High Efficient Virtual Machine Migration Using Glow Worm Swarm Optimization Method for Cloud Computing

A major advantage of cloud computing is the ability to pay as you go, which makes it a perfect fit for a wide range of utility computing needs. The fifth feature of the programme might be used to characterise it. Several well-known services are available on the cloud, some of which include both software and infrastructure. These things can be made available in a variety of ways. Public, private, commodity, and hybrid are the four basic cloud deployment options. Infrastructure and maintenance costs can be decreased by utilising cloud computing. Customers benefit from cloud computing's scalability, dependability, and mobility. The low-level hardware and software infrastructure is often ignored by companies in favour of new product development and creating economic value. The move of all computer services to the cloud is being slowed down by numerous unresolved problems. The security, privacy, and energy efficiency concerns of many companies prevent them from moving their computing services to the cloud. One of the most contentious issues in today's society is energy conservation. There are superior economic incentives for data centre operators, in addition to increased environmental sustainability [1].

It's much easier to manage and use resources more efficiently when servers are virtualized in the datacenter. While this is true for physical systems, virtualized ones benefit from increased stability and availability. To run virtual machines (VMs) independently on the same physical computer, hardware resources are shared (VM). By moving virtual machines back and forth between a datacenter's physical servers, you can save money on datacenter energy costs. There is a feature in nearly all modern CPUs called dynamic voltage/frequency scaling (DVFS), which decreases CPU usage depending on workload. In standby mode, servers can consume up to 66% of their maximum energy capacity. Critical server software services and hardware components require a constant supply of electricity [2].

The two halves of a cloud computing system are linked via the Internet or Intranet. With cloud computing, the primary objective is to maximise the existing computing resources by utilising them to their fullest potential. It is vital to have algorithms for scheduling jobs if you want to optimise anything. In order to avoid this, users must plan their duties using an effective planning strategy. In order to spread the workload across as many processors as possible, scheduling techniques maximise each processor's performance while also minimising the overall execution time. To say that task scheduling is an NP-complete problem is an understatement. In order to perform tasks under problem-specific limits, scheduling arranges things in a logical order. Work scheduling in a cloud context has taken several years to create and solve utilising heuristic optimization techniques. Clonal Selection Algorithm makes advantage of the AI system's clonal selection mechanism as its core mechanism (CSA) [3].

1.1 Problem identification and objectives

VM migration is the process of moving a VM from one physical server to another. In cloud computing, the data center places VM to satisfy the demands of users without compromising the service provided to the user.

VM migration consists of two main steps: (1) VM migration from overloaded hosts to prevent service deprivation (2) VM migration from under loaded hosts to enhance the resource utilization and reduced energy consumption [4].

The main issues of VM migration are:

(1) Selection of VMs for migration. Once the migration choice has been made, you must select one or more VMs from the pool of all the ones currently allocated to the server and move them to new servers. Finding which virtual machines should be moved to servers with the optimal system specifications is the problem [5].

(2) Time and position for migration: Where new virtual machines should be installed on the server or where existing virtual machines should be transferred to other servers has an impact on the quality of VM consolidation and the energy consumption of the system Several migration techniques are available for migrating VM from one host to another. But they fail to consider the migration cost while determining the energy consumption during migration [6].

During VM migration, the main objectives are:

• To determine the optimum number of VMs for migration such that the power consumption is reduced.
• To adaptively determine the optimum locations to migrate the VMs.

This paper proposes a high efficient VM migration using Glow worm Swarm Optimization (GSO) algorithm for Cloud Computing.

3.1 Overview

This technique consists of the following 3 phases:

Phase-1: The VMs to be migrated are selected based on their resource utilization and Fault probability (FP) (ie) the VMs which are underutilized and over utilized are considered. Similarly, the VM with higher FoP are also considered for migration.

Phase-2: In order to minimize the total power consumption involved in migration, we have to determine the optimum number of VMs based on the total power consumption. In other words, after selecting the VMs for migration, the total migration power (based on the distance from server and task size) is computed. The power consumption of the remaining VMs are also computed. Then the total power consumption is the sum of these two powers. If the total power consumption becomes greater than an upper threshold, then the number of VMs from the selected VM are reduced until it again becomes less than the threshold.

Phase-3: Glowworm Swarm Optimization (GSO) is used for finding the target VMs to place the migrated VMs. The fitness function is derived in terms of distance between the main server and the other server, VM capacity and memory size. Then the VMs with best fitness functions are selected as target VMs for placing the migrated VMs.

where, λ is the failure coefficient of cloud computing environment.

T(t_i,vm_j) is the execution time of task t_i at vm_j.

Then fault probability of vm_j for m tasks is given by:

FP(vm_j) = $\sum\limits_{i=1}^{m}{F{{P}_{{}}}{{(ti)}_{{}}}}$     (2)

3.3 Total power consumption

The execution power (EP) of a task T_jwith size TS can be determined as:

EP_j = TS(t_j) * P_j     (3)

where, P_j is the power required for executing a task in unit time.

The execution power of a vm_i running n tasks is given by:

EP(vm_i) = $\sum\limits_{j=1}^{n}{T{{S}_{{}}}(tj)*{{P}_{j}}}$      (4)

The execution power of k VMs are given by:

EP_k = $\sum\limits_{i=1}^{k}{EP(v{{m}_{i}})}$      (5)

The migration power of vm_i is given by:

MP(vm_i) = d(H_s, H_t) * S(vm_i)      (6)

where, d(H_s, H_t) is the distance between the source host (H_s) and target host (H_t).

S(vm_i) is the total task size of vm_i.

Total power consumption is given by the sum of migration power of vm_i and the execution power of remaining k VMs.

totPower = MP(vm_i) + EP_k      (7)

3.4 Selection of VMs for migration

______________________________________________

Algorithm: Optimum VM selection for Migration

_______________________________________________

Let MigrationVMs = NULL

For each vm_i of host H_j

Find FP(vm_i) using Eq. (2)

     If (sizeof(H_j) >upperThresholdhostXCPUTotal) OR

If (sizeof(H_j) < lowerThresholdXhostCPUTotal) OR

If (FP(vm_i) > FPThreshold)

              add vm_i to MigrationVMs

              RemainingVMs = TotalVMs –MigrationVMs

              k = RemainingVMs

              Find totPower using Eq. (7)

              If (totPower>= MaxPower)

                       remove vm_ifromMigrationVMs

              Else

                    H_j = H_j +1

              End if

     End if

End for

_______________________________________________

3.5 GSO algorithm for VM placement

The migrated VMs are placed over the selected target VMs in another host. The target VMs are selected using GSO algorithm.

Glow-worms employ luciferin, a brightness quantity, to communicate with each other and transmit information. The Krishnanand and Ghose [14] method does the same. GSO can avoid missing the perfect response by intelligently adjusting the decision radius. This is an excellent method for locating the global optimum of a function in a small area of vector space.

3.5.1 Basic GSO operations

An initial swarm of glowworms is released into the solution space in GSO, and it spreads out at random. A small amount of luciferin is found in each glowworm, indicating that the objective function in the search space has been solved. The amount of luciferin an agent possesses is associated with their present level of fitness.

In this paradigm, you can only be lured to a nearby neighbour whose lumiferin activity is greater than your own. Because of this, the agent decides to stroll over to their next-door neighbour and introduce himself. The decision radii and radii of a glow-worm’s neighbours are related to its local-decision domain size. Local-decision domain increases to discover new neighbours when the density of neighbours is low; otherwise, it decreases to allow the horde to divide into smaller groups to allow for more efficient communication. Iterating on the algorithm until it reaches the algorithm's termination condition [15].

(i) Luciferin-Update Phase.

This phase depends on the fitness function and the earlier luciferin value. The updating equation is given as:

$\mathop{l}_{i}(t+1)=(1-\rho )\mathop{l}_{i}(t)+\gamma Fitness(\mathop{x}_{i}(t+1))$     (8)

Here, l_i(t) denotes the luciferin value of glowworm i at time t, $\rho$ is the luciferin decay constant, $\gamma$ is the luciferin enhancement constant; $x_{i}(t+1) \in R^{M}$ is the location of glowworm i at time t+1, and Fitness$\left(x_{i}(t+1)\right)$ represents the value of the fitness at glowworm i's location at time t+1.

(ii) Neighborhood-Selection Phase.

In this phase, the neighbours of the glow worm i which are having better brightness can be selected as:

$\mathop{N}_{i}(t)=\{j:\mathop{d}_{ij}(t)\prec \mathop{r}_{d}^{i}(t);\mathop{l}_{i}(t)<\mathop{l}_{j}(t)\}$     (9)

where, d_ij(t) is the distance between the worms i and j at time t, and $r_{d}^{i}(t)$ is the decision making radius of worm i at time t.

(iii) Moving Probability-Computer Phase.

A glowworm uses a probability rule to move towards other glowworms having higher luciferin level. The probability P_ij(t) of glowworm i moving towards its neighbor j can be stated as follows:

$\mathop{p}_{ij}(t)=\frac{\mathop{l}_{j}(t)-\mathop{l}_{i}(t)}{\sum\nolimits_{k\in \mathop{N}_{i}(t)} \quad {\mathop{l}_{k}(t)-\mathop{l}_{i}(t)}}$     (10)

where

$\begin{align}  & j\in \mathop{N}_{i}(t)\ne \Theta ,\mathop{N}_{i}(t) \\ & =\{j:\mathop{d}_{ij}(t)\prec \mathop{r}_{d}^{i}(t);\mathop{l}_{i}(t)<\mathop{l}_{j}(t)\} \\\end{align}$

is the set of neighbors of glowworm i.

(iv) Movement Phase.

Let glowworm i chooses another worm $j \in(t)$ for moving, with $(t)$. Then the movement of i is defined by:

$\mathop{x}_{i}(t+1)=\mathop{x}_{i}(t)+s\left( \frac{\mathop{x}_{j}(t)-\mathop{x}_{i}(t)}{||\mathop{x}_{j}(t)-\mathop{x}_{i}(t)||} \right)$     (11)

where, $(t)$ denotes the location of i at time t, s is the step size, and ‖⋅‖ is the Euclidean norm operator.

(v) Decision Radius Updating Phase.

The decision making radius of i is updated as follows:

$\begin{align}  & \mathop{r}_{d}^{i}(t+1)=\min \{\mathop{r}_{s},\max \{0,\mathop{r}_{d}^{i}(t) \\ & +\beta (\mathop{n}_{t}-|\mathop{N}_{i}(t)|)\}\} \\\end{align}$       (12)

Here, $\beta$ is a constant, $r_{S}$ denotes the sensory radius of i, and $n_{t}$ is a factor for controlling the neighbors.

3.5.2 Determining fitness function

The fitness function F for the GSO algorithm is derived in terms of memory size (Mem), CPU capacity (C), Available bandwidth (BW), Power consumption (P) and distance between the hosts (Dist) [16].

$F=$$\begin{align}  & {{\mu }_{1}}.\left( \frac{C}{{{C}_{\max }}}+\frac{BW}{B{{W}_{\max }}}+\frac{Mem}{Me{{m}_{\max }}}+ \right) \\ & +{{\mu }_{2}}\left( \frac{Dis{{t}_{\max }}}{Dist}+\frac{{{P}_{\max }}}{P} \right) \\\end{align}$      （13）

where, C_max, BW_max, Mem_max, Dist_max and P_max are maximum threshold values of C, BW, Mem, Dist and P. $\mu_{1}$ and $\mu_{2}$ are the weight values ranging from 0 to 1.

3.5.3 GSO algorithm

The fundamental GSO algorithm involves the below steps [15]:

Step 1: (parameters’ definition). The main parameters which affect the performance of GSO algorithm are, $s, \rho, \beta, R 0$, and $R s$.

Step 2: Set up the glow-worms from scratch. To begin, the luciferin and sensor range are distributed equally among the glow-worms, which are randomly deployed throughout the fitness function region. This time period is referred to as the phase. Furthermore, this is the first iteration of the project.

Step 3: This is luciferin update phase. This means that as the glowworms move around in space, their positions vary, which means that as they do, the luciferin must also adjust its value. Each luciferin-updating glowworm follows the Eq. (8).

Step 4: The fourth and last step is (movement phase). rs denotes a radial sensor range, and brighter glowworms are more attractive to each glowworm because they have a larger local-decision domain. Glowworms use a probabilistic technique to find a neighbour with a greater luciferin value and move to that one. When a glowworm I moves toward a neighbour, the probability equation is given in Eq. (10) Then, the glowworm movement equation is given in Eq. (11).

Step 5: The final stage is to: (local-decision domain update). The GSO algorithm's local-decision domain changes based on the number of gathered peaks. Glowworm local-decision domain ranges are adaptively updated according to Eq. (12).

Pseudocode of GSO algorithm

Set number of dimensions m

Set number of glowworms n

Let (t) be the location of glowworm i at time t

Generate initial population of glowworms (i = 1, 2, . . . ,n) randomly

for i= 1 to n do $\ell \dot{\imath}(0)=\ell$

0(0) = r0;

set maximum iteration number = iter max

set t= 1

While (t< iter max) do

{

(t) = 3 − (3 − 0.001) ∗ (t/iter max)¹

for each glowworm i do

$\ell(t+1)$=$(1-\rho) \ell_{i}(t)$+ $\gamma J_{i}(t+1)$;

for each glowworm i do

{

$\mathop{N}_{i}(t)=\{j:\mathop{d}_{ij}(t)\prec \mathop{r}_{d}^{i}(t);\mathop{l}_{i}(t)<\mathop{l}_{j}(t)\}$

for each glowworm $j \in N_{i}(t) d o$

$\mathop{p}_{ij}(t)=\frac{\mathop{l}_{j}(t)-\mathop{l}_{i}(t)}{\sum\nolimits_{k\in \mathop{N}_{i}(t)} \quad {\mathop{l}_{k}(t)-\mathop{l}_{i}(t)}}$

j= select glowworm ($p^{\rightarrow}$)

$\mathop{x}_{i}(t+1)=\mathop{x}_{i}(t)+s\left( \frac{\mathop{x}_{j}(t)-\mathop{x}_{i}(t)}{||\mathop{x}_{j}(t)-\mathop{x}_{i}(t)||} \right)$

$r_{d}^{i}(t+1)=R S=255$

for k = 1 to m do

$x_{i m a x}, \leftarrow x_{i m a x}, k+\alpha($ rand $-0.5)$;

end

}

$t \leftarrow t+1$;

}