research article global detection of live virtual...

Research ArticleGlobal Detection of Live Virtual Machine MigrationBased on Cellular Neural Networks

Kang Xie1 Yixian Yang123 Ling Zhang2 Maohua Jing3 Yang Xin2 and Zhongxian Li4

1 College of Information Science and Engineering Shandong University Jinan 250100 China2 Information Security Center Beijing University of Posts and Telecommunications Beijing 100876 China3Northeastern University amp College of Information Science and Engineering Shenyang 110819 China4National Cybernet Security Ltd Beijing 100088 China

Correspondence should be addressed to Kang Xie xiekang1987sinacom

Received 12 February 2014 Accepted 31 March 2014 Published 6 May 2014

Academic Editor Antonio Puliafito

Copyright copy 2014 Kang Xie et alThis is an open access article distributed under the Creative CommonsAttribution License whichpermits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

In order tomeet the demands of operationmonitoring of large scale autoscaling and heterogeneous virtual resources in the existingcloud computing a newmethod of live virtual machine (VM)migration detection algorithm based on the cellular neural networks(CNNs) is presented Through analyzing the detection process the parameter relationship of CNN is mapped as an optimizationproblem inwhich improved particle swarmoptimization algorithmbased on bubble sort is used to solve the problem Experimentalresults demonstrate that the proposed method can display the VM migration processing intuitively Compared with the best fitheuristic algorithm this approach reduces the processing time and emerging evidence has indicated that this new approach isaffordable to parallelism and analog very large scale integration (VLSI) implementation allowing the VM migration detection tobe performed better

1 Introduction

Cloud computing provides ability to dynamically scale orshrink the provisioned resources as per the dynamic require-ments [1] It is elastic in nature and works on pay as itsclients use model Virtualization as a key concept of cloudcomputing gives an abstract view of hardware by means ofmultiple virtual guest operating systems which are knownas Virtual machines (VMs) and are instanced on a singlephysical machine (PM) That is depending on the capacityof the PM many VMs can be created on it It is alsoclear that the proper management of VMs can improve theresource utilization efficiently and carry out the isolationto multitenancy Virtual machine monitor (VMM) whichis also known as hypervisor is simply a software and canbe used as an interface between PM and VMs One of thewidely used VMM is Xen which allows various VMs to sharecommon hardware in a safe and resource managed conditionbut without sacrificing the performance [2]

VM migration is a key enabler for dynamic resourcemanagement in cloud-based systems Live VM migration as

an extremely powerful tool for the cloud managers transfersstate of a VM from one PM to another and can mitigateoverload conditions and enables uninterrupted maintenanceactivities In themigration process complete state of runningVM which includes the permanent storage (ie the diskimage) volatile storage (the memory) the state of connecteddevices (such as network interface card) and the internal stateof the virtual CPU (VCPU) has to be transferred [2] In thiscase VM locations are varied dynamically and the internalstate of the VCPU and connected devices are a few kilobytesof data and can be easily sent to the VMM and the targetPM For these reasons it is clear that VMs can be migratedfrom overloaded PM to underloaded PM but it will be onlyhelpful when the efficient live migration techniques are usedThus we need an approach to trace and detect the processesmigration in VM and mapping relationship between VCPUsand PM from the hypervisorrsquos point of view in order toverify whether the new scheduler is effective and the resultis consistent with the initial idea

Therefore to meet the demands of processingtime-saving a lightweight algorithm of live VM migration

Hindawi Publishing Corporatione Scientific World JournalVolume 2014 Article ID 829614 8 pageshttpdxdoiorg1011552014829614

2 The Scientific World Journal

detection (VMMD) based on cellular neural network(CNN) is proposed in this paper CNN has features withmultidimensional array of neurons and realizable paradigmof parallel computation the processing time is unrelatedwith the data dimension Moreover it can be implementedby very large scale integration (VLSI) which makes neuralnetworks easily implemented [3] Therefore it has beenwidely used in many fields [3ndash7] such as the classificationand recognition of moving targets [4 5] path detection ofmazes [6] and microarray image reconstruction [7] To thebest of our knowledge few reports have been published onthe application in live VMmigration detection

The rest of this paper is organized as follows Section 2provides an overview of the most adopted VM migrationtechniques In Section 3 the proposed algorithm of liveVMMD based on CNN is discussed Section 4 explains thedetails for designing CNNparameters via bubble sort particleswarm optimization (BSPSO) algorithm After that Section 5presents the experimental results followed by the conclusionsin Section 6

2 Related Works

In the proposed VM migration policy VM migration istriggered when the data transfer time crosses a certainthreshold due to the unstable network Here the thresholdcan be determined by a time-related service level agreement(SLA) between the cloud facility provider and the clouduser [8] To solve the problem of host overload detec-tion a novel approach based on a Markov chain modelis proposed by maximizing the mean intermigration timeunder the specified quality of service (QoS) goal [9] In[10 11] migration strategies are generally based on the CPUutilization and when CPU utilization of a host exceeds acertain threshold tasks implementing on the VM with thehighest CPU utilization on the host will be migrated outThese strategies are generally lack of resources sensitivity andcan not realize the dynamic resource allocation according tothe actual conditions of the load

In [12] Celesti et al propose a composed image cloning(CIC) methodology able to reduce the cost of the VMdisk-image relocation over a WAN In [13] a new live VMmigration strategy is proposed by using the load character-istics to implement hotspots detection selection of VM andmigration destination host according to somemultithresholdpatterns Srikantaiah models the migration problem as amodified bin packing problem in [14] Firstly the optimalpoint is determined from profiling data Then the heuristicalgorithm is used to maximize the sum of the Euclideandistances of the current allocations to the optimal point ateach server Luo et al [15] propose an advanced VM dynamicconsolidation system through live migration approach Thesystem consists of three main components load monitorwhich collects resources usage statistics from each servernode relocation planner which calculates the relocation planby analyzing the resource usage histories and VM controllerwhich activates live migration to server nodes according tothe results from relocation planner In [16] Li et al propose

a live migration strategy for VM which combines withperformance predicting algorithm According to the averageutilization of theCPUmemory IO and network bandwidththe proposed strategy makes a serial of judgments aboutmigration such as whether a migration is triggered whatVM should be migrated and where is the destination PM ofthe VM But these existing researches are generally heuristicbased or heavily rely on statistical analysis of historical datathus the optimization procedure has to be conducted after arelatively long period to obtain statistics

To trace the migration process in VM Zhang et al[17] present a demo process migration tracing engine formonitoring the migration of process on VCPUs and VCPUson the cores of physical processor based on Linux 26 andXen32 In [18] Liu et al design a novel approach CRTR-Motionthat adopts recovery and trace technology to provide fasttransparent VM migration With execution trace logged onthe source PM a synchronization algorithm is performed toorchestrate the running source and target VM until they geta consistent state According to the above issue it is clear thatit is important for the proper management of virtualizationresources information and security events monitoring andanalysis of VM situations and migration policies Thereforewe study the tracing method for history of infected VMsfollowing the analysis of the VM migration lifecycle statusand running states

3 The Processing of VMMD Based on CNN

CNN is a large scale nonlinear locally connected analogcircuit which processes signal in real time that was firstproposed by the Professor Chua andDr Yangwho came fromthe University of California in 1988 [3] A CNN is composedof basic processing units called cells Each cell denoted by119888(119894 119895) is connected to its neighboring ones therefore onlythe adjacent cells can directly connect with each other othersinteraction are through dynamic continuous propagationeffects Each cell has the same structure composed by a smallamount of linear and nonlinear circuit elements

The circuit equations of a cell which satisfy the KCL andKVL are easily derived as follows

State equation is

119862

119889V119909119894119895(119905)

119889119905

= minus

1

119877

V119909119894119895(119905) + sum

119888(119896119897)isin119873119903(119894119895)

119860 (119894 119895 119896 119897) V119910119896119897

(119905)

+ sum

119888(119896119897)isin119873119903(119894119895)

119861 (119894 119895 119896 119897) V119906119896119897

(119905) + 119868

(1)

Output equation is

V119910119896119897

(119905) = 119891 (V119909119894119895) =

1

2

(

10038161003816100381610038161003816V119909119894119895(119905) + 1

10038161003816100381610038161003816minus

10038161003816100381610038161003816V119909119894119895(119905) minus 1

10038161003816100381610038161003816) (2)

Constraint conditions are10038161003816100381610038161003816V119909119894119895(0)

10038161003816100381610038161003816le 1

10038161003816100381610038161003816V119906119894119895(0)

10038161003816100381610038161003816le 1 (3)

where 1 le 119894 le 119872 1 le 119895 le 119873 The dynamics of a CNNhas both output feedback and input control mechanismsThe

The Scientific World Journal 3

dXijdt

1Q0Q1Q2Q3Xij

(a) The trajectory of Local Rule 1

dXijdt

minus1 P0 P1 P2 P3

Xij

(b) The trajectory of Local Rule 2

Figure 1 The dynamic trajectory of cell for different local rules

output feedback effect depends on the interactive parameter119860(119894 119895 119896 119897) and the input control effect depends on 119861(119894 119895 119896 119897)The key of CNN-based solution is to design the appropriateparameters of the feedback matrix 119860 and control matrix 119861which represent the connection weight between cells

Virtualization provides virtualized view of resources usedto instantiate VMs Once a VM is instantiated a resourcemonitoring engine which is called VMM or hypervisorwill track the resource usage and performance indicatorsrelated to the applications of the VM it also manages andmultiplexes access to the physical resources maintainingisolation between VMs at all times

Let VM119899= V1198981119899cup V1198982119899cup sdot sdot sdot V119898119898

119899denote the static state

matrix of VM numbered 119899 composed of the union of 119898continue time state components V1198981

119899 V1198982119899 and V119898119898

119899 The

allocation matrix of VM 119875119892= 11987511 119875

119894119895 119875

119897119904 can be

divided into the state set of VM 119881 = VM1 VM

119897 and

physical host (PH)119867 = 1198671 119867

119904 if a VM numbered V is

allocated into the host named ℎ 119875ℎV = 1 otherwise 119875

ℎV = minus1Any allocate solutions must meet the capacity constraints ofPHs as follows

forallℎ isin 1 119899

119897

sum

V=1119875ℎV times CPU (119881V) le CPU (119867

ℎ)


119897

sum

V=1119875ℎV timesMem (119881V) le Mem (119867

ℎ)

(4)

where the idle host will be closed to save energyLet VM

0denote theVMMwhich is called triggering cells

and let VM1015840 = VM VM0as the initial state denote the

remainder pattern to be deleted from VM VM1015840 is appliedas input to the global state detection of VM by the cloningtemplates given by (5) [3] Since VM1015840 isin VM each stateof VM at the initial time 119905 = 0 can assume the threecombinations of (input initial state) = (119906

119894119895 119909119894119895(0)) namely

(VMsuspendVMsuspend) = (minus1 minus1) (VMcopyVMsuspend) =

(1 minus1) (VMcopyVMcopy) = (1 1) Consider

119860 =

[

[

[

[

[

0

119888

2

0

119888

2

119886

119888

2

0

119888

2

0

]

]

]

]

]

119861 =

[

[

[

[

[

0 minus

119888

2

0

minus

119888

2

119887 minus

119888

2

0 minus

119888

2

0

]

]

]

]

]

119885 = 119911

(5)

where 119886 119887 119888 and 119911 are real numbers and 119888 gt 0The steady state output 119910

119894119895(infin) of each cell 119875

119894119895belonging

to any connected component obeys two static local rules andone dynamic global rule

Local Rule 1 If 119911 le 119886 + 119887 minus 1 (119906119894119895 119909119894119895(0)) = (minus1 minus1) thus

119910119894119895(infin) = minus1 independent of (119906

119896119903 119909119896119903(0)) of its neighbors

states

Local Rule 2 If 119911 ge 1 minus 119886 minus 119887 (119906119894119895 119909119894119895(0)) = (1 minus1) thus

119910119894119895(infin) = 1 independent of (119906

119896119903 119909119896119903(0)) of its neighbors

states

Proof Consider

119909∙

119894119895(119905) = minus119909

119894119895(119905) + 119886119910

119894119895(119905) + 119887119906

119894119895

+

119888

2

sum

(119896119903)isin119868

[119910119894+119896119895+119903

(119905) minus 119906119894+119896119895+119903

(119905)] + 119911

= minus119909119894119895(119905) + 119886119910

119894119895(119905) + 119908

119894119895(119905)

(6)

where 119894 = 1 2 119897 119895 = 1 2 119904Assume that the parameters satisfy 119886 gt 1119877

119909= 1

and then each cell of the CNN must settle at a stableequilibrium point after the transient has decayed to zero andlim119905rarrinfin

V119910119894119895(119905) = plusmn1 1 le 119894 le 119872 1 le 119895 le 119873 guarantee that

our CNNs have binary-value outputs This property is veryimportant for detection problems in live VMmigration

(i) If (119906119894119895 119909119894119895(0)) = (minus1 minus1) in order to guarantee Local

Rule 1 to be hold 119910119894119895(infin) = minus1 From Figure 1(a) 119909∙

119894119895le 0


dXijdt

P+ P1 P2

Xij

QminusQ1Q2

Figure 2 The dynamic trajectory of cell for global rule

the dynamic trajectory of cell119875119894119895must tend to an equilibrium

1198760= minus1 and the following inequality must be satisfiedThus

1 minus 119886 + 119908119894119895(119905) le 0

119911 le 119886 + 119887 minus 1

(7)

(ii) If (119906119894119895 119909119894119895(0)) = (1 minus1) in order to guarantee Local

Rule 2 to be hold 119910119894119895(infin) = 1 From Figure 1(b) 119909∙

119894119895ge 0

and the dynamic route of cell 119875119894119895is not below the curve with

119908119894119895= 1 minus 119886 and finally tends to an equilibrium 119875

0= 1 the

following inequality must be satisfied

119886 minus 1 + 119908119894119895(119905) ge 0 (8)

Thus

119911 ge 1 minus 119886 minus 119887 (9)

Global Rule If 119911 gt minus119886 minus 119887 minus 2119888 minus 1 (119906119894119895 119909119894119895(0)) = (1 1)

and there exists an adjacent triggering cell or a directionalconnected path defined by I from cell119875

119894119895to somemarked cell

119875119894lowast119895lowast 119910119894119895(infin) = 1 Otherwise 119911 lt 1minus119886minus119887+2119888 119910

119894119895(infin) = minus1

Proof If there exists a directional connected path fromcell 119875

119894119895to some marked cell 119875

119894lowast119895lowast then there exists some

connected path from cell 119875119894119895to some cell 119875

11989410158401198951015840 and there is at

least one adjacent triggering cell 11987511989410158401198951015840 connected to cell 119875

119894lowast119895lowast

(119906119894119895 119909119894119895(0)) = (1 1) and 119910

119894119895(infin) = minus1 In this case 119909∙

119894119895gt 0

Thus

1 + 119886 + 119908119894119895(119905) gt 0 (10)

119911 gt minus119886 minus 119887 minus 2119888 minus 1 (11)

If there neither exit an adjacent triggering cell or adirectional connected path defined by I from cell 119875

119894119895to

some marked cell 119875119894lowast119895lowast (119906119894119895 119909119894119895(0)) = (1 1) The dynamic

trajectory of cell must tend to an equilibrium 119875+indicated in

Figure 2 that is 119909∙119894119895lt 0

Hence

1 + 119886 + 119908119894119895(119905) lt 0 (12)

119911 lt minus1 minus 119886 minus 119887 + 2119888 (13)

If (11) and (13) are satisfied then the global rule holds Insummary we complete the proof of Local Rules 1 and 2 andglobal rule

4 The Design of CNN Template ParametersBased on BSPSO Algorithm

CNN is a typical nonlinear dynamic system with ability ofoptimization The Lyapunov function 119864(119905) of a CNN by thescalar function is defined as follows

119864 (119905) = minus

1

2

sum

(119894119895)

sum

(119896119897)

119860 (119894 119895 119896 119897) V119910119894119895(119905) V119910119896119897

(119905)

+

1

2119877119909

sum

(119894119895)

V119910119894119895(119905)2

minus sum

(119894119895)

sum

(119896119897)

119861 (119894 119895 119896 119897) V119910119894119895(119905) V119906119896119897

minus sum

(119894119895)

119885V119910119894119895(119905)

(14)

where 1 le 119894 119896 le 119872 1 le 119895 119897 le 119873 And it is amonotone decreasing function always converges to a localminimum where the CNN produces the desired outputThus the detected problem of live VM migration based onCNN can be changed into constrained optimization problem(COP) as follows

min120597119864CNN (119905)

120597119905

= minus119909119894119895(119905) + sum

119896119897isin119873119894119895(119903)

119860119896119897119910119896119897(119905)

+ sum

119896119897isin119873119894119895(119903)

119861119896119897119906119896119897+ 119885119894119895

(15)

st

119911 le 119886 + 119887 minus 1

119911 ge 1 minus 119886 minus 119887

119911 gt minus119886 minus 119887 minus 2119888 minus 1

119911 lt minus1 minus 119886 minus 119887 + 2119888

(16)

Most existing algorithms for COP use the penalty func-tion method to handle constrains which depends stronglyon the penalty parameter [19 20] PSO as a global searchoptimization algorithm has been widely used in COP [21]The algorithm was inspired by the social behavior of a flockof birds when searching for food The potential solutions aredenoted as particles fly in the search space exploring forbetter regions In PSO each particle has a current positionand a velocity and the particle finds the best solution throughexchanges information with other experienced particles

Assume that a swarm 119875(119896) is composed of119873 particles inthe 119863-dimensional solution space where the position vectorof particle 119894 is119883

119894= (1199091198941 1199091198942 119909

119894119889) and119881

119894= (V1198941 V1198942 V

119894119889)

denotes the velocity vector 119875119894= (119901

1198941 1199011198942 119901

119894119889) is the


optimal solution vector experienced that is the best fitnesssolution 119901best found by particle 119894 In other words 119875

119892=

(1199011198921 1199011198922 119901

119892119889) denotes the best fitness solution 119892best

found by all of particles The velocity and position updateequations of each particle are shown as follows

119881119894(119905 + 1) = 120596 lowast 119881

119894(119905) + 119888

1lowast 1199031lowast (119901best minus 119881119894 (119905))

+ 1198882lowast 1199032lowast (119892best minus 119881119894 (119905))

119883119894(119905 + 1) = 119883

119894(119905) + 119881

119894(119905 + 1)

(17)

where 119894 = 1 2 119889 The inertia weight 120596 is a factor used tocontrol the balance of the search algorithm exploitation Theacceleration coefficient 119888

1moderates the maximum step size

toward the global best particle while another coefficient 1198882

moderates the step size toward the best personal position ofthat particle They are bounded by 0 lt 119888

1 1198882lt 2 119903

1and 1199032

are random value in the range of [0 1] and usually selected itsrange artificially to reduce the probability that119881

119894(119905) and119883

119894(119905)

escape from the search spaceThere are no selection crossover and mutation opera-

tions to train the parameters of neural network by usingPSO The algorithm is simple quick convergence and thetraining precision is high but when a particle finds thelocal optimal solution it will stop searching in the solutionspace while other particles will quickly move closer to thisparticle therefore the algorithm is easy to fall into prematureconvergence In addition when the algorithm gets into alocal optimum an infeasible solution replacing mechanismis given to improve the search capability in this paper

The basic idea of the proposed algorithm was to putfeasible solutions and infeasible solutions into two differentcontainers respectively Let 119866fs = 119909

1 1199092 119909

1198991 denote

the set of feasible solution and the set of infeasible solutionis 119866ifs = 119910

1 1199102 119910

1198992 where 1198991 + 1198992 = 119873 119865fs(119909) and

119865ifs(119910) are the fitness function of 119866fs and 119866ifs respectivelyConsidering that the global optimal solutions often locateon or near the boundary of the feasible region for manyCOPs we choose some better solutions in 119866ifs based onbubble sort algorithm (BS) and add to119866fs in order to improvethe searching ability According to this the value of 119865fs(119909)should be sorted from small to large based on BS algorithmthat is after BS processing in the ordered set of feasiblesolution 1198661015840fs = 119909

1015840

1 1199091015840

2 119909

1015840

1198991 11990910158401is the best fitness solution

Similarly 11991010158401is the best fitness solution in 1198661015840ifs

In the sort processing the competitive choice followsthese rules

(1) any feasible solution is better than the infeasiblesolution

(2) in two feasible solutions it gets ahead whose value of119865fs(119909) is superior

(3) in two infeasible solutions smaller degree of 119865ifs(119910) issuperior

The processing of improved PSO based on BS algorithmis given as follows

Step 1 In accordance with the restraint condition the parti-clesrsquo velocity and position are to be initialized Iteration time(IT) IT = 1

Step 2 For each particle calculate the current value of thefitness function 119865(119909) from the formula (15)

Step 3 Compare 119865(119909) which is current experienced with thebest fitness function 119901best and if current 119865(119909) is smaller than119865(119901best) the particle will be set to 119866fs and become the new119901best Otherwise the particle will be set to 119866ifs Then IT =

IT + 1

Step 4 Update the velocity and position of the particlesaccording to formulas (5) and (6)

Step 5 After 119899 times of iteration obtain some superiorparticles in 1198661015840ifs according to 119865ifs(119910) and BS algorithm

Step 6 Put the choosing particles into 119866fs and get the bestfitness function 119892best of current globally experienced in 119866fsbased on BS algorithm

Step 7 Determine whether the algorithm met the rule ofiteration times if it satisfied the condition then go to Step 8otherwise return to Step 2

Step 8 Output the value of parameters 119886 119887 119888 and 119911

The flow diagram of BSPSO is as Figure 3

5 The Simulation Results and Analysis

In this Section we first introduce the performances of ourBSPSOalgorithm for the design ofCNN template parametersincluding the ability to get the template solution and thecomparison results withGA PSOBSPSO and sort algorithmThen the performance of our VMMDmodel based on CNNalgorithm is evaluated such as the effects of VMMD modelon migration lifecycle and the comparison results with otherpublished existing algorithms

51 The Experimentation Results of BSPSO Algorithm In ourstudy we use MATLAB 760 software on the PC of 2Gmemory to carry out this simulation experiments In orderto ensure stability and convergence of the BSPSO algorithmaccelerating factors 119888

1and 1198882are made to 149 inertia factor

is equal to 0729 1199031and 119903

2are the random number in

the interval [0 1] the initial state V119909119894119895(0) and V

119910119894119895(0) of the

randomly selected center cells are set to minus1 other parameters119862 = 10

minus9

119865 119877119909= 1Ω and the maximum iterations is 2000

times Finally the algorithm gets the template solution asfollows

119860 =[

[

0 064 0

064 632 064

0 064 0

]

]

119861 =[

[

0 minus064 0

minus064 219 minus064

0 minus064 0

]

]

119885 = minus736

(18)


Initialize the particlesrsquo velocity

Obtain superior particles based on BS

and position

Calculate F(x)

F(pbest )

No

No

No

Yes

Yes

Yes

Gfs

Gifs

F(gbest )

Update the particlesrsquo velocityand position

Output the value of parameters

IT

a b c and z

Figure 3 Flow diagram of BSPSO

2 4 6 8 10 12 14 16 18 200

1

2

3

4

5

6

7

Iterations

Fitn

ess v

aria

nce

BSPSOPSO

GASORT

times10minus3

times102

Figure 4 Simulation result of the optimal solution efficiency

Then we optimize the template parameter by usingGA PSO BSPSO and sort algorithm In order to analyzethe superiority of the proposed algorithm quantitatively wedefine an evaluation criteria formula as (19) by using thenumber of iterations and fitness variance

1205902

=

119899

sum

119894=1

(

119891119894minus 119891avg

119891

) (19)

where 119891 is a normalization factor and can be taken asarbitrary values119891

119894represents the fitness of the particle whose

0 5 10 15t (min)

20 25 30 350

255075

100125150175

PH1PH2PH3

CPU

util

izat

ion

()

Figure 5 The sum of CPU percentage usage of VMs on each PH

number is 119894 and119891avg denotes the average fitness of the particleswarmThe simulation result of the quantitative evaluation ofthe optimal solution efficiency is shown in Figure 4

As depicted in Figure 4 the convergence speed of GA-based parameter optimization algorithm is faster than thatof sort-based algorithm and the performance of convergencespeed of PSO algorithm is better than GA but when the PSOalgorithm finds a local optimal solution it stops doing thesearching and sinks into the state of premature convergenceIn the meantime by adding the BS operating the BSPSOmodel is slower than PSO in the first period of time but it canjump out of the premature convergence and find the globaloptimum eventually

52 The Experimentation Results of VMMD Based on CNNThis experimental environment contains four PHs each host


32 34 36 38 40 42 440

1

2

3

4

5

t (s)

PH

VM1VM2VM3

(a) The result of VMMD from 32 s to 44 s

t (s)

0

1

2

3

4

5

PH

1552 1554 1556 1558 1560 1562 1564

VM1VM2VM3

(b) The result of VMMD from 1552 s to 1564 s

Figure 6 The result of VMMD based on CNN in different times

Idle Httperf RUBiS UnixBench 0

5

10

15

20

25

30

35

40

45

Tota

l pro

cess

ing

time (

s)

Our approachBFH algorithm

Figure 7 Total processing time of VMMD for different workloads

consists of Intel (R) Core (TM) 2Duo E8400 30GHz 16GBmemory and is divided into a number of VMs with differentconfigurations The PH4 acts as VMM and connected withPH1 2 and 3 with NFS server by 1 Gbps network bandwidthline The virtualization software is Xen412 based on Linuxplatform

By ldquotoprdquo order we obtain the extent of access to a resourcesuch as memory allocated or CPU allocated to a VM In 35minutes the CPU percentage use of PHs is shown in Figure 5andwhen the sumof CPUpercentage use of VMs on each PHexceeds 100 like PH1 andPH2 theVMs instantiated on thatPHwill migrate to other PMs because of the violation of SLA

By ldquopublic Map adjust VM Position (String umid)lowastparam VM Namerdquo order we choose three VMs instantiatedon different PH which are running EPA-HTTP benchmarktests and the migration detection process based on CNN isshown in Figure 6

As depicted in Figure 6(a) in the first period time thestate (memory pages) of VM3 is transferred from PH1 to

PH2 and then the memory pages of VM1 migrate from PH3to PH2 that is VM1 and VM3 will suspend they copy allits pages and then resume the VMs on the target machinePH2 The migration downtime is proportional to the size ofthe VMs and network resources available for state transferWhen the physical resources can not satisfy the need of VMsexecuted on PH2 each of which is self-contained with itsown operating system the targets of VM2 and VM3 willtransfer from PH2 to PH1 in order and meet the sequenceconstraints Similarly Figure 6(b) also verifies the live VMmigration policy satisfying the sequence constraints and SLAThenewPMcan be added to offset the load of overloaded PMby migrating VM2 from PH3 to PH1 Likewise hosting newVMs may result in future overloads of PH1 which will causethe migration requires to be triggered

Figure 7 shows the total processing time of VMMDbetween our approach and traditional best fit heuristicalgorithm (BFH) [22] for different workloads in 30 minutesCompared with BFH algorithm our approach reduced theprocessing time by 3678 221 3086 and 2648for the workload of dynamic application (Idle HttperfRUBiS and UnixBench) respectively The reason is that themigration policy based on BFH should determine when aPH is considered being underloaded hence it becomes agood candidate for hosting VMs that are being migratedfrom overloaded PHs that cause the migration downtimeof dynamic application to be extremely long Whereas ourdetection approach only executes iterations to perform theprocess of the suspending and copy phase it is a feasiblemethod with satisfying lightweight performance and globallive monitoring advantage

6 Conclusions

In this paper we have proposed a live VMMD model basedon CNN which can monitor the security of target operatingsystems and it also provides the ability to inspect the VMsrsquo


state The migration procedure can be emerged as locallyconnected nonlinear processor arrays while the outputsreach their steady state values at an equilibrium point whichrepresents a desirable feature in viewofVLSI hardware imple-mentations of real time networks On the basis of analyzingthe Local Rules 1 and 2 and the global rule the parameterrelationship can bemapped as aCOPThen BSPSOalgorithmhas been designed to find out better optimum avoiding thePSO algorithm trapping into local optimum Experiments arecarried out to demonstrate the performance of the proposedmodel and the comparative results show that the proposedmodel exhibits superior performance with shorter processingtime compared with BFH algorithm

Conflict of Interests

The authors declare that there is no conflict of interestsregarding the publication of this paper

Acknowledgments

This work is supported by the National Natural ScienceFoundation of China (no 61121061) the National NaturalScience Foundation of China (no 61161140320) and theNational Key Technology RampD Program (2012BAH37B05)

References

[1] K J Ye H Zh Wu X H Jiang and Q M He ldquoPowermanagement of virtualized cloud computing platformrdquo ChineseJournal of Computers vol 6 pp 1262ndash1285 2012

[2] S Sangeeta and C Meenu ldquoA technical review for efficientvirtual machine migrationrdquo in Proceedings of the InternationalConference on Cloud and Ubiquitous Computing and EmergingTechnologies pp 20ndash25 2013

[3] L O Chua and L Yang ldquoCellular neural networks applica-tionsrdquo IEEE Transactions on Circuits and Systems vol 35 no10 pp 1273ndash1290 1988

[4] I Rodero J Jaramillo A Quiroz M Parashar and F GuimldquoTowards energy-aware autonomic provisioning for virtual-ized environmentsrdquo in Proceedings of the 19th ACM Interna-tional Symposium on High Performance Distributed Computing(HPDC rsquo10) pp 320ndash323 June 2010

[5] S Parmaksizoglu E Gunay and M Alci ldquoDetermining clon-ning templates of CNN via MOGA edge detectionrdquo in Pro-ceedings of the IEEE 19th Signal Processing and CommunicationsApplications Conference (SIU rsquo11) pp 758ndash761 Antalya TurkeyApril 2011

[6] L Li GMa and X Du ldquoNewmethod of horizon recognition inseismic datardquo IEEE Geoscience and Remote Sensing Letters vol9 no 6 pp 1066ndash1068 2012

[7] L Wan X Liu T-T Wong and C-S Leung ldquoEvolving mazesfrom imagesrdquo IEEETransactions onVisualization andComputerGraphics vol 16 no 2 pp 287ndash297 2010

[8] M Mishra A Das P Kulkarni and A Sahoo ldquoDynamicresource management using virtual machine migrationsrdquo IEEECommunications Magazine vol 50 no 9 pp 34ndash40 2012

[9] A Beloglazov and R Buyya ldquoManaging overloaded hostsfor dynamic consolidation of virtual machines in cloud datacenters under quality of service constraintsrdquo Journal of IEEE

Transactions on Parallel and Distributed Systems vol 24 no 7pp 1366ndash1379 2013

[10] J J Liu G L Chen and X Ch Hu ldquoVirtual machine migrationscheduling strategy based on load characteristicrdquo ComputerEngineering vol 37 no 17 pp 276ndash278 2011

[11] W Liu and T Fan ldquoLive migration of virtual machine basedon recovering system and CPU schedulingrdquo in Proceedingsof the 6th IEEE Joint International Information Technologyand Artificial Intelligence Conference (ITAIC rsquo11) pp 303ndash307Chongqing China August 2011

[12] A Celesti F Tusa M Villari and A Puliafito ldquoImprovingvirtual machine migration in federated cloud environmentsrdquoin Proceedings of the 2nd International Conference on EvolvingInternet (Internet rsquo10) pp 61ndash67 Valencia Spain September2010

[13] Ch Chen H X Zhang L Zh Yu Y Fan and L N Liu ldquoA newlive virtual machine migration strategyrdquo in Proceedings of theInternational Symposiumon InformationTechnology inMedicineand Education (ITME rsquo12) pp 173ndash176 2012

[14] C-H Hsu S-C Chen C-C Lee et al ldquoEnergy-aware taskconsolidation technique for cloud computingrdquo in Proceedingsof the 3rd IEEE International Conference on Cloud ComputingTechnology and Science (CloudCom rsquo11) pp 115ndash121 AthensGreece December 2011

[15] Y Luo B Zhang XWang ZWang Y Sun and H Chen ldquoLiveand incremental whole-system migration of virtual machinesusing block-bitmaprdquo in Proceedings of the IEEE InternationalConference on Cluster Computing (CCGRID rsquo08) pp 99ndash106Tsukuba Japan October 2008

[16] H Zh Li W Luo X J Lu and J W Yin ldquoA live migrationstrategy for virtual machine based on performance predictingrdquoin Proceedings of the International Conference on ComputerScience and Service System pp 72ndash76 2012

[17] L Zhang Y B Bai and X Wei ldquoA tracing approach to processmigration for virtual machine based on multicore platformrdquo inProceedings of the 10th International Conference on Algorithmsand Architectures for Parallel Processing (ICA3PP rsquo10) vol 6081of Lecture Notes in Computer Science pp 391ndash403 2010

[18] H K LiuH Jin X F Liao L THu andCh Yu ldquoLivemigrationof virtual machine based on full system trace and replayrdquo inProceedings of the 18th ACM International Symposium on HighPerformance Distributed Computing (HPDC rsquo09) pp 101ndash110June 2009

[19] R H Shang L Ch Jiao X Ch Hu and J J Ma ldquoModifiedimmune clonal constrained multi-objective optimization algo-rithmrdquo Journal of Software vol 23 no 7 pp 1773ndash1786 2012

[20] X L Li G Ch Guo X Y Li and H M Hua ldquoA constraintoptimization based mapping method for virtual networkrdquo inProceedings of the International Conference on Graphic andImage Processing (ICGIP rsquo12) vol 49 pp 1601ndash1610 2012

[21] K K Wang Q Lv H Q Zhao and W Zhang ldquoHybrid algo-rithm for solving complex constrained optimization problemsbased on PSO and ABCrdquo Systems Engineering and Electronicsvol 34 no 6 pp 1193ndash1199 2012

[22] T Mofolo and R Suchithra ldquoHeuristic based resource allo-cation using virtual machine migration a cloud computingperspectiverdquo International Refereed Journal of Engineering andScience vol 2 no 5 pp 40ndash45 2013

Submit your manuscripts athttpwwwhindawicom

Computer Games Technology

International Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014


Distributed Sensor Networks


Advances in

FuzzySystems

Hindawi Publishing Corporationhttpwwwhindawicom

Volume 2014


ReconfigurableComputing

Hindawi Publishing Corporation httpwwwhindawicom Volume 2014


Applied Computational Intelligence and Soft Computing

thinspAdvancesthinspinthinsp

Artificial Intelligence

HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014

Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014


Electrical and Computer Engineering

Journal of

Journal of

Computer Networks and Communications


Hindawi Publishing Corporation

httpwwwhindawicom Volume 2014

Advances in

Multimedia


Biomedical Imaging


ArtificialNeural Systems

Advances in


RoboticsJournal of



Computational Intelligence and Neuroscience

Industrial EngineeringJournal of


Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014


Human-ComputerInteraction

Advances in

Computer EngineeringAdvances in



detection (VMMD) based on cellular neural network(CNN) is proposed in this paper CNN has features withmultidimensional array of neurons and realizable paradigmof parallel computation the processing time is unrelatedwith the data dimension Moreover it can be implementedby very large scale integration (VLSI) which makes neuralnetworks easily implemented [3] Therefore it has beenwidely used in many fields [3ndash7] such as the classificationand recognition of moving targets [4 5] path detection ofmazes [6] and microarray image reconstruction [7] To thebest of our knowledge few reports have been published onthe application in live VMmigration detection

The rest of this paper is organized as follows Section 2provides an overview of the most adopted VM migrationtechniques In Section 3 the proposed algorithm of liveVMMD based on CNN is discussed Section 4 explains thedetails for designing CNNparameters via bubble sort particleswarm optimization (BSPSO) algorithm After that Section 5presents the experimental results followed by the conclusionsin Section 6

2 Related Works

In the proposed VM migration policy VM migration istriggered when the data transfer time crosses a certainthreshold due to the unstable network Here the thresholdcan be determined by a time-related service level agreement(SLA) between the cloud facility provider and the clouduser [8] To solve the problem of host overload detec-tion a novel approach based on a Markov chain modelis proposed by maximizing the mean intermigration timeunder the specified quality of service (QoS) goal [9] In[10 11] migration strategies are generally based on the CPUutilization and when CPU utilization of a host exceeds acertain threshold tasks implementing on the VM with thehighest CPU utilization on the host will be migrated outThese strategies are generally lack of resources sensitivity andcan not realize the dynamic resource allocation according tothe actual conditions of the load

In [12] Celesti et al propose a composed image cloning(CIC) methodology able to reduce the cost of the VMdisk-image relocation over a WAN In [13] a new live VMmigration strategy is proposed by using the load character-istics to implement hotspots detection selection of VM andmigration destination host according to somemultithresholdpatterns Srikantaiah models the migration problem as amodified bin packing problem in [14] Firstly the optimalpoint is determined from profiling data Then the heuristicalgorithm is used to maximize the sum of the Euclideandistances of the current allocations to the optimal point ateach server Luo et al [15] propose an advanced VM dynamicconsolidation system through live migration approach Thesystem consists of three main components load monitorwhich collects resources usage statistics from each servernode relocation planner which calculates the relocation planby analyzing the resource usage histories and VM controllerwhich activates live migration to server nodes according tothe results from relocation planner In [16] Li et al propose

a live migration strategy for VM which combines withperformance predicting algorithm According to the averageutilization of theCPUmemory IO and network bandwidththe proposed strategy makes a serial of judgments aboutmigration such as whether a migration is triggered whatVM should be migrated and where is the destination PM ofthe VM But these existing researches are generally heuristicbased or heavily rely on statistical analysis of historical datathus the optimization procedure has to be conducted after arelatively long period to obtain statistics

To trace the migration process in VM Zhang et al[17] present a demo process migration tracing engine formonitoring the migration of process on VCPUs and VCPUson the cores of physical processor based on Linux 26 andXen32 In [18] Liu et al design a novel approach CRTR-Motionthat adopts recovery and trace technology to provide fasttransparent VM migration With execution trace logged onthe source PM a synchronization algorithm is performed toorchestrate the running source and target VM until they geta consistent state According to the above issue it is clear thatit is important for the proper management of virtualizationresources information and security events monitoring andanalysis of VM situations and migration policies Thereforewe study the tracing method for history of infected VMsfollowing the analysis of the VM migration lifecycle statusand running states

3 The Processing of VMMD Based on CNN

CNN is a large scale nonlinear locally connected analogcircuit which processes signal in real time that was firstproposed by the Professor Chua andDr Yangwho came fromthe University of California in 1988 [3] A CNN is composedof basic processing units called cells Each cell denoted by119888(119894 119895) is connected to its neighboring ones therefore onlythe adjacent cells can directly connect with each other othersinteraction are through dynamic continuous propagationeffects Each cell has the same structure composed by a smallamount of linear and nonlinear circuit elements

The circuit equations of a cell which satisfy the KCL andKVL are easily derived as follows

State equation is

119862

119889V119909119894119895(119905)

119889119905

= minus

1

119877

V119909119894119895(119905) + sum

119888(119896119897)isin119873119903(119894119895)

119860 (119894 119895 119896 119897) V119910119896119897

(119905)

+ sum

119888(119896119897)isin119873119903(119894119895)

119861 (119894 119895 119896 119897) V119906119896119897

(119905) + 119868

(1)

Output equation is

V119910119896119897

(119905) = 119891 (V119909119894119895) =

1

2

(

10038161003816100381610038161003816V119909119894119895(119905) + 1

10038161003816100381610038161003816minus

10038161003816100381610038161003816V119909119894119895(119905) minus 1

10038161003816100381610038161003816) (2)

Constraint conditions are10038161003816100381610038161003816V119909119894119895(0)

10038161003816100381610038161003816le 1

10038161003816100381610038161003816V119906119894119895(0)

10038161003816100381610038161003816le 1 (3)

where 1 le 119894 le 119872 1 le 119895 le 119873 The dynamics of a CNNhas both output feedback and input control mechanismsThe


dXijdt

1Q0Q1Q2Q3Xij


dXijdt

minus1 P0 P1 P2 P3

Xij








119899 V1198982119899 and V119898119898

119899 The


119894119895 119875

119897119904 can be


119897 and

physical host (PH)119867 = 1198671 119867





119897

sum


ℎ)


119897

sum


ℎ)

(4)





119894119895 119909119894119895(0)) namely



119860 =

[

[

[

[

[

0

119888

2

0

119888

2

119886

119888

2

0

119888

2

0

]

]

]

]

]

119861 =

[

[

[

[

[

0 minus

119888

2

0

minus

119888

2

119887 minus

119888

2

0 minus

119888

2

0

]

]

]

]

]

119885 = 119911

(5)







119896119903 119909119896119903(0)) of its neighbors

states



119896119903 119909119896119903(0)) of its neighbors

states

Proof Consider

119909∙

119894119895(119905) = minus119909

119894119895(119905) + 119886119910

119894119895(119905) + 119887119906

119894119895

+

119888

2

sum

(119896119903)isin119868

[119910119894+119896119895+119903

(119905) minus 119906119894+119896119895+119903

(119905)] + 119911

= minus119909119894119895(119905) + 119886119910

119894119895(119905) + 119908

119894119895(119905)

(6)


119909= 1






119894119895le 0


dXijdt

P+ P1 P2

Xij

QminusQ1Q2




1 minus 119886 + 119908119894119895(119905) le 0

119911 le 119886 + 119887 minus 1

(7)



119894119895ge 0



0= 1 the


119886 minus 1 + 119908119894119895(119905) ge 0 (8)

Thus

119911 ge 1 minus 119886 minus 119887 (9)





119894119895(infin) = minus1





11989410158401198951015840 and there is at



(119906119894119895 119909119894119895(0)) = (1 1) and 119910


119894119895gt 0

Thus

1 + 119886 + 119908119894119895(119905) gt 0 (10)



119894119895to



Figure 2 that is 119909∙119894119895lt 0

Hence

1 + 119886 + 119908119894119895(119905) lt 0 (12)





119864 (119905) = minus

1

2

sum

(119894119895)

sum

(119896119897)

119860 (119894 119895 119896 119897) V119910119894119895(119905) V119910119896119897

(119905)

+

1

2119877119909

sum

(119894119895)

V119910119894119895(119905)2

minus sum

(119894119895)

sum

(119896119897)

119861 (119894 119895 119896 119897) V119910119894119895(119905) V119906119896119897

minus sum

(119894119895)

119885V119910119894119895(119905)

(14)


min120597119864CNN (119905)

120597119905

= minus119909119894119895(119905) + sum

119896119897isin119873119894119895(119903)

119860119896119897119910119896119897(119905)

+ sum

119896119897isin119873119894119895(119903)

119861119896119897119906119896119897+ 119885119894119895

(15)

st

119911 le 119886 + 119887 minus 1

119911 ge 1 minus 119886 minus 119887



(16)



119894= (1199091198941 1199091198942 119909

119894119889) and119881

119894= (V1198941 V1198942 V

119894119889)


1198941 1199011198942 119901

119894119889) is the



119892=

(1199011198921 1199011198922 119901



119881119894(119905 + 1) = 120596 lowast 119881

119894(119905) + 119888



119883119894(119905 + 1) = 119883

119894(119905) + 119881

119894(119905 + 1)

(17)





1 1198882lt 2 119903

1and 1199032


119894(119905) and119883

119894(119905)




1 1199092 119909

1198991 denote


1 1199102 119910

1198992 where 1198991 + 1198992 = 119873 119865fs(119909) and


1015840

1 1199091015840

2 119909

1015840











IT + 1











is equal to 0729 1199031and 119903



119910119894119895(0) of the


minus9



119860 =[

[

0 064 0

064 632 064

0 064 0

]

]

119861 =[

[

0 minus064 0


0 minus064 0

]

]

119885 = minus736

(18)




and position

Calculate F(x)

F(pbest )

No

No

No

Yes

Yes

Yes

Gfs

Gifs

F(gbest )



IT

a b c and z


2 4 6 8 10 12 14 16 18 200

1

2

3

4

5

6

7

Iterations

Fitn

ess v

aria

nce

BSPSOPSO

GASORT

times10minus3

times102



1205902

=

119899

sum

119894=1

(

119891119894minus 119891avg

119891

) (19)



0 5 10 15t (min)

20 25 30 350

255075

100125150175

PH1PH2PH3

CPU

util

izat

ion

()






32 34 36 38 40 42 440

1

2

3

4

5

t (s)

PH

VM1VM2VM3


t (s)

0

1

2

3

4

5

PH

1552 1554 1556 1558 1560 1562 1564

VM1VM2VM3




5

10

15

20

25

30

35

40

45

Tota

l pro

cess

ing

time (

s)









6 Conclusions






Acknowledgments


References































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in




dXijdt

1Q0Q1Q2Q3Xij


dXijdt

minus1 P0 P1 P2 P3

Xij








119899 V1198982119899 and V119898119898

119899 The


119894119895 119875

119897119904 can be


119897 and

physical host (PH)119867 = 1198671 119867





119897

sum


ℎ)


119897

sum


ℎ)

(4)





119894119895 119909119894119895(0)) namely



119860 =

[

[

[

[

[

0

119888

2

0

119888

2

119886

119888

2

0

119888

2

0

]

]

]

]

]

119861 =

[

[

[

[

[

0 minus

119888

2

0

minus

119888

2

119887 minus

119888

2

0 minus

119888

2

0

]

]

]

]

]

119885 = 119911

(5)







119896119903 119909119896119903(0)) of its neighbors

states



119896119903 119909119896119903(0)) of its neighbors

states

Proof Consider

119909∙

119894119895(119905) = minus119909

119894119895(119905) + 119886119910

119894119895(119905) + 119887119906

119894119895

+

119888

2

sum

(119896119903)isin119868

[119910119894+119896119895+119903

(119905) minus 119906119894+119896119895+119903

(119905)] + 119911

= minus119909119894119895(119905) + 119886119910

119894119895(119905) + 119908

119894119895(119905)

(6)


119909= 1






119894119895le 0


dXijdt

P+ P1 P2

Xij

QminusQ1Q2




1 minus 119886 + 119908119894119895(119905) le 0

119911 le 119886 + 119887 minus 1

(7)



119894119895ge 0



0= 1 the


119886 minus 1 + 119908119894119895(119905) ge 0 (8)

Thus

119911 ge 1 minus 119886 minus 119887 (9)





119894119895(infin) = minus1





11989410158401198951015840 and there is at



(119906119894119895 119909119894119895(0)) = (1 1) and 119910


119894119895gt 0

Thus

1 + 119886 + 119908119894119895(119905) gt 0 (10)



119894119895to



Figure 2 that is 119909∙119894119895lt 0

Hence

1 + 119886 + 119908119894119895(119905) lt 0 (12)





119864 (119905) = minus

1

2

sum

(119894119895)

sum

(119896119897)

119860 (119894 119895 119896 119897) V119910119894119895(119905) V119910119896119897

(119905)

+

1

2119877119909

sum

(119894119895)

V119910119894119895(119905)2

minus sum

(119894119895)

sum

(119896119897)

119861 (119894 119895 119896 119897) V119910119894119895(119905) V119906119896119897

minus sum

(119894119895)

119885V119910119894119895(119905)

(14)


min120597119864CNN (119905)

120597119905

= minus119909119894119895(119905) + sum

119896119897isin119873119894119895(119903)

119860119896119897119910119896119897(119905)

+ sum

119896119897isin119873119894119895(119903)

119861119896119897119906119896119897+ 119885119894119895

(15)

st

119911 le 119886 + 119887 minus 1

119911 ge 1 minus 119886 minus 119887



(16)



119894= (1199091198941 1199091198942 119909

119894119889) and119881

119894= (V1198941 V1198942 V

119894119889)


1198941 1199011198942 119901

119894119889) is the



119892=

(1199011198921 1199011198922 119901



119881119894(119905 + 1) = 120596 lowast 119881

119894(119905) + 119888



119883119894(119905 + 1) = 119883

119894(119905) + 119881

119894(119905 + 1)

(17)





1 1198882lt 2 119903

1and 1199032


119894(119905) and119883

119894(119905)




1 1199092 119909

1198991 denote


1 1199102 119910

1198992 where 1198991 + 1198992 = 119873 119865fs(119909) and


1015840

1 1199091015840

2 119909

1015840











IT + 1











is equal to 0729 1199031and 119903



119910119894119895(0) of the


minus9



119860 =[

[

0 064 0

064 632 064

0 064 0

]

]

119861 =[

[

0 minus064 0


0 minus064 0

]

]

119885 = minus736

(18)




and position

Calculate F(x)

F(pbest )

No

No

No

Yes

Yes

Yes

Gfs

Gifs

F(gbest )



IT

a b c and z


2 4 6 8 10 12 14 16 18 200

1

2

3

4

5

6

7

Iterations

Fitn

ess v

aria

nce

BSPSOPSO

GASORT

times10minus3

times102



1205902

=

119899

sum

119894=1

(

119891119894minus 119891avg

119891

) (19)



0 5 10 15t (min)

20 25 30 350

255075

100125150175

PH1PH2PH3

CPU

util

izat

ion

()






32 34 36 38 40 42 440

1

2

3

4

5

t (s)

PH

VM1VM2VM3


t (s)

0

1

2

3

4

5

PH

1552 1554 1556 1558 1560 1562 1564

VM1VM2VM3




5

10

15

20

25

30

35

40

45

Tota

l pro

cess

ing

time (

s)









6 Conclusions






Acknowledgments


References































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in




dXijdt

P+ P1 P2

Xij

QminusQ1Q2




1 minus 119886 + 119908119894119895(119905) le 0

119911 le 119886 + 119887 minus 1

(7)



119894119895ge 0



0= 1 the


119886 minus 1 + 119908119894119895(119905) ge 0 (8)

Thus

119911 ge 1 minus 119886 minus 119887 (9)





119894119895(infin) = minus1





11989410158401198951015840 and there is at



(119906119894119895 119909119894119895(0)) = (1 1) and 119910


119894119895gt 0

Thus

1 + 119886 + 119908119894119895(119905) gt 0 (10)



119894119895to



Figure 2 that is 119909∙119894119895lt 0

Hence

1 + 119886 + 119908119894119895(119905) lt 0 (12)





119864 (119905) = minus

1

2

sum

(119894119895)

sum

(119896119897)

119860 (119894 119895 119896 119897) V119910119894119895(119905) V119910119896119897

(119905)

+

1

2119877119909

sum

(119894119895)

V119910119894119895(119905)2

minus sum

(119894119895)

sum

(119896119897)

119861 (119894 119895 119896 119897) V119910119894119895(119905) V119906119896119897

minus sum

(119894119895)

119885V119910119894119895(119905)

(14)


min120597119864CNN (119905)

120597119905

= minus119909119894119895(119905) + sum

119896119897isin119873119894119895(119903)

119860119896119897119910119896119897(119905)

+ sum

119896119897isin119873119894119895(119903)

119861119896119897119906119896119897+ 119885119894119895

(15)

st

119911 le 119886 + 119887 minus 1

119911 ge 1 minus 119886 minus 119887



(16)



119894= (1199091198941 1199091198942 119909

119894119889) and119881

119894= (V1198941 V1198942 V

119894119889)


1198941 1199011198942 119901

119894119889) is the



119892=

(1199011198921 1199011198922 119901



119881119894(119905 + 1) = 120596 lowast 119881

119894(119905) + 119888



119883119894(119905 + 1) = 119883

119894(119905) + 119881

119894(119905 + 1)

(17)





1 1198882lt 2 119903

1and 1199032


119894(119905) and119883

119894(119905)




1 1199092 119909

1198991 denote


1 1199102 119910

1198992 where 1198991 + 1198992 = 119873 119865fs(119909) and


1015840

1 1199091015840

2 119909

1015840











IT + 1











is equal to 0729 1199031and 119903



119910119894119895(0) of the


minus9



119860 =[

[

0 064 0

064 632 064

0 064 0

]

]

119861 =[

[

0 minus064 0


0 minus064 0

]

]

119885 = minus736

(18)




and position

Calculate F(x)

F(pbest )

No

No

No

Yes

Yes

Yes

Gfs

Gifs

F(gbest )



IT

a b c and z


2 4 6 8 10 12 14 16 18 200

1

2

3

4

5

6

7

Iterations

Fitn

ess v

aria

nce

BSPSOPSO

GASORT

times10minus3

times102



1205902

=

119899

sum

119894=1

(

119891119894minus 119891avg

119891

) (19)



0 5 10 15t (min)

20 25 30 350

255075

100125150175

PH1PH2PH3

CPU

util

izat

ion

()






32 34 36 38 40 42 440

1

2

3

4

5

t (s)

PH

VM1VM2VM3


t (s)

0

1

2

3

4

5

PH

1552 1554 1556 1558 1560 1562 1564

VM1VM2VM3




5

10

15

20

25

30

35

40

45

Tota

l pro

cess

ing

time (

s)









6 Conclusions






Acknowledgments


References































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in





119892=

(1199011198921 1199011198922 119901



119881119894(119905 + 1) = 120596 lowast 119881

119894(119905) + 119888



119883119894(119905 + 1) = 119883

119894(119905) + 119881

119894(119905 + 1)

(17)





1 1198882lt 2 119903

1and 1199032


119894(119905) and119883

119894(119905)




1 1199092 119909

1198991 denote


1 1199102 119910

1198992 where 1198991 + 1198992 = 119873 119865fs(119909) and


1015840

1 1199091015840

2 119909

1015840











IT + 1











is equal to 0729 1199031and 119903



119910119894119895(0) of the


minus9



119860 =[

[

0 064 0

064 632 064

0 064 0

]

]

119861 =[

[

0 minus064 0


0 minus064 0

]

]

119885 = minus736

(18)




and position

Calculate F(x)

F(pbest )

No

No

No

Yes

Yes

Yes

Gfs

Gifs

F(gbest )



IT

a b c and z


2 4 6 8 10 12 14 16 18 200

1

2

3

4

5

6

7

Iterations

Fitn

ess v

aria

nce

BSPSOPSO

GASORT

times10minus3

times102



1205902

=

119899

sum

119894=1

(

119891119894minus 119891avg

119891

) (19)



0 5 10 15t (min)

20 25 30 350

255075

100125150175

PH1PH2PH3

CPU

util

izat

ion

()






32 34 36 38 40 42 440

1

2

3

4

5

t (s)

PH

VM1VM2VM3


t (s)

0

1

2

3

4

5

PH

1552 1554 1556 1558 1560 1562 1564

VM1VM2VM3




5

10

15

20

25

30

35

40

45

Tota

l pro

cess

ing

time (

s)









6 Conclusions






Acknowledgments


References































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in






and position

Calculate F(x)

F(pbest )

No

No

No

Yes

Yes

Yes

Gfs

Gifs

F(gbest )



IT

a b c and z


2 4 6 8 10 12 14 16 18 200

1

2

3

4

5

6

7

Iterations

Fitn

ess v

aria

nce

BSPSOPSO

GASORT

times10minus3

times102



1205902

=

119899

sum

119894=1

(

119891119894minus 119891avg

119891

) (19)



0 5 10 15t (min)

20 25 30 350

255075

100125150175

PH1PH2PH3

CPU

util

izat

ion

()






32 34 36 38 40 42 440

1

2

3

4

5

t (s)

PH

VM1VM2VM3


t (s)

0

1

2

3

4

5

PH

1552 1554 1556 1558 1560 1562 1564

VM1VM2VM3




5

10

15

20

25

30

35

40

45

Tota

l pro

cess

ing

time (

s)









6 Conclusions






Acknowledgments


References































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in




32 34 36 38 40 42 440

1

2

3

4

5

t (s)

PH

VM1VM2VM3


t (s)

0

1

2

3

4

5

PH

1552 1554 1556 1558 1560 1562 1564

VM1VM2VM3




5

10

15

20

25

30

35

40

45

Tota

l pro

cess

ing

time (

s)









6 Conclusions






Acknowledgments


References































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in







Acknowledgments


References































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in










Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in



research article global detection of live virtual...

Documents