. Here, we calculate the CPU utilization for each incoming load after execution of the different types of VM instances. Figure 6a–d illustrates the performance of the load balancing algorithms for 10, 15, 20 and 25 distinct types of VMs. With this instance of experiment, it is concluded that our proposed approach achieves higher CPU utilization as compared with the existing load balancing algorithms.

Figure 7.

VM utilization.

VM Utilization: This parameter shows that how many times VMs are rescheduled for the efficient execution of the incoming loads. If VM utilization is higher than resource then utilization is also higher. Here, we calculate the VM utilization for each incoming load after execution of the different types of VM instances. Figure 7a–d depicts the performance of the load balancing algorithms for 10, 15, 20 and 25 distinct types of VMs. With this experiment, it is proved that the proposed approach HFQ-LB achieves higher VM utilization as compared with the existing load balancing algorithms.

After the execution scenario, finally, we made a statistical comparison among all the load balancing algorithms, which are depicted in Table 2. With this, it is concluded that (i) the proposed methodology provides a fair allocation of VMs. Hence, the efficiency of the load balancing algorithm is achieved. (ii) On the other hand, it provides the maximum utilization of resources through heuristic rules.

6. Conclusion and future work

In this paper, an integrated heuristic and SFQ based load balancing algorithm has been designed. On the other hand, the related factors such as VM consolidation, fair allocation of VMs and heuristic rules for properly balancing the incoming loads have been introduced. SFQ based VM configuration method gives an optimized decision for selecting a VM to migrate from one host to some other host. As illustrated in Table 2, after the simulation and comparison against existing methods, it is found that the proposed method outperformed as compared with the other existing methods of load balancing in all perspectives (i.e., makespan, waiting time, CPU Utilization and VM Utilization). With these functional parameters, it is stated that the objective of the proposed approach has been achieved in an efficient manner. As a future work, the proposed methodology can be further extended with suitable optimization strategies to enhance the functional parameters in a heterogeneous cloud datacentre environment.

Footnotes

Acknowledgements

This work is partially supported by Indian Institute of Technology (ISM), under the grant of TEQIP-III/18 supported by MHRD, Govt. of India. The authors wish to express their gratitude and heartiest thanks to the Department of Computer Science and Engineering, Indian Institute of Technology (ISM), Dhanbad, India for providing their research support.

Authors’ Bios

	Dharavath Ramesh received the B.Tech degree from Kakatiya University, Warangal, India, in 2004 and M.Tech degree from the Jawaharlal Nehru Technological University, Hyderabad, India, in 2009. He received Ph.D.degree from Indian Institute of Technology (ISM) (formerly known as Indian School of Mines), Dhanbad, India in 2015. He is currently an assistant professor at the Department of Computer Science and Engineering, Indian Institute of Technology (ISM), Dhanbad, India. He works and publishes widely in the areas of Embedded Systems, Distributed Systems, Cloud Computing, Resource Management and Big Data. He is a member of the ACM and IEEE.
	Sweta Dey received the B.Tech degree in Inormtion Technolog from Maulana Abul Kalam Azad University of Technology, West Bengal, in the year 2016. She is currently an M.Tech. scholar of Computer Science and Engineering with spl. in Information Security, Indian Institute of Technology (ISM), Dhanbad, India. Her research interests include cloud resource and management, load balancing and big data security.
	Raju Bhukya has received his B.Tech in Computer Science and Engineering from Nagarjuna University in the year 2003, M.Tech degree in Computer Science and Engineering from Andhra University in the year 2005 and P.hD in Computer Science and Engineering from National Institute of Technology (NIT) Warangal in the year 2014. He is currently working as an Assistant Professor in the Department of Computer Science and Engineering in National Institute of Technology, Warangal, Telangana, India. He is currently working in the areas of Cloud Resource Management, Data Mining, and Bio-Informatics.

References

Singh

Juneja

and Malhotra

, A novel agent based autonomous and service composition framework for cost optimization of resource provisioning in cloud computing, Journal of King Saud University-Computer and Information Sciences 29(1) (2017), 19–28.

Suresh

and Sakthivel

, A novel performance constrained power management framework for cloud computing using an adaptive node scaling approach, Computers & Electrical Engineering 60 (2017), 30–44.

Chatterjee

Ojha

V.K.

Adhikari

Banerjee

Biswas

, and Snášel

, Design and implementation of an improved data center broker policy to improve the QoS of a cloud, in: Proceedings of the Fifth International Conference on Innovations in Bio-Inspired Computing and Applications IBICA 2014 (2014) 281–290. Springer, Cham.

Banerjee

Adhikari

Kar

and Biswas

, Development and analysis of a new cloudlet allocation strategy for QoS improvement in cloud, Arabian Journal for Science and Engineering 40(5) (2015), 1409–1425.

Zhu

Tian

and Wang

, Dynamic provisioning modeling for virtualized multi-tier applications in cloud data center, in; Cloud Computing (CLOUD) (2010), IEEE 3rd International Conference on (2010, July) 370–377.

Fox

Griffith

Joseph

Katz

Konwinski

Lee

and Stoica

, Above the clouds: A Berkeley view of cloud computing, Dept Electrical Eng and Comput Sciences, University of California, Berkeley, Rep UCB/EECS 28(13) (2009).

Foster

Zhao

Raicu

and Lu

, Cloud computing and grid computing 360-degree compared, in: Grid Computing Environments Workshop (2008), GCE’08 (2008 November) 1–10.

Garg

S.K.

Toosi

A.N.

Gopalaiyengar

S.K.

and Buyya

, SLA-based virtual machine management for heterogeneous workloads in a cloud datacenter, Journal of Network and Computer Applications 45 (2014), 108–120.

Buyya

C.S.

Yeo

and Venugopal

Broberg

, Brandic, Cloud computing and emerging it platforms: Vision, hype, and reality for delivering computing as the 5th utility, Future Gener. Comput. Syst. 25 (2009), 599–616.

10.

Beloglazov

, (2013), Energy-efficient Management of Virtual Machines in Data Centers for Cloud Computing (Doctoral dissertation), Link: http://beloglazovInfo/ThesisPdfAccessed04Jul2016.

11.

Beloglazov

and Buyya

, Energy efficient allocation of virtual machines in cloud data centers, in: Cluster, Cloud and Grid Computing (CCGrid) (2010), 10th IEEE/ACM International Conference on (May 2010), 577–578.

12.

Barroso

LA.

, Holzle

,The case for energy-proportional computing,Computer,40(12)(2007),33–37.

13.

Singh

and Hemalatha

, An approach on semi-distributed load balancing algorithm for cloud computing system, International Journal of Computer Applications 56(12) (2012), 5–10.

14.

Alakeel

A.M.

, A guide to dynamic load balancing in distributed computer systems, International Journal of Computer Science and Information Security 10(6) (2010), 153–160.

15.

Calheiros

R.N.

Ranjan

, and Buyya

, CloudSim: A novel framework for modeling and simulation of cloud computing infrastructures and services, (2009), arXiv Preprint ArXiv:09032525.

16.

Kondo

and Cirne

, Host load prediction in a Google compute cloud with a Bayesian model, In High Performance Computing, Networking, Storage and Analysis (SC) (2012), International Conference for (2012 November), 1–11.

17.

Aluvalu

Vardhaman

M.J.

and Kantaria

, Performance evaluation of clustering algorithms for dynamic VM allocation in cloud computing, in: Smart Technologies For Smart Nation (SmartTechCon) (2017), International Conference on (August 2017), 1560–1563.

18.

Gupta

Samvatsar

and Singh

, Cloud computing through dynamic resource allocation scheme, in: Electronics, Communication and Aerospace Technology (ICECA) (2017), International Conference of 2 (April 2017 ), 544–548.

19.

Pang

and Fu

, A load balancing model based on cloud partitioning for the public cloud, Tsinghua Science and Technology 18(1) (2013), 34–39.

20.

Y.W.

Chen

J.L.

and Lin

C.J.

, Automated network load balancing and capacity enhancing mechanism in future network, IEEE Access 6, (2018), 19407–19418.

21.

García-Lozano

Sallent

Pérez-Romero

Gomes

Á.

d’Orey

P.M.

and Ruiz

, Automated Up-and downlink capacity balancing in WCDMA networks, In Vehicular Technology Conference (2007), VTC-2007 Fall (2007), IEEE 66th (September 2007), 336–340.

22.

Preet

and Kaur

, Behavior pattern analysis for resource utilization in cloud computing, In Next Generation Computing Technologies (NGCT) (2016), 2nd International Conference on (October 2016), 124–130.

23.

Maguluri

S.T.

and Srikant

, Scheduling jobs with unknown duration in clouds, IEEE/ACM Transactions on Networking, 22(6) (2014), 1938–1951.

24.

Pham

Tran

N.H.

C.T.

Huh

E.N.

and Hong

C.S.

, Joint consolidation and service-aware load balancing for datacenters, IEEE Communications Letters 20(2) (2016), 292–295.

25.

Borhani

A.H.

Hung

Lee

B.S.

Qin

and Bagheri

, Network-Aware VM migration heuristics for improving the sla violation of multi-tier Web applications in the cloud, in: Parallel, Distributed and Network-based Processing (PDP) (2017), 25th Euromicro International Conference on (March 2017), 454–462.

26.

Wand

Tao

and Kuang

, A dynamic load balancing method of cloud-center based on SDN, China Communications 13(2) (2016), 130–137.

27.

Doverspike

Clapp

Douyon

Freimuth

D.M.

Gullapalli

Han

and Pastor

, Using SDN technology to enable cost-effective bandwidth-on-demand for cloud services, Journal of Optical Communications and Networking 7(2) (2015), A326–A334.

28.

Paolucci

Sgambelluri

Sambo

Cugini

and Castoldi

, Hierarchical OAM infrastructure for proactive control of SDN-based elastic optical networks, in: Global Communications Conference (GLOBECOM) (2015), IEEE (December 2015), 1–6.

29.

Y.F.

Blake

R.J.

and Emerson

D.R.

, An optimal migration algorithm for dynamic load balancing, Concurrency and Computation: Practice and Experience 10(6) (1998), 467–483.

30.

Cao

and Stojmenovic

, Optimal power allocation and load distribution for multiple heterogeneous multicore server processors across clouds and data centers, IEEE Transactions on Computers 63(1) (2014), 45–58.

31.

Devi

D.C.

and Uthariaraj

V.R.

, Load balancing in cloud computing environment using improved weighted round robin algorithm for non pre-emptive dependent tasks, The Scientific World Journal, 2016.

32.

Aldawsari

Baker

and England

, Trusted energy efficient cloud-based services brokerage platform, Int J Intell Comput Res 6 (2015), 630–639.

33.

Faraji

Kang

J.M.

Bannazadeh

and Leon-Garcia

, Identity access management for multi-tier cloud infrastructures, in: Network Operations and Management Symposium (NOMS) (2014), IEEE (May 2014), 1–9.

34.

Castro

Villamizar

Sotelo

Diaz

C.O.

Pecero

J.E.

, and Bouvry

, Green flexible opportunistic computing with task consolidation and virtualization, Cluster Computing 16(3) (2013), 545–557.

35.

Lixin

and Qiang

, Stability analysis of gyro based on Weibull cloud model, in: Chinese Automation Congress (CAC), 2017 (October 2017), 310–315.

36.

Buyya

Ranjan

and Calheiros

R.N.

, Modeling and simulation of scalable Cloud computing environments and the CloudSim toolkit: Challenges and opportunities, in: High Performance Computing & Simulation (2009), HPCS’09 International Conference on (June 2009), 1–11.

37.

Long

and Zhao

, A toolkit for modeling and simulating cloud data storage: An extension to CloudSim, in: Control Engineering and Communication Technology (ICCECT) (2012), International Conference on (December 2012), 597–600.

38.

Corradi

Fanelli

and Foschini

, VM consolidation: A real case based on OpenStack Cloud, Future Generation Computer Systems 32 (2014), 118–127.

39.

Goswami

Erol

Mukhi

Pajarola

and Gobbetti

, An efficient multi-resolution framework for high quality interactive rendering of massive point clouds using multi-way kd-trees, The Visual Computer 29(1) (2013), 69–83.

40.

Gao

Guan

Hou

and Liu

, A multi-objective ant colony system algorithm for virtual machine placement in cloud computing, Journal of Computer and System Sciences 79(8) (2013), 1230–1242.

41.

Bhoi

and Ramanuj

P.N.

, Enhanced max-min task scheduling algorithm in cloud computing, International Journal of Application Or Innovation in Engineering and Management (IJAIEM) 2(4) (2013), 259–264.

42.

Chen

Wang

Helian

and Akanmu

, User-priority guided Min-Min scheduling algorithm for load balancing in cloud computing, in: Parallel Computing Technologies (PARCOMPTECH) (2013), National Conference on (February 2013), 1–8.

43.

Zhao

Ding

Dong

and Fu

, A location selection policy of live virtual machine migration for power saving and load balancing, The Scientific World Journal, (2013).

44.

Midya

Roy

Majumder

and Phadikar

, Multi-objective optimization technique for resource allocation and task scheduling in vehicular cloud architecture: A hybrid adaptive nature inspired approach, Journal of Network and Computer Applications 103 (2018), 58–84.

45.

Dashora

Bajaj

H.P.

Dube

and Narayanamoorthy

, ParaRMS algorithm: A parallel implementation of rate monotonic scheduling algorithm using OpenMP, in: Advances in Electrical Engineering (ICAEE) (2014), International Conference on (January 2014), 1–6.

46.

Pradeep

and Jacob

T.P.

, Comparative analysis of scheduling and load balancing algorithms in cloud environment, in: Control, Instrumentation, Communication and Computational Technologies (ICCICCT) (2016), International Conference on (December 2016 ), 526–531.

47.

Zuo

Shu

L.E.I.

Dong

Zhu

and Hara

, A multi-objective optimization scheduling method based on the ant colony algorithm in cloud computing, IEEE Access 3 (2015), 2687–2699.

48.

Beloglazov

Abawajy

and Buyya

, Energy-aware resource allocation heuristics for efficient management of data centers for cloud computing, Future Generation Computer Systems 28(5) (2012), 755–768.

49.

Beloglazov

and Buyya

, Optimal online deterministic algorithms and adaptive heuristics for energy and performance efficient dynamic consolidation of virtual machines in cloud data centers, Concurrency and Computation: Practice and Experience 24(13) (2012), 1397–1420.

50.

Fang

Wang

and Ge

, A task scheduling algorithm based on load balancing in cloud computing, in: International Conference on Web Information Systems and Mining (October 2010), 271–277. Springer, Berlin, Heidelberg.

51.

Zhao

Dong

and Wang

, Cloud task scheduling based on load balancing ant colony optimization, in: Chinagrid Conference (ChinaGrid) (2011), Sixth Annual (August 2011), 3–9.

52.

Krishna

P.V.

, Honey bee behavior inspired load balancing of tasks in cloud computing environments, Applied Soft Computing 13(5) (2013), 2292–2303.

53.

Deng

Zhang

Zeng

and Zhou

, A task scheduling algorithm based on QoS-driven in cloud computing, Procedia Computer Science 17 (2013), 1162–1169.

54.

Sindhu

and Mukherjee

, Efficient task scheduling algorithms for cloud computing environment, in: High Performance Architecture and Grid Computing (2011), 79–83. Springer, Berlin, Heidelberg.

55.

Zhao

Yang

Wei

Ding

and Xu

, A heuristic clustering-based task deployment approach for load balancing using bayes theorem in cloud environment, IEEE Transactions on Parallel and Distributed Systems 27(2) (2016), 305–316.

56.

Tsai

C.W.

Huang

W.C.

Chiang

M.H.

Chiang

M.C.

and Yang

C.S.

, A hyper-heuristic scheduling algorithm for cloud, IEEE Transactions on Cloud Computing 2(2) (2014), 236–250.

57.

Zuo

Shu

L.E.I.

Dong

Zhu

and Hara

, A multi-objective optimization scheduling method based on the ant colony algorithm in cloud computing, IEEE Access 3 (2015), 2687–2699.

58.

Monil

M.A.H.

and Rahman

R.M.

, VM consolidation approach based on heuristics, fuzzy logic, and migration control, Journal of Cloud Computing 5(1) (2016), 8.

Heuristic and fair-queuing based VM load balancing strategy for cloud data centers: A hybrid approach

Abstract

Keywords

1. Introduction

3. Problem formulation

Table 1 notations

3.2 System architecture

4.1 Heuristic rules

4.1.1 Heuristic rules based on CPU utilization

Rule 1. Rule 4. VM_Load = Bw + RAM + MIPS Rule 5. Lowloaded VM = Maximum Capacity - (load/capacity) Rule 6. Overloaded VM = (load/capacity) - Maximum Capacity 4.2 VM configuration based on start-time fair queuing (SFQ)

Table 2 Statistical comparison between load balancing approaches

Footnotes

Acknowledgements

Authors’ Bios

References

Table 1
notations

Rule 1. Rule 4.
VM_Load $=$ Bw $+$ RAM $+$ MIPS
Rule 5.
Lowloaded VM $=$ Maximum Capacity $-$ (load/capacity)
Rule 6.
Overloaded VM $=$ (load/capacity) $-$ Maximum Capacity

4.2 VM configuration based on start-time fair queuing (SFQ)

Table 2
Statistical comparison between load balancing approaches