Inter national J our nal of Electrical and Computer Engineering (IJECE) V ol. 9, No. 1, February 2019, pp. 512 524 ISSN: 2088-8708, DOI: 10.11591/ijece.v9i1.pp512-524 512 Bin packing algorithms f or virtual machine placement in cloud computing: a r e view K umaraswamy S 1 and Mydhili K Nair 2 1 Department of Computer Science and Engineering, Global Academy of T echnology , Beng aluru 560098, India 2 Department of Information Science and Engineering, Ramaiah Institute of T echnology , Beng aluru 560054, India Article Inf o Article history: Recei v ed Dec 31, 2017 Re vised Jul 25, 2018 Accepted Jul 29, 2018 K eyw ords: Cloud computing V irtual machine placement Bin packing ABSTRA CT Cloud computing has become more comm ercial and f amiliar . The Cloud data centers ha v e huge challenges to maintain QoS and k eep the Cloud perf ormance high. The placing of virtua l machines among ph ysical machines in Cloud is significant in opti- mizing Cloud performance. Bin packing based algorithms a re most used concept to achie v e virtual machine placement (VMP). This paper presents a rigorous surv e y and comparisons of the bin packing based VMP methods for the Cloud computing en viron- ment. V arious methods are discussed and the VM placement f actors in each methods are analyzed to understand the adv antages and dra wbacks of each method. The scope of future research and studies are also highlighted. Copyright c 2019 Institute of Advanced Engineering and Science . All rights r eserved. Corresponding A uthor: K umarasw amy S, Dept. of Computer Science and Engineering, Global Academy of T echnology , Rajarajeshw ari Nag ar , Beng aluru - 560098 India. +919886363412 Email: sksw amy99@gmail.com 1. INTR ODUCTION 1.1. Back Gr ound In recent years Cloud computing architectures ha v e recei v ed an increasing attention due to their great promises in enabling distrib uted computing paradigm [1]. It pro vides pool of computing resources enabling the Cloud user’ s application to use it [2]. These resources can be rented to users or customers by Cloud Service Pro viders (CSPs) on pay-as-you-go model, lik e public utility such as g as, w ater and electricity [3, 4]. V irtualization is an enabled technology for cloud computing. It pro vides the fle xibility to cloud. It minimizes the ener gy cost, maximizes the CPU utilization, and maximizes the usage of memory/disk. Multiple VMs on a PM may de grades the performance of PM or o v erutilize the PM. T o o v ercome these issues, virtualization layer pro vides the mobility to pl ace or mo v e VM to other PMs within cloud. The plan for placing VMs among PMs plays an important role in optimi zing cloud perform ance is c alled V irtual Machine Placement (VMP) which finds the optimal VM placement/mapping with PMs with v arious objecti v es and constraints. In principle, the VMP problem is related to the classical Bin P acking (BP) problem, where the aim is to pack or place or map set of items / objects of dif ferent size into a minimum number of unit-capacity bins. It is pro v ed in literature that both BP and VMP are NP-hard problems [5]. If there are ’m’ PMs and ’n’ VMs, then, m n placement solutions. If m and n are v ery big numbers, solutions are huge. One can easily say that VMP is as hard as BP . The dif ferences between Classical BP and BP-based VMP are detailed in T able 1 J ournal Homepage: http://iaescor e .com/journals/inde x.php/IJECE Evaluation Warning : The document was created with Spire.PDF for Python.
IJECE ISSN: 2088-8708 513 Since VMP problem is an NP hard problem, e xact algorithms tak es more time to get solutions, hence this problem can be ef fecti v ely approximated using approximation algorithms such as greedy algorithms, ge- netic algorithms to accomplish results. These pro vide a solution, which though v ery good in most cases, may not be an optimal solution b ut near -optimal. P acking algorithms lik e First Fit, Best Fit, W orst Fit , First Fit Decreasing (FFD), Best Fit Dec reasing (BFD), W orst Fit Decreasing (WFD), greedy algorithms etc., are used to place VMs optimally . Also, authors in Ref. [6, 7, 8, 9] present fe w heuristics lik e First-Fit (FF), Best-Fit (BF), and W orst-Fit (WF) for load balancing, some of which could be tuned to serv e the purpose of VMP . In the FF , the VM is placed in the PM which is selected as first machine with a v ailable capacity . In BF , the name itself says fit to be best when left o v er space is least after placing the VM in the PM. In the WF , left o v er space is more compared to pre vious methods when the VM is pack ed in the PM. In First Fit Decreasing (FFD), Best Fit Decreasing (BFD), and W orst Fit Decreasing (WFD) heuristics VMs are sorted in decreasing order before packing. Once sorted, the VMs are placed according to original heuristics. T able 1. Dif ferences between Classical BP and BP-based VMP schemes Classical BP BP-based VMP Item size is fix ed once placed Item size will v ary e v en after place- ment due to SLAs mono-objecti v e optimization e.g., to minimize number of bins multi-objecti v e optimization e.g., minimize number of bins and max- imize VMs performance single-dimensional resource pack- ing e.g., CPU or memory multi-dimensional resource pack- ing e.g., CPU and memory The VMP problem can also be considered as a multi-dimensional or multi-capacity problem wi thin a PM [10]. In Multi-Capacity Bin P acking (MCBP) problem, once item is allocated to bin, its resources lik e CPU, memory , bandwidth etc., cannot be allocated or a v ailable to other items in a bin. In multi-dimensional problem, capac ity is shared and is not dedicated to single item. This leads to conflict among items to get dedicated resources. Figure 1 sho ws the comparison between multi-dimensional and multi-capacity bin packing. Figure 1. multi-dimensional vs. multi-capacity bin packing [10] 1.2. Pr oblem Description There has been considerable contrib utions in the literature re g arding BP problem and its applicat ion for the VMP issue. Man y approximation techniques to solv e BP problem for VMP i ssue ha v e been presented; ho we v er , comparati v e study on all these techniques is still lacking in the literature. Comparing all the major techniques for VMP issue through BP problem modeling, w ould aid in understanding the relati v e merits and design limi tations of all these techniques, and w ould also aid in framing future problems in this area. The main focus of this paper is to pro vide comprehensi v e surv e y and simulation study on v arious approximation technique–re g arding BP solution techniques–used for VMP issue; also, to pro vide the merits and limitations for these techniques, and to identify the future research problems. Bin pac king algorithms for virtual mac hine ... (K umar aswamy S) Evaluation Warning : The document was created with Spire.PDF for Python.
514 ISSN: 2088-8708 1.3. Contrib utions In this surv e y paper , the important considerations and challenges such as: resource utilization, rel iabil- ity e.t.c, to ef ficiently design BP based solution for VMP issue are outlined. The three important approximation solutions for BP problem: First Fit , Best Fit and Best Fit Decreasing approaches, are described; also, the recently e xtended techniques for these approximation solutions presented in the literature are also described. Heuristics for VMP issue, usually do not g ua rantee performa n c e bounds; ne v ertheless, the y pro vide empirically demonstrated performance merits, and dif ferent heuristic techniques used for VMP issue are also outlined. Rel- ati v e simulation oriented performance comparison for dif ferent approximation solutions for BP problem based VMP issue is pres ented, and corresponding merits and limitations are outlined. Finally , the future directions for the VMP issue are highlighted. This paper is or g anized as follo ws; we start with a discussion on the considerations and challenges for VMP (Section 2), follo wed by BP-based problem definition in Section 3. Section 4 presents BP based VMP . Challenges of BP algorithms are presented in section 5. Section 6 presents performance e v aluation of VMP schemes. This is follo wed by future studies (Section 7), and conclusion (Section 8). 2. CONSIDERA TIONS AND CHALLENGES OF VMP Se v eral placement problem e xist with re g ard to determining as to where data needs to be stored and where the job in VMs can be e x ecuted. Here we discuss the considerations and challenges of VMP . 2.1. Considerations of VMP There are se v eral f actors in v olv ed in deciding as to when and where to place or reallocate VMs for performing computations in Cloud computing en vironment. The main f actors are as follo ws. (a) Performance: T o see that the utilization of ph ysical infrast ructures are impro v ed, dat a centers (DC) need t o emplo y virtualization that could f acilitate lar ge number of applications to run simultane- ously [11]. (b) Cost: Recent trend in Cloud mark et sho ws that dynamic pricing schemes utilization is being in- creased [12] to reduce the in v estment in the DC. There fore, internal cost for VMP also needs to be considered. (c) Locality: If accessibility and usability for users need to be considered, VMs should be closely located to users. (d) Reliability and continuous a v ailability: Reliability and a v ailability of the dif ferent DC, and its e xpected usage frequenc y must be tak en into account. 2.2. Challenges of VMP V ariation of scenarios in which the applications are to be deplo yed need rele v ant parameters consid- erations that results in the follo wing challenges while placing VMs. (a) Firstly , non-e xistence of generic model for representing v arious scenarios for resource scheduling. (b) Secondly , parameterization of model for bigger problem size. (c) Thirdly , the VMP problem is typically mapped to multiple-knapsack problem, belonging to a cate- gory of NP hard problems [13]. Thus, tradeof fs between e x ecution time and quality of solution is an important issue to be tackled, gi v en the size of real life DC. 3. BIN P A CKING PR OBLEM Let, S = S 1 ; S 2 ; :::S m be the set of bins and all of them ha v e the same size V . Let, a 1 ; a 2 ; ::::a n be the set of n items that need to be pack ed in the bins. The task is to disco v er inte ger number of bins B and a B partition of the set (1 ; 2 ; ::n ) gi v en by , S 1 [ S 2 [ ; ::: [ S B , so that, P i 2 S k a i V 8 k = 1 ; 2 ; ::B . The optimal IJECE V ol. 9, No. 1, February 2019 : 512 524 Evaluation Warning : The document was created with Spire.PDF for Python.
IJECE ISSN: 2088-8708 515 solution is to find minimal B . The inte ger Liner Programming formulation of Bin P acking problem is sho wn in Equation 1. min B = n X i =1 y i (1) subject to B 1 , P n j =1 a j x ij V y i , 8 i 2 (1 ; 2 ; ::n ) , P n i =1 x ij = 1 , 8 j 2 (1 ; 2 ; ::n ) , y i 2 (0 ; 1) , 8 i 2 (1 ; 2 :::n ) , x ij 2 (0 ; 1) , 8 i 2 (1 ; 2 ; :::n ) , 8 j 2 (1 ; 2 ; :::n ) Here, y i = 0 if bin i is not used, otherwise, y i = 1 and x ij = 0 if item j is put into bin i , otherwise, x ij = 1 . 4. BP-B ASED VMP W e vie w the problem of placement of VMs in data centers(DC)s as similar to a BP problem. W e visulaize the classic BP Problem [5] in the conte xt of VMP in the follo wing w ay: VMs (objects) of dif ferent sizes are placed in hosts (bins ) to ensure minimum hosts being emplo yed for placing all VMs. In other w ords, the problem statement can be stated as follo ws: W ith ’x’ PMs being made a v ailable with resource capacities in terms of memory , CPU and netw ork bandwidth resources, we require ’y’ VMs to be placed such that the total resource requirement of the VMs placed on a PM should not e xceed its capacity . Se v eral BP-based VMP algorithms are used in finding optimal VMP are discussed ne xt. 4.1. First Fit (FF) appr oach The First Fit Bin P acking algorithm is described in Algorithm 1. The items are scanned in an y order and e v ery item is attempted to be placed in the a v ailable bins sequentially . If it does not get fit into e xisting bins then, ne w bin is created to place the item. This algorithm achie v es an approximation ratio of 2 . Algorithm 1 First Fit Algorithm f or i = 1 to n do f or j = 1 to m do if Item i with size a i fits in Bin S j then Place the item a i in Bin S j . j + + end if if Item i does not fit in Bin S j then j + + end if end f or if Item i does not fit in an y Bin then Open a ne w bin and place the item in it. m = m + 1 end if end f or Authors in [14] introduce a dynamic serv er and consolidation management algorithm that meas ures historical data, forecas ts future demand and re-maps VMs to PMs. T o forecast future demand, time series method is used. The BP heuristic method based on FF approximation is used to perform mapping between VMs and PMs. Authors in [15] in v estig ated VMP problem in a Cloud as a generalized assignment problem (GAP) and formulated a multi-le v el generalized assignment problem (MGAP). It uses FF heuristic to solv e MGAP and maximizes the profit under the SLA. Bin pac king algorithms for virtual mac hine ... (K umar aswamy S) Evaluation Warning : The document was created with Spire.PDF for Python.
516 ISSN: 2088-8708 In the FF approach, the VMs are placed in the host where the y first fit, according to a predefined order between acti v e hosts. The FF algorithm [16, 5] accomplishes this by acti v ating a single host at a time, as it is filled up with VMs. This results in DC sa ving ener gy , since only hosts that contain some VMs need to be switched on. Ho we v er , since host resource usage le v els tend to be close to the maximum, VM migration is more lik ely to happen in the presence of elasticity , that de grades the performance, besides consuming some e xtra ener gy . 4.2. Best Fit (BF) appr oach The Best Fit technique is described in Algorithm 2. The algorithm w orks similar to First Fit, b ut, the items are placed in that bin which has the lo west residual capacity . The Best Fit algorithm achie v es an approximation ratio of 2 . Algorithm 2 Best Fit Algorithm Let r 1 = V 0 ; r 2 = V 0 ; ::r m = V 0 be the initial residual capacity of each bin. f or i = 1 to n do f or j = 1 to m do if Item i with size a i fits in Bin S j then Calculate scor e ij = r j a i j + + end if end f or Fit item i into that bin which has the lo west scor e ij . r j = r j a i if Item i does not fit in an y Bin then Open a ne w bin and place the item in it. m = m + 1 end if end f or Authors in Ref. [17] consider greedy algorithm with tw o-stage to solv e VMP for maximizing ener gy- ef ficienc y and netw ork performance. In first stage, i f no congestion, ener gy optimization took priority . Authors combined minimum cut hierarchical clustering with Best Fit(BF) algorithm, enabled VMs with lar ge traf fic to be placed on the same PM or the same access switch. In second stage, if there is congestion, the y applied local search algorithm to minimize Maximum Link Utilization(MLU) and link congestion with equal distrib ution of netw ork traf fic. Authors in [18] consider F at-tree DC topology to propose a solution for VMP problem and migration. The solut ion reduces the po wer cos t and job delay by consolidating VMs to a fe w PM s and mi grating VMs to close locations. In VMP stage, it considers tw o mechanisms, namely random placement mechanism, and po wer -ef ficient placement. In random placement, PM is selected randomly to place a VM. In po wer -ef ficient mechanism, three algorithms, i.e., FF , BF and WF are discussed; an OpenFlo w controller uses these algorithms to find proper PM to deplo y a VM based on the weighted dif ference. The authors in [19] discuss a static disk threshold-based migration algorithm aiming at optimizing the performance of the VMs. T o pre v ent the de gradation and congestion problems in the netw ork, authors in [20] in v estig ate a VM placement method, called ener gy ef ficienc y and quality of service a w are VM placement (EQVMP). T o predict the future beha vior of a VM on destination host, authors in Ref. [21, 17] present a VM placement scheme named Backw ard Speculati v e Placement (BSP), that monitors the historical demand traces of the deplo yed VMs. 4.3. First Fit Decr easing (FFD) and Best Fit Decr easing(BFD) appr oach The Fit Fit Decreasing and Best Fit decreasing techniques w ork similar to First Fit and Best Fit tech- niques respecti v ely . But, the items are arranged in decreasing order of their sizes to impro v e the approximation IJECE V ol. 9, No. 1, February 2019 : 512 524 Evaluation Warning : The document was created with Spire.PDF for Python.
IJECE ISSN: 2088-8708 517 ratio. The approximation ratio of both these techniques is < 2 . Algorithms 3 and 4 describe these tw o tech- niques. Algorithm 3 First Fit Decreasing Algorithm Item Set (1 ; 2 ; ::n ) is arranged in decreasing order of their sizes. a 1 a 2 ::: a n f or i = 1 to n do f or j = 1 to m do if Item i with size a i fits in Bin S j then Place the item a i in Bin S j . j + + end if if Item i does not fit in Bin S j then j + + end if end f or if Item i does not fit in an y Bin then Open a ne w bin and place the item in it. m = m + 1 end if end f or Algorithm 4 Best Fit Decreasing Algorithm Item Set (1 ; 2 ; ::n ) is arranged in decreasing order of their sizes. a 1 a 2 ::: a n Let r 1 = V 0 ; r 2 = V 0 ; ::r m = V 0 be the initial residual capacity of each bin. f or i = 1 to n do f or j = 1 to m do if Item i with size a i fits in Bin S j then Calculate scor e ij = r j a i j + + end if end f or Fit item i into that bin j which has the lo west scor e ij . r j = r j a i if Item i does not fit in an y Bin then Open a ne w bin and place the item in it. m = m + 1 end if end f or In the simplified model, since only single dimension is considered (e.g., CPU usage), VMs could be ordered according to its resource requirement. In a complicated model, se v eral dimensions are considered to establish a ranking amongst the set of VMs. The ranking is established in the follo wing manner . (a) The v olume of a VM is calculated by mult iplying its demands in all dimens ions (by emplo ying the v olume method). (b) The rank of each VM is determined with respect to a reference host which is, normally , the least occupied acti v e host or the last one to be acti v ated. In this approach, during rank calculation, the VM placement problem can be modeled as an instance of the 0-1, or Binary , Knapsack Problem [17]. Bin pac king algorithms for virtual mac hine ... (K umar aswamy S) Evaluation Warning : The document was created with Spire.PDF for Python.
518 ISSN: 2088-8708 F or complicated model, directly we can not apply the FFD, generalization of FFD algorithm is essential. Con- v ersion of multi dim ension in to single dimension is done. Such FFD algorithms are called FFD-based algo- rithms. Research in Refs. [22] considers V irtual Machine Monitor (VMMs) CPU c ycles and migration o v er - head in designing VMP problem, to maximize the performance of applications and reduce the number of PMs. The e v aluation of the problem is done in three dif ferent FFD based heuristics viz. Dot Product, Euclidean Distance and Resource Imbalance V ector method. It reduces the number of migrations by more than 84%. Authors [23] considered deterministic resources viz., CPU, memory , po wer consumption and netw ork bandwidth as a stochastic resource for formulating Multi-Dimensional Stochastic V irtual Machine Placement Problem (MSVP). Since it is a NP-hard problem, authors proposed polynomial time algorithm called Max- Min Multi-Dimensional Stochastic Bin P acking (M3SBP) to maximize the minimum utilizati on ratio of all the resources of a PMs in lar ge scale data centers. This algorithm is inspired by First Fit Decreasing (FFD) and Dominant Resource First (DRF) This paper [24] used data model for each VM, the data model accurately gi v es the resource usage of the VM. Based on this forecast, VM placement algorithm with po wer -a w are BFD heuristic algorithm is designed. The simulation results of this algorithm sho w the ef fecti v e reduction in po wer consumption, number of VM migrations, and pre v ent SLA violations. Authors [25] proposed the ener gy-ef ficient and QoS a w are VM placement (EQVMP) mechanism. This mechanism is a combination of three modules viz., hop reduction, ener gy sa ving and load balancing. The ener gy sa ving module uses VM placement technique inspired by Best-Fit Decreasing (BFD) and max-min multi-dimensional stochastic bin packing (M3SBP), so that it minimizes the number of serv er in the datacenter without SLA violation In [26], the First Fit Decreasing algorithm is used to solv e the VMP problem. In this, for each bin priority is pro vided. The highest priority bins consume too man y resources or too fe w resources. So, the virtual machines in these bins will be subjected to placement. 4.4. Other Heuristics Fe w of the authors consider dif fer ent packing t yp e for VM consolidation in their w orks. F or e.g., in Ref. [5], authors consider consolidation problem as a r andom variable pac king problem with bandwidth constraints, and solv e it by approximation algorithm. Ref. [10] considers VM consolidation problem as multi- capacity stochastic bin packing problem. It proposes heuristic method in order to increase the ef ficienc y of placement algorithm. [9] propose similar w ay of VM placement using bin packing, to increase resource utiliza- tion and profit of C SP . Fe w authors also consider the Ne xt Fit (NF) approach, in which the VM is placed into the current PM (if VM resource demands are satisfied in current PM), otherwise ne w PM is started. It pro vides performance ratio of 2. In the paper [5], authors suggest ener gy sa ving by minimizing the number of idle resources in a cloud computing en vironment. It has been achie v ed by f ar -reaching e xperiments, so as to quantify the performance of the suggested algorithm. The same has also bee n compared with the FCFSMaxUtil and Ener gy a w are T ask Consolidation (ETC) algorithm. The outcomes ha v e sho wn that the suggested algorithm surpass the FCFSMaxUtil and ETC algorithm in terms of the CPU utilization and ener gy consumption. According to Mark o v Decision Proces s (MDP) and based on reinforcement learning for auto-scaling resources, authors[27] propose an automatic resource pro visioning system. Proposed system simulation re- sults sho ws bett er performance when compared to similar approaches with respect to rate of Service Le v el Agreement (SLA) violation and stability . The authors [28] proposed optimal technique based on four -adapti v e threshold model to reduce ener gy consumption. Hosts are clustered into v e clusters using K-Means Clustering algorithm viz., 1. Hosts with lo w load 2. Hosts with light load 3. Hosts with middle load 4. Hosts with hea vy load and 5 hosts with hea vy load. VMs are migrated between these clusters on the basis threshold v alues using mathematical modeling approach to reduce ener gy consumption. T able 2 summarizes the already discussed heuristics in pre vious section by comparing their basi c at- trib utes lik e: the problem type, the type of heuristics used, the objecti v e type (ener gy or bandwidth or QoS a w are), type of placement, and finally objecti v e function composite or single. IJECE V ol. 9, No. 1, February 2019 : 512 524 Evaluation Warning : The document was created with Spire.PDF for Python.
IJECE ISSN: 2088-8708 519 T able 2. Bin P acking Algorithms Surv e yed P aper Pr oblem type Heuristics Ener gy awar e Band width awar e QoS awar e Static/ Dy- namic place- ment Objecti v e function [25] Ener gy-ef ficient and QoS a w are VMP problem BFD Y es Y es Y es Dynamic Dynamic [24] Cloud serv er consolidation problem BFD Y es No Y es Dynamic Composite [18] Po wer -ef ficient VM placement and migration problem FF ,BF ,WF Y es Y es No Dynamic Composite [17] multi- dimensional resource con- straints packing problem BF based Y es Y es No Dynamic Composite [15] multile v el gener - alized assignment problem (MGAP) FF No Y es Static and Dy- namic single no [14] Dynamic re- source allocation problem FF N A No Y es Dynamic Single [22] N A FFD-based N A No Y es Dynamic Single [23] multi dimen- sional stochastic VM Placement MSVP FFD-based N A Y es Y es Dynamic Single [10] multi-capacity stochastic bin packing opti- mization problem Hierarchical based Y es Y es No Dynamic Composite [5] VM consolida- tion problem Random V ariable P ack- ing(R VP) No Y es No Dynamic Composite [9] Deterministic bin packing W orst Fit Y es No Y es Static N A [26] VM consolida- tion problem for po wer sa ving e xtended FFD Y es No No rank- based N A Bin pac king algorithms for virtual mac hine ... (K umar aswamy S) Evaluation Warning : The document was created with Spire.PDF for Python.
520 ISSN: 2088-8708 5. CHALLENGES OF BP ALGORITHMS Some of the important challenge of BP algorithms are discussed as follo ws. (a) Interference between items: Due to the multi-dimensional packing of resources used by se v eral VMs at same t ime, there is content ion for resources among VMs. There may be af finity between tw o VMs due to which the y may be required to be placed together . There can be v arious possible reasons for such af finity . F or e xample, the netw ork traf fic between tw o communicating VMs may suggest that the y be placed together so as to reduce the netw ork o v erheads. There may be b usiness constraints lik e the requirement of web-serv er and database layers to be together . The contention may af fect the performance of co-located VMs. There are se v eral approaches that ha v e been proposed to miti g ate the ef fect of contention such as resource isolat ion among VMs and optimal mapping of VMs and PMs. There could also be interference between tw o VMs which requires that the y should not be placed together . This may be due to technical constraints. F or e xample, say there is a resource (some remote data center) mirrored at tw o places, that the tw o VMs need to access. The VMs are pro- grammed to access the nearest resource. No w , there may be a technical requirement to k eep the tw o VMs separately , so that the y do not access the same cop y . There may be a disk contention: tw o VMs trying to access the same data on disk. Interference can also be due to b usiness constraints lik e not k eeping VMs of competi tors together . Or , s o that one of the VM remains a v ailable e v en if the ph ysical host hosting the other VM f ails. (b) Item size and bin size: VM capacity cannot be al w ays static o v er its lifetime due to SLAs. Dy- namic size of VM can be considered in BP based algorithm. This is also true for bin/PM size. F or v ariable-sized BP , there are man y open questions. In terms of asymptotic w orst-case ratio, follo wing question arises: i. which combination of sizes produces the smallest w orst-case ratio? ii. what can be said about the problem with at least three bin sizes? iii. in terms of absolute w orst-case ratio, what is a lo wer bound for of f-line algorithms? i v . ho w to design an optimal on-line algorithm? (c) P artial packing: Existing research w orks do not discuss filling PMs/bins in optimal w ay , lea ving lar ge resource fra g m ents or residue. Hence, resources are underutilized. Better BP algorithms need to be designed so to a v oid resource fragments. (d) No migration: most of BP based w orks doesn’ t supports migration in VMs. T asks cannot be relo- cated once the y ha v e been allocated to an y processor (i.e. no migration). 6. PERFORMANCE EV ALU A TION In this section, the simulation analysis results for analyzing the relati v e performance of dif ferent VMP algorithms are presented. The simulation w as carried out through open source simulation tool kit: CloudSim. T w o performance metrics are utilized in simulation study: No of Se v ers and CPU utilization. The first metric indicates the number of serv ers on which all the VMs are running. It is clear that, less number of serv ers indicate that, ef ficient resource utilization has been achie v ed. The second metric indicates the total CPU utilization in the cloud serv er . Higher v alues of CPU utilization indicates that, proper resource utilization is being achie v ed. The simulation analysis results w .r .t. no of serv ers is illustrated in Figure 2. It is clear that, FF , FFD and Max-Min algorithms achie v e the maximum perform ance, mainly due to the theoretical performance bounds establ ished for thes e algorithms. Sim ilarly , NF and S-NF achie v e poor performance mainly due to lack of ef fecti v e theoretical performance bounds. Figure 3 illustrates the simulation analysis results w .r .t. CPU utilization. All the VMP techniques sho w similar performance, mainly due to their primary design focus of on achie ving ef ficient CPU utilization. From this simulation study , it can be inferred that, FF , FFD and Max-Min are the most f a v orable techniques for achie ving ef ficient load balancing in cloud. IJECE V ol. 9, No. 1, February 2019 : 512 524 Evaluation Warning : The document was created with Spire.PDF for Python.
IJECE ISSN: 2088-8708 521 0.5%_mean 1%_mean  300  350  400  450  500 NF FF FFD MaxMin S−NF S−FF S−FFD Number of servers  Algorithms  Figure 2. Number of serv ers vs placement algorithms Server_1 Server_2 Server_3 Server_4 Server_5 Server_6  0  20  40  60  80  100 EBFA BFA FFA NFA % of CPU Utilization  Algorithms  Figure 3. % of CPU utilization vs placement algorithms 7. DISCUSSIONS FOR FUTURE DIRECTIONS This section pro vides VMP schemes that can be considered for the Cloud en vironment in the future, with respect to the follo wing issues. (a) Consideration of appropriate schemes. (b) VM placement f actors consideration. (c) Resource consideration. (d) Po wer , cost and netw ork related f actors. 7.1. Resour ce-awar e VMP schemes Managing virtual resource is an important issues in Cloud computing. Normally , for job e x ecution, VM may require v arious resources. Resource-a w are VMP schemes are responsible for considering the resource requirements (hardw are etc.) of the VMs in the placement decisions. The optimized placement of VMs on the PMs can be achie v ed through ef ficient resource-a w are placement scheme. Resource contention amongst multiple co-hosted neighboring VMs ensures maximum resource utilization in this scheme. Bin pac king algorithms for virtual mac hine ... (K umar aswamy S) Evaluation Warning : The document was created with Spire.PDF for Python.