Adaptive Service Provisioning for Mobile Edge Cloud - ZTE

DEMAG2017-04-56VOL15F2VFTmdashmdash9PPSP

Adaptive Service Provisioning for Mobile Edge CloudAdaptive Service Provisioning for Mobile Edge CloudHUANG Huawei1 and GUO Song2

（1 School of Computer and Engineering The University of Aizu Aizuwakamatsu 9650006 Japan2 Department of Computing The Hong Kong Polytechnic University Hong Kong SAR 852 China）

Abstract

A mobile edge cloud provides a platform to accommodate the offloaded traffic workload generated by mobile devices It can significantly reduce the access delay for mobile application users However the high user mobility brings significant challenges to theservice provisioning for mobile users especially to delaysensitive mobile applications With the objective to maximize a profitwhich positively associates with the overall admitted traffic served by the local edge cloud and negatively associates with the access delay as well as virtual machine migration delay we study a fundamental problem in this paper how to update the serviceprovisioning solution for a given group of mobile users Such a profitmaximization problem is formulated as a nonlinear integerlinear programming and linearized by absolute value manipulation techniques Then we propose a framework of heuristic algorithms to solve this Nondeterministic Polynomial (NP)hard problem The numerical simulation results demonstrate the efficiencyof the devised algorithms Some useful summaries are concluded via the analysis of evaluation results

edge cloud mobile computing service provisioningKeywords

DOI 103969j issn 167310490205188 2017 02 001httpknscnkinetkcmsdetail341294TN201704181708002html published online April 18 2017

This work was partially supported by JSPS KAKENHI under Grant NumberJP16J07062

1 Introductionn recent years the fast development of mobile cloudtechnologies [1]- [3] has incubated large varieties ofmobile online applications to facilitate our daily lifeeg mobile online games big data applications [4]

[5] More importantly most of them are normally highly delaysensitive when executed in smartphones [6] Nowadays the mobile devices are facing numbers of challenges such as sufferingthe shortage of computing capacity [4] and the battery poverty[7] Therefore the computationalintensive workload generatedfrom the mobile devices is suggested to offload to a remote private cloud [8]-[11] for execution

To alleviate these challenges recent studies [9] [12]- [19]pay particular attentions to the cluster of distributed servers inthe intermediate layered edge cloud network called cloudletHowever in a cloudlet based network such as a metropolitanarea network [18] a certain group of mobile users normallyjoin in (or become online) and leave (or become offline) the network randomly when they are using a particular mobile application as shown in Fig 1 Therefore the disruption of connection between the mobile device and the server under a mobileapplication frequently occurs at different locations and differ

ent time frames This brings a frequent churn to the serviceprovisioning in cloudlet based network Furthermore in a realworld the access delay between each mobile device and thebase station often dynamically changes in different locationseven in a same cell (macrocell or smallcell)I

Figure 1 An example of service provisioning for mobile users under acloudlet based network The workload generated from a mobile devicecan be offloaded to a VM which resides in the local edge cloud or in a remote private cloud Meanwhile this figure also demonstrates the dynamic characteristics of an edge network eg a mobile user alternates in online and offline status frequently

BS base stationVM virtual machine

Mobile device is onlineMobile device is offline

Connection between a macrocell BS and a mobile deviceConnection between a BS and a VMTrajectory of a mobile device

Local edge cloudt 2

t 3

t 4

t 1

u 1

BSVM

Localservers

VMRemoteprivate cloud

Special Topic

ZTE COMMUNICATIONSZTE COMMUNICATIONS02 April 2017 Vol15 No 2

1


Via an extensive survey in the next section over the existingrelated studies we find out that the challenge to deal with thedynamic characteristics of the mobile cloudlet based networkshas not been well addressed so far Therefore we are motivated to study a fundamental problem in this paper how to update(partially or entirely) the service provisioning solution for a certain group of online mobile application users in a cloudletbased network supposed that the trajectory of each mobile device can be obtained according to the daily routine of each user We try to answer the following two questions 1) when to update the service provisioning solution for each mobile userand 2) how to make a trade off between the admitted trafficrate offloaded by the local edge cloud and the induced accessdelay and VMmigration delay while updating the current configuration

Our study leads to the major contributions as followsbullWe study a service provisioning problem in the cloudlet

based network and try to find a near optimal update schemefor updating the service provisioning solution for each mobileuser at each timeframe if the trajectory of each mobile user isprovided

bullWith the objective to maximize a weighted profit for network operators we first formulate this problem to a nonlinearprogramming problem which is then transformed to a solvableinteger linear programming using the absolute value manipulation techniques

bullBecause of the NP hardness of the formulated problemwe have designed a series of heuristic Algorithms to solve theproblem Extensive numerical simulation results show that thedevised algorithms can yield a near optimal solution We alsoconclude some useful findings via the discussion of evaluationresults

The remaining paper is organized as follows Section 2 reviews the related work Section 3 presents the system modeland gives the problem statement The heuristic algorithms areelaborated in Section 4 Section 5 demonstrates the numericalevaluation Finally Section 6 concludes this paper

2 Related Work

21 Cloudlet Based Edge ComputingRecently edge computing has attracted wide spread re

search efforts [9] [12]-[20] for the mobile computing For instance Xia et al [9] [12] explored a locationbased offloadingproblem aiming to permit requests offloaded to a cloudlet network Then authors proposed several efficient online algorithms that can dynamically handle the requests from users Anovel hierarchical edge cloud architecture constituted withmultiple cloudlets has been proposed in [17] to efficientlyserve the peak loads originating from mobile users Then toadaptively balance the tradeoff between response delay of mobile applications and energy efficiency Tong et al [20] pro

posed both offline and online algorithms to schedule the transmission in mobile cloud computing

In wireless networks the cloudlet placement problem alsohas been studied in [13] [14] [16] [18] For example in awireless metropolitan area network (WMAN) in order to solvethe problem of cloudlet placement Jia et al [14] proposed aplacement scheme for a number of limited cloudlets This approach is proved to greatly improve the mobile cloud performance Similarly Xu et al [13] [16] also focused on the cloudlet placement problem in which capacitated cloudlets need tofind the best deployment locations within a given set of candidate locations The objective is to minimize the average accessdelay between these activated cloudlets and mobile devicesTo this end some approximate algorithms have been devisedwith approximation ratios proved if all the cloudlet servers ownthe identical computing capability22 Task Offloading Using Edge Cloud

Wang et al [21] studied a cost reduction problem in mobileedge clouds by deciding the assignment of mobile offloadedtasks The authors formulated such a problem as a mixed integer program at first Then by introducing admission controlthe problem is simplified and solved by the proposed efficienttwophase scheduling algorithm To solve the decision makingproblem of computation offloading among multiple mobile users Chen et al [22] first formulated the problem as a multiuser computation offloading game and proved that the game always assures a Nash equilibrium Then a game theoretic distributed algorithm is proposed to offload computation intensivetasks over the mobile edge could23 Comparison

Different from all efforts made by existing work mentionedabove this paper particularly studies the service provisioningupdate problem while considering the online and offline statusof mobile users during their trajectories as well as the highlydynamic characteristics of edge cloud networks We find thatthis problem has not been well studied yet To fill this gap inthis paper we strive to design highly effective update schemesof service provisioning for edge cloud network operators

3 Network Model and Problem Statement

31 System ModelThe network that we focus on includes a cloudlet based

edge cloud and a remote private cloud The former networkconsists of a set S of local edge servers Without loss of generality as shown in Fig 2 we assume that a powerful edge serverlocates at each macrocell Therefore a mobile user connectingwith a macrocell base station is equivalent to connecting withthe corresponding local edge server In such a cloudlet basednetwork a set U of mobile application users traverse at differ

Adaptive Service Provisioning for Mobile Edge CloudHUANG Huawei and GUO Song

Special Topic

ZTE COMMUNICATIONSZTE COMMUNICATIONS 03April 2017 Vol15 No 2

2


ent places in different time slots Meanwhile each of them becomes online and offline randomly while using the applicationon their mobile devices such as smartphone tablet etc Suppose that the given trajectory of each mobile user is traced withthe ID of its associated macrocell and onlineoffline status ateach time slot As a result a timeslot labeled trajectory of a mobile user is constituted of a consecutive list of macrocell IDsFor example a mobile userrsquos trajectory looks like [〈t1cella〉〈t2cellb〉〈tn0〉〈tn+1cellp〉〈tn+2cellq〉] where〈tn0〉particularly represents that this user is offline at timeslot tn Whenthe granularity of trace is quite fine a same macrocell ID maycontinually appears many times if the mobile user keeps onlinein the macrocell area

With the provided trajectories of all mobile users the network operator needs to make a decision on where to deploy therequired VM for each user at each time slot only when the useronline There are generally three categories of optimizationmodels [15] when planning a service provisioning solution inthe cloudlet based networks 1) static planning in which boththe user mobility and VM mobility are not taken into account2) planning with nonreal time VM migrations in which bothuser mobility and M migrations are considered 3) planningwith delay sensitive live VM migrations in which the difference from the previous category is that the live VMmigrationsare taken into account In this paper the mobile applicationsare assumed as highly delay sensitive ones Therefore weadopt the optimization scenario under the third category ieconsidering the live VM migrations However according topractice we only concern the live VM migrations between theremote cloud and the local cloudlet network and ignore the delay of intra cloudlet VM migrations Table 1 shows the symbols and variables used in this paper32 Problem Statement and Formulation

We first define a binary variable xtu to denote the location todeploy the VM for an online mobile user u isinU at the timeslot

t isinT during its trajectory

xtu =igrave

iacute

icirc

iumliumliumliuml

1 if a VM is deployed for an online user uin a local edge server at the time slot t 0 if a VM is deployed for an online user uin the romote cloud at the time slot t

It can be seen that different VM deployments for an onlineuser indicate different access delays and VMmigration delaysTo represent such two terms of delays we then define an eventnamed intercloud VMmigration in which the VM serving anonline user u isinU is migrated between the remote private cloudand the local edge cloud Then another binary variable ztu is defined to denote whether the inter cloud VMmigration eventoccurs at the timeslot t isinT

ztu =igraveiacuteicirc

iuml

iuml

1 if an inter - cloud VM-migration event occursfor an online mobile user u at the time slot t 0 otherwise

By analyzing the given trajectory of each mobile user u isinUwe find that in some time slots u becomes online from the offline status Such a set of the onlineactivating time slots is denoted by F(u) Naturally we consider there is no inter cloudVMmigration event occurring in each time slot t isinF(u)

The objective is to maximize a weighted profit which positively associates with the overall admitted traffic rate that isserved by the local cloudlet network and negatively associateswith the total access delay and the migration delay In particular letting ϕt

u denote the access delay of user u isinU at the timeslot t isinT we can calculate it asϕt

u = xtumiddotCtu + ( )1 - xtu middotRt

u forallt isin Tu isinU (1)

VM virtual machineFigure 2 System model

Table 1 Symbols and variables

NotationUSTDu

Cs

F (u )Rt

u

C tu

Δt

ζΓ t

u

xtu

ztu

Descriptionthe set of mobile users in network

the set of servers in the local cloudlet based networkthe set of candidate time slots when to update the provisioning solution for

each online mobile userthe demanded traffic rate of user u isinU

the traffic processing capacity of server s isinSa set of timeslots in each of which user u becomes online from offline status

according to its given trajectorythe access delay from user u to the remote private cloud at time slot tthe access delay from user u to the local edge server at time slot t

total access delay of all mobile users at timeslot tthe normalized VMmigration delay between the private cloud and a local

edge servertotal VMmigration delay of all mobile users at timeslot t

binary variable indicating the location where to deploy a VM for an onlineuser u isin U at timeslot t isinT

binary variable denoting whether to migrate a VM between the remote privatecloud and the local cloudlet network for an online user u isinU at timeslot t isinT

VM

VM

hellip

Remote cloudAlternative

choiceLocalserverVM

t +1t

t +1ts t +1ts tt

Local server

Local serverOnline at

time slot (ts) t

Offline at t +1Connection between a VMand a user at time slot tTrajectory of a mobile user

t

Special Topic



VM

3


where Ctu and Rt

u represents the access delay from user u tothe local edge server and to the remote private cloud respectively

Then we compute the access delay which is denoted by Δtat the time slot t in the following manner

∆t =sumu isinU

ϕtu forallt isin T (2)

On the other hand we let Γt indicate the total VMmigration delay of all mobile users at the time slot t and it can becalculated as

Γt =sumt isin T

ztumiddotζ forallt isin T (3)where ζ denotes the normalized VMmigration delay betweenthe private cloud and a local edge server

Then a profit maximization is formulated as the followingnonlinear programming

maxP =sumt isin Tsumu isinU

Duxtu -sum

t isin T(w1Δt +w2Γt) (4a)

stsumu isinU

xtu∙Du∙1| (s = L(u t)) leCsforallt isin Ts isin S (4b)ztu = || xtu - xt - 1

u forallu isinU forallt t - 1 isin TF( )u (4c)ztu = 0 forallt isinF(u) forallu isinU (4d)xtu ztu isin 0 1 u isinU forallt isin T (4e)In the objective function (4a) the first term sumt isin Tsumu isinUDux

tu

calculates the total admitted traffic rate that is served by the local cloudlet network and w1 and w2 in the second term indicatethe weight coefficients of the overall access delay and migration delay respectively Constraint (4b) expresses that the capacity of each server should not be expired Note that 1| () is abinary indicator which returns 1 if and only if the given condition is satisfied and L(u t) is a location function that returnsthe cell where user u locates Equation (4c) describes the relationship between variables ztu and xtu As shown in this constraint in any two successive time slots that user u is active inboth the case under || xtu - xt - 1

u = 0 indicates that both xtu andxt - 1u have the same binary value meaning that there is no inter

cloud VMmigration event occurring at the time slot t for useru On the other hand once the intercloud VMmigration eventoccurs at the time slot t we have the situation || xtu - xt - 1

u = 1 which implies xtu and xt - 1

u must take different binary valuesenforcing ztu = 1 Furthermore (4d) imposes the aforementioned special rule for variable ztu when user u is in each timeslot of set F(u)

It is worth noting that the objective function of (4) containsztu which is decided by the constraints (4c) and (4d) However(4c) involves the absolute value functions making (4) becomenonlinear and not able to be solved using linear programmingmethods Therefore we particularly transform (4c) to two linearconstraints through the following manipulation of the absolutevalue expression

|| xtu - xt - 1u = 0 (5)

Finally the nonlinear profit maximization (4) can be reformulated as the following linear programming

maxPst (4b) (5) and (4d)

xtuztu isin 01 u isinUforallt isin T (6)

4 Heuristic AlgorithmsConventionally the service provisioning problem under the

constraints of resource capacity is known as NP hard [23]-[26] To solve the aforementioned profitmaximization problemin this section we present two types of fast heuristic algorithmsand their variants aiming to yield the service provisioning solutions in each time frame for each mobile user The major contribution of this section is the proposal of the framework of heuristic algorithms ie Algorithm 1 using which many variants ofheuristic algorithms can be devised41 The Framework of Heuristic Algorithms

We first present a framework of the heuristic algorithms inAlgorithm 1 based on which we are going to devise severalheuristic algorithms in the third subsection

In line 1 the empty solution xtu ztu is generated at first

Then it is initialized in line 3 according to a feasibility specification which is going to be presented afterwards Line 4 is tofind the set of mobile users who locate at each macrocell wherethe local server sisinS is deployed Then in line 5 algorithmssort all the mobile users decreasinglyincreasingly by their demanded rates and decide the priority to use the local edgeserver After that a priority set U t

s is obtained in line 6 to denote the priority of users at each time slot t isinT Next the VMdeployment for each server at each time slot can be decided asfollows Lines 9-15 show the operation under the case that a local server s is still capable to serve the traffic demanded by user uprime while lines 16-22 demonstrate the opposite situation Finally algorithms deploy traffic demands in each local cloudletserver until the capacity of the server expires and then deploythe remaining users to the remote cloud42 Structure and Feasibility Specification of a Solution

As mentioned we have to specify a special feasibility specification to judge the feasibility of any element in a solutionSuch a feasibility specification is elaborated with the explana


Special Topic


4


tion of solution structure in the followingAn example of the structure in a solution is shown in Fig

3a We can see that each solution particularly includes tworow of binary codes The intention of each row is illustrated inFig 3b The first row indicates the variable xtu(forallu isinUforallt isin T) while the second row represents the offlineonline status in each time slot Only the bits in the first row labeledwith an online indicator in the second row are valid bits whichare highlighted with shadow in Fig 3a The bit labeled with denotes aldquodonotcarerdquoinvalid bit which will not be includedin solution x A valid binary bit in the first row implies that aVM is deployed in the local edge server for the current timeslot if it is equal to 1 Otherwise it indicates that the VM serving a mobile user is deployed to the remote cloud Accordingto the given trajectory of each mobile user the second row of asolution can be retrieved quickly In the next step each validbit in the first row can be initialized randomly After the initial

ization of solutions x and z only the valid bits in the first roware need to be decided according to the chosen algorithm

We then explain how to retrieve the solution of inter cloudVMmigration event ie variables ztu(forallu isinUforallt isin T) when asolution x is provided According to the definition of ztu and constraints (4d) and (4c) the rules are as follows 1) to an invalidbit in the first row we consider no intercloud VMmigrationevent occurs at this current corresponding time slot 2) to anytwo adjacent valid bits in the first row if the bit correspondingto the second time slot is labeled with 1 while the bit corresponding to the first time slot is labeled with 0 we still consider that there is no intercloud VMmigration event occurring atthe second time slot 3) if any two adjacent valid bits in thefirst row are labeled with different binary values we considerthe intercloud VMmigration event occurs at the second timeslot For the example shown in Fig 3b once b1 = 0 we definitely have zt - 1

u = 0 On one hand if b1 = 0 both b2 and b3 are labeled with 1 the cases under a2 = 0 a3 = 1 and a2 = 1 a3 = 0both yield ztu = 0 and zt + 1

u = 1 On the other hand when b1 b2

and b3 are all equal to 1 the same cases under a2 = 0 a3 = 1and a2 = 1 a3 = 0 will both yield zt + 1

u = 1 for sure and the value of ztu depends on a143 Heuristic Algorithms and Variants

Based on the algorithm framework we now present twotypes of heuristic algorithms and their variants The first one iscalled OnlineFirst algorithm the basic idea of which is to tryto assign higher priority to the set of mobile users who are stillin online status at the previous one timeslot As a result a mobile user who just becomes online at the current time slot has alower priority than other local online mobile users Finally allthe mobile users located at a local cell are classified into twogroups by their priorities We further get the final sequentialset of users according to their demanded traffic rates By sorting them decreasingly or increasingly by the demanded trafficrates we finally receive the variants of such OnlineFirst algorithm which are named as OnlineFirstDecreasing and Online

Algorithm 1 Framework of Heuristic AlgorithmsInput U T S and trajectory tracesOutput xtuztu isin 01 u isinUforallt isin T

1 for t isin T u isinU do2 xtuztu larrempty3 Initialize xtuztu according to the given trajectory trace4 Find the set of mobile users located at each macrocell

where foralls isin S is deployed5 Check the priority to use the local edge server of each user

sort them decreasinglyincreasingly by their demanded rates6 Obtain a sequential set Ucirct

s of mobile users by theirpriorities for each server s isin S at each time slot t isin T

7 Decide the VM deployment for each mobile user at eachtime slot

8 for t isin T s isin S uprime isin Ucircts do9 if s is feasible to serve the traffic demanded by uprime

then10 Deploy a VM locally at s for uprime 11 xt

uprime larr 112 if t ge1 and 1 = xt - 1

uprime then13 zt

uprime larr 014 else if t ge1 and 0 = xt - 1

uprime then15 zt

uprime larr 116 else17 Deploy a VM remotely for uprime 18 xt


uprime then20 zt


uprime then22 zt

uprime larr 0

(a) An example to show the structure of a solution

Figure 3 The structure and the feasibility specification of a solution

(b) The feasibility specification to retrieve a solution z

Special Topic



1 0 0 hellip 1 1 1 hellip0 0

1 11 0 10

0 01 1 1hellip

1 hellip

Partial solution for the first userwith a length that is equal to thesize of the given time slot set

The remaining solution for other users

a1 a2

b1 b2 b3

a3hellip hellip

hellip hellip

t -1 t +1tTime slot

01 (offlineonline) indicatorVariable x tu

5


FirstIncreasing respectivelyAnother heuristic algorithm is called First Fit which is

widely adopted to solve the binpacking problem [24] Similarly according to the decreasinglyincreasingly sorting mannertowards the demanded traffic rate of each mobile user the variants of FirstFit are labeled as FirstFitDecreasing and FirstFitIncreasing respectively

5 Performance EvaluationIn this section we conduct extensive numerical simulations

to evaluate the presented 4 heuristic algorithms FirstFitDecreasing (FFD) First Fit Increasing (FFI) Online First Decreasing (OFD) and OnlineFirstIncreasing (OFI)

The basic ideas of these 4 heuristic algorithms have beenwidely used by existing studies related to the resource allocation in cloud Here we mainly compare the performance differences of the 4 heuristic algorithms designed under our proposed algorithmframework Furthermore we are also interested in the performance gaps between such 4 algorithms and theOptimal one under different system settings Finally we wouldlike to draw some useful conclusions over their performance byanalyzing the simulation results and try to suggest the serviceproviders which heuristic algorithm is the best choice under anetwork configuration51 Simulation Settings

The network topology adopted inour simulations is a cloudlet basedurban access network with 10 adjacent macrocells each of which hasan isolated local server that can onlyserve the mobile users located in thecurrent macrocell We randomly generate a traffic demand trace for eachmobile user within [10 100] MbsThe access delay to the remote cloudis fixed to 10 ms while the local access delay of any mobile user to itslocal edge server is randomly generated within [1 3] ms Furthermorethe inter cloud VMmigration delayis normalized to 10 ms

We then generate a sequential trajectory for each mobile user within20 time slots At each time slot wefirst decide the online status of anymobile user using a predefined probability which is fixed to 08 in thispaper If a user is offline in a timeslot we mark its traversed cell ID to0 Otherwise we find a cell locationfollowing a twofold rule 1) when a

mobile user becomes online from an offline status we randomly find a cell that it appears at 2) when a mobile user keeps online from the previous one time slot we find a cell for the current time slot within its located cell and the neighboring cellsas well On the other hand as a benchmark to compare performance with our devised heuristic algorithms we also solve (6)to retrieve the Optimal solution using Gurobi 60 [27] undereach simulation setting We compare heuristic algorithms andthe optimal solution in terms of 4 metrics ie total numericalprofit total traffic rate allocated to the local edge cloud theweighted access delay and the weighted migration delay52 Effect of Traffic Processing Capacity of Edge Servers

In the first group of the simulations we study the effect ofserverrsquos traffic processing capacity by varying Cs isin｛600 9001200 1500｝Mbs and fixing both w1 and w2 to 3 FromFigs4a and 4b we can observe that the profit and total numerical cloudlet traffic rate are increasing functions over the capacity of servers When the capacity is insufficient eg whenCs = 600 Mbs algorithms FFI and OFI perform better than theother two heuristics This is because in the previous two algorithms more mobile users who request traffic demands withsmall rates can be served in the local cloudlet servers resultingin smaller total access delay and migration delay

Furthermore in Figs 4c and 4d we can see that the access

FFD FirstFitDecreasing FFI FirstFitIncreasing OFD OnlineFirstDecreasing OFI OnlineFirstIncreasing

Figure 4 Performance of algorithms when the serving capacity of a local server (ie Cs)varies in a range 600 Mbs-1500 Mbs

(d) Weighted migration delay vs Cs

(a) Profit vs Cs (b) Cloudlet traffic rate vs Cs

(c) Weighted access delay vs Cs

15001200900600

times104757065605550

Nume

ricalpr

ofit

Capacity of local server (Mbs)

OptimalOFDOFIFFDFFI

15001200900600

times104

Nume

ricalcl

oudlet

traffic


848280787674

OptimalOFDOFIFFDFFI

15001200900600

times10416

08

Nume

ricalw

gtacc

essdel

ay


14

12

10

OptimalOFDOFI

FFDFFI

15001200900600

8000

0Nume

ricalw

gtmig

ration

delay


6000

4000

2000

OptimalOFDOFIFFDFFI


Special Topic


6


and migration delays decrease as the traffic processing capacity grows As expected the algorithms considering the demandswith small traffic rates to be first served ie FFI and OFIhave the lower delays than FFD and OFD algorithms

Finally once the processing capacity of local edge serversgrows sufficiently the performance of all algorithms becomessame This can be explained by the reason that every algorithmyields a similar solution and performs close to the optimal solution when the processing capacity of edge servers is not thebottleneck resource any more53 Effect of w1

Using the same traces we evaluate the effect of the weightof access delay by varying w1 isin｛1 2 3 4 5｝and fixing w2 = 1and Cs = 500 Mbs Fig 5 illustrates the same four metrics ofthe previous group of simulations Because the access delaycontributes negatively to the objective function we observe thedecreasing profits in Fig 5a and the increasing numericalweighted (shorten as wgt) access delay in Fig 5c while enlarging the weight of access delay from 1 to 5 FFI and OFI showthe larger profits than that of the other two algorithms The reason is same with the previous simulation

Interestingly Figs 5b and 5d demonstrate that improvingthe weight of access delay has no effect to the total cloudlettraffic and the weighted migrations delay This is becausechanging w1 will not significantly affect the task allocation to

the local edge cloud or to the remote cloud This is a usefulfinding to network operators54 Effect of w 2

By varying w2 isin｛1 2 3 4 5｝and setting w1 to 1 we thenstudy the effect of the weight of migration delay in this group ofsimulations Fig 6 presents the 4 metrics of the four heuristicalgorithms and the optimal solution as well In Figs 6a and 6bwe have similar observations on both the total profit and the total cloudlet traffic rate compared with the previous group ofsimulations This is because w2 plays a similar role with w1 tothe system objective

Although w2 in all heuristic algorithms has no effect on theweighted access delay from Fig6c the increasing weight of migration delay makes the weighted migration delay higherThus the total profit is reduced significantly Especially underFFD more traffic demands with small traffic rates have to experience the intercloud VMmigration than that under otheralgorithms This is because when the server capacity is limited only a small number of requests can be provisioned in thelocal edge cloud The VMs serving other users with tiny ratedemands have to be migrated to the remote cloud thus incurring higher migration delay when performing the FFD andOFD algorithms

In a summary via all the simulation results we can alwaysobserve that the FFI and OFI have a similar performance and

outperform the other two heuristicsin terms of total profit the weightedaccess delay and the weighted migration delay

6 ConclusionsIn this paper we study the up

date problem of service provisioning in the cloudlet based mobileedge network We try to find anadaptive update scheme to decidewhen to update the service provisioning solution for each mobile user at each timeframe if the trajectory of each mobile user is knownWith the objective of maximizing aweighted profit for network operators we first formulate this problemas nonlinear programming problemThen it is transformed to solvableinteger linear programming usingthe absolute value manipulationtechnique Next to solve this problem we devise a series of heuristicalgorithms Extensive numericalsimulation results demonstrate that


Figure 5 Performance of algorithms when the weight of access delay (ie w1) varies in a range 1-5

(a) Profit vs w 1 (b) Cloudlet traffic rate vs w 1

(c) Weighted access delay vs w 1 (d) Weighted migration delay vs w 1

5

times104

70

Nume

ricalpr

ofit

Weight of access delay

60

50

40

30 4321

Optimal

OFDOFI

FFDFFI

5

times104

71

Nume

ricalcl

oudlet

traffic

Weight of access delay4321

70

69

68

67

Optimal FFDFFI

OFDOFI

5

times104

Nume

ricalw

gtacc

essdel

ay


3

2

1

0

OptimalFFDFFI

OFDOFI

5

Nume

ricalw

gtmig

ration

delay


30003500

2500200015001000

Optimal

FFDFFI

OFDOFI

Special Topic



7


the devised algorithms could yield near optimal solutionsSome useful findings have been also revealed through the evaluation of simulation results

(a) Profit vs w 2


Figure 6Performance of algorithms whenthe weight of migration delay(ie w 2) varies in a range 1-5

(b) Cloudlet traffic rate vs w 2

c) Weighted access delay vs w 2 (d) Weighted migration delay vs w 2

References[1] T Verbelen P Simoens F De Turck and B DhoedtldquoCloudlets bringing the

cloud to the mobile userrdquoin Third ACM Workshop on Mobile Cloud Computingand Services Low Wood Bay Lake District UK 2012 pp 29-36 doi 10114523078492307858

[2] D Meilander F Glinka S Gorlatch et alldquoUsing mobile cloud computing forreal time online applicationsrdquoin IEEE International Conference on MobileCloud Computing Services and Engineering Oxford UK 2014 pp 48-56 doi101109MobileCloud201419

[3] N Fernando W L Seng and W RahayuldquoMobile cloud computing A surveyrdquoFuture Generation Computer Systems vol 29 no 1 pp 84-106 2016

[4] A T Lorsquoai W Bakheder and H SongldquoA mobile cloud computing model using the cloudlet scheme for big data applicationsrdquoin IEEE First InternationalConference on Connected Health Applications Systems and Engineering Technologies Washington DC USA 2016 pp 73-77 doi 101109CHASE201640

[5] A T Lorsquoai R Mehmood E Benkhelifa and H SongldquoMobile cloud computingmodel and big data analysis for healthcare applicationsrdquoIEEE Access vol 4pp 6171-6180 2016 doi 101109ACCESS20162613278

[6] K Ha Z Chen W Hu et alldquoTowards wearable cognitive assistancerdquoin International Conference on Mobile Systems Bretton Woods USA 2014 pp 68-81doi 10114525943682594383

[7] K Yang S Ou and H H ChenldquoOn effective offloading services for resourceconstrained mobile devices running heavier mobile internet applicationsrdquoIEEECommunications Magazine vol 46 no 1 pp 56- 63 2008 doi 101109MCOM20084427231

[8] D Kovachev T Yu and R KlammaldquoAdaptive computation offloading from mobile devices into the cloudrdquoin IEEE International Symposium on Parallel andDistributed Processing with Applications Madrid Spain 2012 pp 784-791 doi101109ISPA2012115

[9] Q Xia W Liang and W XuldquoThroughput maximization for online request admissions in mobile cloudletsrdquoin IEEE 38th Conference on Local Computer Networks Sydney Australia 2013 pp 589-596 doi 101109LCN20136761295

[10] W Gao Y Li H Lu T Wang and C LiuldquoOn exploiting dynamic executionpatterns for workload offloading in mobile cloud applicationsrdquoin IEEE 22ndInternational Conference on Network Protocols (ICNP) Raleigh USA 2014pp 1-12 doi 101109ICNP201422

[11] E J HaughnldquoMobile device management through an offloading networkrdquoUS Patent 8626143 Jan 7 2014

[12] Q Xia W Liang Z Xu and B ZhouldquoOnline algorithms for locationawaretask offloading in twotiered mobile cloud environmentsrdquoin IEEEACM 7th International Conference on Utility and Cloud Computing London UK 2014 pp109-116 doi 101109UCC201419

[13] Z Xu W Liang W Xu M Jia and S GuoldquoCapacitated cloudlet placementsin wireless metropolitan area networksrdquoin IEEE 40th Conference on LocalComputer Networks Clearwater Beach USA 2015 pp 570-578 doi 101109LCN20157366372

[14] M Jia J Cao and W LiangldquoOptimal cloudlet placement and user to cloudletallocation in wireless metropolitan area networksrdquoIEEE Transactions onCloud Computing vol 6 no 25 pp 1-14 2015 doi 101109TCC20152449834

[15] A Ceselli M Premoli and S SeccildquoCloudlet network design optimizationrdquoin IFIP Networking Conference Toulouse France 2015 pp 1-9 doi 101109IFIPNetworking20157145315

[16] Z Xu W Liang W Xu M Jia and S GuoldquoEfficient algorithms for capacitated cloudlet placementsrdquoIEEE Transactions on Parallel and Distributed Systems vol 27 no 10 pp 2866-2880 2016

[17] L Tong Y Li and W GaoldquoA hierarchical edge cloud architecture for mobilecomputingrdquoin IEEE International Conference on Computer CommunicationsSan Francisco USA 2016 pp 1-9 doi 101109INFOCOM20167524340

[18] M Jia W Liang Z Xu and M HuangldquoCloudlet load balancing in wirelessmetropolitan area networksrdquoin IEEE International Conference on ComputerCommunications San Francisco USA 2016 pp 1- 9 doi 101109INFOCOM20167524411

[19] X Sun and N Ansari (2016) Green cloudlet network A distributed green mo

5

times10465

Nume

ricalpr

ofit

Weight of migration delay45 4321

60

55

50

Optimal

OFDOFI

FFDFFI

5

times10471

Nume

ricalcl

oudlet

traffic


70

69

68

Optimal

OFDOFI

FFDFFI

5

times10420

Nume

ricalw

gtmig

ration

delay


15

10

05

OFDOFI

FFDFFI

Optimal

5

6500

Nume

ricalw

gtacc

essdel

ay


6000

5500

5000OFDOFI

FFDFFI

Optimal


Special Topic


8


bile cloud network [Online] Available httpsarxivorgabs160507512[20] L Tong and W GaoldquoApplication aware traffic scheduling for workload

offloading in mobile cloudsrdquoin IEEE International Conference on ComputerCommunications San Francisco USA 2016 pp 1- 9 doi 101109INFOCOM20167524520

[21] L Wang L Jiao D Kliazovich and P BouvryldquoReconciling task assignmentand scheduling in mobile edge cloudsrdquoin IEEE 24th International Conferenceon Network Protocols Singapore 2016 pp 1-6

[22] X Chen L Jiao W Li and X FuldquoEfficient multiuser computation offloading for mobileedge cloud computingrdquoIEEEACM Transactions on Networkingvol 24 no 5 pp 2795-2808 2016 doi 101109TNET20152487344

[23] N M K Chowdhury M R Rahman and R BoutabaldquoVirtual network embedding with coordinated node and link mappingrdquoin IEEE International Conference on Computer Communications San Francisco 2009 pp 783-791 doi101109INFOCOM20095061987

[24] H Huang D Zeng S Guo and H YaoldquoJoint optimization of task mappingand routing for service provisioning in distributed datacentersrdquoin IEEE International Conference on Communications Sydney Australia Jun 2014 pp4196-4201

[25] H Huang P Li S Guo and B YeldquoThe joint optimization of rules allocationand traffic engineering in software defined networkrdquoin IEEE 22nd International Symposium of Quality of Service Hong Kong China 2014 pp 141- 146doi 101109IWQoS20146914313

[26] H Huang S Guo P Li B Ye and I StojmenovicldquoJoint optimization of ruleplacement and traffic engineering for QoS provisioning in software defined networkrdquoIEEE Transactions on Computers vol 64 no 12 pp 3488-3499 2015doi 101109TC20152401031

[27] Gurobi Optimization (2016) Gurobi optimizer reference manual [Online] Available httpwww gurobi com

Manuscript received 20170115

HUANG Huawei (davyhwangcuggmailcom) received his PhD in computer science from the University of Aizu Japan His research interests mainly include network optimization and algorithm designanalysis for wiredwireless networks He isa member of IEEE and a JSPS Research FellowGUO Song (songguopolyueduhk) received his PhD in computer science fromUniversity of Ottawa Canada He is currently a full professor at Department of Computing The Hong Kong Polytechnic University (PolyU) China Prior to joiningPolyU he was a full professor with the University of Aizu Japan His research interests are mainly in the areas of cloud and green computing big data wireless networks and cyberphysical systems He has published over 300 conference and journal papers in these areas and received multiple best paper awards from IEEEACMconferences His research has been sponsored by JSPS JST MIC NSF NSFC andindustrial companies Dr GUO has served as an editor of several journals includingIEEE TPDS IEEE TETC IEEE TGCN IEEE Communications Magazine and Wireless Networks He has been actively participating in international conferences serving as general chairs and TPC chairs He is a senior member of IEEE a senior member of ACM and an IEEE Communications Society Distinguished Lecturer

BiographiesBiographies

Special Topic



Call for Papers

ZTE Communications Special Issue on

Motion and Emotion Sensing Driven by Big Data

Motion and emotions are two critical features of humanpresence and activities Recent developments in the field ofindoor motion and emotion sensing have revealed their potentials in enhancing our living experiences through applications like public safety and smart health However existingsolutions still face several critical downsides such as theavailability (specialized hardware) reliability (illuminationand line of sight constraints) and privacy issues (beingwatched) To this end this special issue seeks original articles describing development relevant trends challengesand current practices in the field of applying Artificial Intelligence to address various issues of motion and emotion sensing brought by theldquoBig datardquo Position papers technologyoverviews and case studies are also welcome

Appropriate topics include but not limited tobullMotion and Emotion ModelTheory driven by Big DatabullMotion and Emotion Sensing Algorithms driven by Big

DatabullMultimodality Data Processing for Motion and Emotion

SensingbullMulti modality Data Mining for Motion and Emotion

SensingbullNovel Motion and Emotion ApplicationsSystems Sup

ported by Big Databull Evaluation Metrics and Empirical Studies of Motion

and Emotion Sensing SystemsbullQuality enhanced and adaptive sensing models driven

by Big Databull Inherent Relationship Modeling between Motion and

Emotions driven by Big DataFirst submission due August 15 2017Peer review September 30 2017Final submission October 15 2017Publication date December 25 2017Submission Guideline

Submission should be made electronically by email inWORD formatGuest Editors

ProfFuji Ren (renistokushima uacjp) University ofTokushima Japan

ProfYixin Zhong (zyxbupteducn) Beijing UniversityOf Posts and Telecommunications China

ProfYu Gu (yugubruceieeeorg) Hefei University ofTechnology China

9


Via an extensive survey in the next section over the existingrelated studies we find out that the challenge to deal with thedynamic characteristics of the mobile cloudlet based networkshas not been well addressed so far Therefore we are motivated to study a fundamental problem in this paper how to update(partially or entirely) the service provisioning solution for a certain group of online mobile application users in a cloudletbased network supposed that the trajectory of each mobile device can be obtained according to the daily routine of each user We try to answer the following two questions 1) when to update the service provisioning solution for each mobile userand 2) how to make a trade off between the admitted trafficrate offloaded by the local edge cloud and the induced accessdelay and VMmigration delay while updating the current configuration

Our study leads to the major contributions as followsbullWe study a service provisioning problem in the cloudlet

based network and try to find a near optimal update schemefor updating the service provisioning solution for each mobileuser at each timeframe if the trajectory of each mobile user isprovided

bullWith the objective to maximize a weighted profit for network operators we first formulate this problem to a nonlinearprogramming problem which is then transformed to a solvableinteger linear programming using the absolute value manipulation techniques

bullBecause of the NP hardness of the formulated problemwe have designed a series of heuristic Algorithms to solve theproblem Extensive numerical simulation results show that thedevised algorithms can yield a near optimal solution We alsoconclude some useful findings via the discussion of evaluationresults

The remaining paper is organized as follows Section 2 reviews the related work Section 3 presents the system modeland gives the problem statement The heuristic algorithms areelaborated in Section 4 Section 5 demonstrates the numericalevaluation Finally Section 6 concludes this paper

2 Related Work

21 Cloudlet Based Edge ComputingRecently edge computing has attracted wide spread re

search efforts [9] [12]-[20] for the mobile computing For instance Xia et al [9] [12] explored a locationbased offloadingproblem aiming to permit requests offloaded to a cloudlet network Then authors proposed several efficient online algorithms that can dynamically handle the requests from users Anovel hierarchical edge cloud architecture constituted withmultiple cloudlets has been proposed in [17] to efficientlyserve the peak loads originating from mobile users Then toadaptively balance the tradeoff between response delay of mobile applications and energy efficiency Tong et al [20] pro

posed both offline and online algorithms to schedule the transmission in mobile cloud computing

In wireless networks the cloudlet placement problem alsohas been studied in [13] [14] [16] [18] For example in awireless metropolitan area network (WMAN) in order to solvethe problem of cloudlet placement Jia et al [14] proposed aplacement scheme for a number of limited cloudlets This approach is proved to greatly improve the mobile cloud performance Similarly Xu et al [13] [16] also focused on the cloudlet placement problem in which capacitated cloudlets need tofind the best deployment locations within a given set of candidate locations The objective is to minimize the average accessdelay between these activated cloudlets and mobile devicesTo this end some approximate algorithms have been devisedwith approximation ratios proved if all the cloudlet servers ownthe identical computing capability22 Task Offloading Using Edge Cloud

Wang et al [21] studied a cost reduction problem in mobileedge clouds by deciding the assignment of mobile offloadedtasks The authors formulated such a problem as a mixed integer program at first Then by introducing admission controlthe problem is simplified and solved by the proposed efficienttwophase scheduling algorithm To solve the decision makingproblem of computation offloading among multiple mobile users Chen et al [22] first formulated the problem as a multiuser computation offloading game and proved that the game always assures a Nash equilibrium Then a game theoretic distributed algorithm is proposed to offload computation intensivetasks over the mobile edge could23 Comparison

Different from all efforts made by existing work mentionedabove this paper particularly studies the service provisioningupdate problem while considering the online and offline statusof mobile users during their trajectories as well as the highlydynamic characteristics of edge cloud networks We find thatthis problem has not been well studied yet To fill this gap inthis paper we strive to design highly effective update schemesof service provisioning for edge cloud network operators

3 Network Model and Problem Statement

31 System ModelThe network that we focus on includes a cloudlet based

edge cloud and a remote private cloud The former networkconsists of a set S of local edge servers Without loss of generality as shown in Fig 2 we assume that a powerful edge serverlocates at each macrocell Therefore a mobile user connectingwith a macrocell base station is equivalent to connecting withthe corresponding local edge server In such a cloudlet basednetwork a set U of mobile application users traverse at differ


Special Topic


2






xtu =igrave

iacute

icirc

iumliumliumliuml




iuml

iuml









NotationUSTDu

Cs

F (u )Rt

u

C tu

Δt

ζΓ t

u

xtu

ztu










VM

VM

hellip


choiceLocalserverVM

t +1t

t +1ts t +1ts tt

Local server


time slot (ts) t


t

Special Topic



VM

3


where Ctu and Rt



∆t =sumu isinU



Γt =sumt isin T




Duxtu -sum


stsumu isinU



tu







|| xtu - xt - 1u = 0 (5)












Special Topic


4




















uprime then13 zt


uprime then15 zt



uprime then20 zt


uprime then22 zt

uprime larr 0




Special Topic




1 11 0 10

0 01 1 1hellip

1 hellip



a1 a2

b1 b2 b3

a3hellip hellip

hellip hellip

t -1 t +1tTime slot


5

















15001200900600

times104757065605550

Nume

ricalpr

ofit


OptimalOFDOFIFFDFFI

15001200900600

times104

Nume

ricalcl

oudlet

traffic


848280787674

OptimalOFDOFIFFDFFI

15001200900600

times10416

08

Nume

ricalw

gtacc

essdel

ay


14

12

10

OptimalOFDOFI

FFDFFI

15001200900600

8000

0Nume

ricalw

gtmig

ration

delay


6000

4000

2000

OptimalOFDOFIFFDFFI


Special Topic


6

















5

times104

70

Nume

ricalpr

ofit


60

50

40

30 4321

Optimal

OFDOFI

FFDFFI

5

times104

71

Nume

ricalcl

oudlet

traffic


70

69

68

67

Optimal FFDFFI

OFDOFI

5

times104

Nume

ricalw

gtacc

essdel

ay


3

2

1

0

OptimalFFDFFI

OFDOFI

5

Nume

ricalw

gtmig

ration

delay


30003500

2500200015001000

Optimal

FFDFFI

OFDOFI

Special Topic



7



(a) Profit vs w 2

























5

times10465

Nume

ricalpr

ofit


60

55

50

Optimal

OFDOFI

FFDFFI

5

times10471

Nume

ricalcl

oudlet

traffic


70

69

68

Optimal

OFDOFI

FFDFFI

5

times10420

Nume

ricalw

gtmig

ration

delay


15

10

05

OFDOFI

FFDFFI

Optimal

5

6500

Nume

ricalw

gtacc

essdel

ay


6000

5500

5000OFDOFI

FFDFFI

Optimal


Special Topic


8














Special Topic



Call for Papers
















9






xtu =igrave

iacute

icirc

iumliumliumliuml




iuml

iuml









NotationUSTDu

Cs

F (u )Rt

u

C tu

Δt

ζΓ t

u

xtu

ztu










VM

VM

hellip


choiceLocalserverVM

t +1t

t +1ts t +1ts tt

Local server


time slot (ts) t


t

Special Topic



VM

3


where Ctu and Rt



∆t =sumu isinU



Γt =sumt isin T




Duxtu -sum


stsumu isinU



tu







|| xtu - xt - 1u = 0 (5)












Special Topic


4




















uprime then13 zt


uprime then15 zt



uprime then20 zt


uprime then22 zt

uprime larr 0




Special Topic




1 11 0 10

0 01 1 1hellip

1 hellip



a1 a2

b1 b2 b3

a3hellip hellip

hellip hellip

t -1 t +1tTime slot


5

















15001200900600

times104757065605550

Nume

ricalpr

ofit


OptimalOFDOFIFFDFFI

15001200900600

times104

Nume

ricalcl

oudlet

traffic


848280787674

OptimalOFDOFIFFDFFI

15001200900600

times10416

08

Nume

ricalw

gtacc

essdel

ay


14

12

10

OptimalOFDOFI

FFDFFI

15001200900600

8000

0Nume

ricalw

gtmig

ration

delay


6000

4000

2000

OptimalOFDOFIFFDFFI


Special Topic


6

















5

times104

70

Nume

ricalpr

ofit


60

50

40

30 4321

Optimal

OFDOFI

FFDFFI

5

times104

71

Nume

ricalcl

oudlet

traffic


70

69

68

67

Optimal FFDFFI

OFDOFI

5

times104

Nume

ricalw

gtacc

essdel

ay


3

2

1

0

OptimalFFDFFI

OFDOFI

5

Nume

ricalw

gtmig

ration

delay


30003500

2500200015001000

Optimal

FFDFFI

OFDOFI

Special Topic



7



(a) Profit vs w 2

























5

times10465

Nume

ricalpr

ofit


60

55

50

Optimal

OFDOFI

FFDFFI

5

times10471

Nume

ricalcl

oudlet

traffic


70

69

68

Optimal

OFDOFI

FFDFFI

5

times10420

Nume

ricalw

gtmig

ration

delay


15

10

05

OFDOFI

FFDFFI

Optimal

5

6500

Nume

ricalw

gtacc

essdel

ay


6000

5500

5000OFDOFI

FFDFFI

Optimal


Special Topic


8














Special Topic



Call for Papers
















9


where Ctu and Rt



∆t =sumu isinU



Γt =sumt isin T




Duxtu -sum


stsumu isinU



tu







|| xtu - xt - 1u = 0 (5)












Special Topic


4




















uprime then13 zt


uprime then15 zt



uprime then20 zt


uprime then22 zt

uprime larr 0




Special Topic




1 11 0 10

0 01 1 1hellip

1 hellip



a1 a2

b1 b2 b3

a3hellip hellip

hellip hellip

t -1 t +1tTime slot


5

















15001200900600

times104757065605550

Nume

ricalpr

ofit


OptimalOFDOFIFFDFFI

15001200900600

times104

Nume

ricalcl

oudlet

traffic


848280787674

OptimalOFDOFIFFDFFI

15001200900600

times10416

08

Nume

ricalw

gtacc

essdel

ay


14

12

10

OptimalOFDOFI

FFDFFI

15001200900600

8000

0Nume

ricalw

gtmig

ration

delay


6000

4000

2000

OptimalOFDOFIFFDFFI


Special Topic


6

















5

times104

70

Nume

ricalpr

ofit


60

50

40

30 4321

Optimal

OFDOFI

FFDFFI

5

times104

71

Nume

ricalcl

oudlet

traffic


70

69

68

67

Optimal FFDFFI

OFDOFI

5

times104

Nume

ricalw

gtacc

essdel

ay


3

2

1

0

OptimalFFDFFI

OFDOFI

5

Nume

ricalw

gtmig

ration

delay


30003500

2500200015001000

Optimal

FFDFFI

OFDOFI

Special Topic



7



(a) Profit vs w 2

























5

times10465

Nume

ricalpr

ofit


60

55

50

Optimal

OFDOFI

FFDFFI

5

times10471

Nume

ricalcl

oudlet

traffic


70

69

68

Optimal

OFDOFI

FFDFFI

5

times10420

Nume

ricalw

gtmig

ration

delay


15

10

05

OFDOFI

FFDFFI

Optimal

5

6500

Nume

ricalw

gtacc

essdel

ay


6000

5500

5000OFDOFI

FFDFFI

Optimal


Special Topic


8














Special Topic



Call for Papers
















9




















uprime then13 zt


uprime then15 zt



uprime then20 zt


uprime then22 zt

uprime larr 0




Special Topic




1 11 0 10

0 01 1 1hellip

1 hellip



a1 a2

b1 b2 b3

a3hellip hellip

hellip hellip

t -1 t +1tTime slot


5

















15001200900600

times104757065605550

Nume

ricalpr

ofit


OptimalOFDOFIFFDFFI

15001200900600

times104

Nume

ricalcl

oudlet

traffic


848280787674

OptimalOFDOFIFFDFFI

15001200900600

times10416

08

Nume

ricalw

gtacc

essdel

ay


14

12

10

OptimalOFDOFI

FFDFFI

15001200900600

8000

0Nume

ricalw

gtmig

ration

delay


6000

4000

2000

OptimalOFDOFIFFDFFI


Special Topic


6

















5

times104

70

Nume

ricalpr

ofit


60

50

40

30 4321

Optimal

OFDOFI

FFDFFI

5

times104

71

Nume

ricalcl

oudlet

traffic


70

69

68

67

Optimal FFDFFI

OFDOFI

5

times104

Nume

ricalw

gtacc

essdel

ay


3

2

1

0

OptimalFFDFFI

OFDOFI

5

Nume

ricalw

gtmig

ration

delay


30003500

2500200015001000

Optimal

FFDFFI

OFDOFI

Special Topic



7



(a) Profit vs w 2

























5

times10465

Nume

ricalpr

ofit


60

55

50

Optimal

OFDOFI

FFDFFI

5

times10471

Nume

ricalcl

oudlet

traffic


70

69

68

Optimal

OFDOFI

FFDFFI

5

times10420

Nume

ricalw

gtmig

ration

delay


15

10

05

OFDOFI

FFDFFI

Optimal

5

6500

Nume

ricalw

gtacc

essdel

ay


6000

5500

5000OFDOFI

FFDFFI

Optimal


Special Topic


8














Special Topic



Call for Papers
















9

















15001200900600

times104757065605550

Nume

ricalpr

ofit


OptimalOFDOFIFFDFFI

15001200900600

times104

Nume

ricalcl

oudlet

traffic


848280787674

OptimalOFDOFIFFDFFI

15001200900600

times10416

08

Nume

ricalw

gtacc

essdel

ay


14

12

10

OptimalOFDOFI

FFDFFI

15001200900600

8000

0Nume

ricalw

gtmig

ration

delay


6000

4000

2000

OptimalOFDOFIFFDFFI


Special Topic


6

















5

times104

70

Nume

ricalpr

ofit


60

50

40

30 4321

Optimal

OFDOFI

FFDFFI

5

times104

71

Nume

ricalcl

oudlet

traffic


70

69

68

67

Optimal FFDFFI

OFDOFI

5

times104

Nume

ricalw

gtacc

essdel

ay


3

2

1

0

OptimalFFDFFI

OFDOFI

5

Nume

ricalw

gtmig

ration

delay


30003500

2500200015001000

Optimal

FFDFFI

OFDOFI

Special Topic



7



(a) Profit vs w 2

























5

times10465

Nume

ricalpr

ofit


60

55

50

Optimal

OFDOFI

FFDFFI

5

times10471

Nume

ricalcl

oudlet

traffic


70

69

68

Optimal

OFDOFI

FFDFFI

5

times10420

Nume

ricalw

gtmig

ration

delay


15

10

05

OFDOFI

FFDFFI

Optimal

5

6500

Nume

ricalw

gtacc

essdel

ay


6000

5500

5000OFDOFI

FFDFFI

Optimal


Special Topic


8














Special Topic



Call for Papers
















9

















5

times104

70

Nume

ricalpr

ofit


60

50

40

30 4321

Optimal

OFDOFI

FFDFFI

5

times104

71

Nume

ricalcl

oudlet

traffic


70

69

68

67

Optimal FFDFFI

OFDOFI

5

times104

Nume

ricalw

gtacc

essdel

ay


3

2

1

0

OptimalFFDFFI

OFDOFI

5

Nume

ricalw

gtmig

ration

delay


30003500

2500200015001000

Optimal

FFDFFI

OFDOFI

Special Topic



7



(a) Profit vs w 2

























5

times10465

Nume

ricalpr

ofit


60

55

50

Optimal

OFDOFI

FFDFFI

5

times10471

Nume

ricalcl

oudlet

traffic


70

69

68

Optimal

OFDOFI

FFDFFI

5

times10420

Nume

ricalw

gtmig

ration

delay


15

10

05

OFDOFI

FFDFFI

Optimal

5

6500

Nume

ricalw

gtacc

essdel

ay


6000

5500

5000OFDOFI

FFDFFI

Optimal


Special Topic


8














Special Topic



Call for Papers
















9



(a) Profit vs w 2

























5

times10465

Nume

ricalpr

ofit


60

55

50

Optimal

OFDOFI

FFDFFI

5

times10471

Nume

ricalcl

oudlet

traffic


70

69

68

Optimal

OFDOFI

FFDFFI

5

times10420

Nume

ricalw

gtmig

ration

delay


15

10

05

OFDOFI

FFDFFI

Optimal

5

6500

Nume

ricalw

gtacc

essdel

ay


6000

5500

5000OFDOFI

FFDFFI

Optimal


Special Topic


8














Special Topic



Call for Papers
















9














Special Topic



Call for Papers
















9

Adaptive Service Provisioning for Mobile Edge Cloud - ZTE

Documents