COMPUTE MOELLIN & NEW TECHNOLOIES 2014 1(11) 279-24 Le Zheng Vrtual achne resource allocaton algorth n cloud envronent 1, 2 Le Zheng 1 School of Inforaton Engneerng, Shandong Youth Unversty of Poltcal Scence, Jnan 250103, Shandong, Chna 2 Key Laboratory of Inforaton Securty and Intellgent Control n Unverstes of Shandong, Jnan 250103, Shandong, Chna eceved 1 July 2014, www.cnt.lv Abstract To resolve the proble that vrtual achne deployent reservaton schee waste a lot of resources and sngle-objectve deployent algorth s not coprehensve, a vrtual achne resource allocaton algorth based on vrtual achnes group ult-objectve genetc algorth s proposed. The algorth s dvded nto group codng and resources codng. esources codng ntegrated codng accordng to the hstory resource need of vrtual achnes to physcal achne and ntegrate nuber of physcal achne and resource need of physcal achne occuped by vrtual achne through proved crossover and utaton operatons. The experental results show that the algorth s effectve to reduce the nuber of physcal achne and resource utlzaton of physcal achne, savng energy as uch as possble. Keywords: Cloud coputng, esource allocaton, Vrtualzaton, Energy-savng, enetc algorths 1 Introducton Cloud coputng s a type of new coputng odel, provdng all knds of servced through Internet. Users can gan access to the cloud servce anyte, anywhere and on any devce. Vrtualzaton technology plays a crtcal role n anageent of cloud resource and dynac confguraton, for all knds of underlyng hardware resources can be encapsulated by vrtualzaton technology and provdes servces to users wth vrtual achnes as the basc resource unt [1]. However, snce cloud coputng platfor ncludes hghly dynac and heterogeneous resources, vrtual achnes has to adapt to the dynac cloud coputng envronent [2]. The purpose of vrtual achnes deployent strategy s realzng deal result by changng layout and placeent of overall vrtual achnes to optze the objectve n eetng constrant condton. The proble of vrtual achnes placeent proved to be NP proble. Thus, t s a research hotspot n current cloud coputng fled how to conduct vrtual achnes placeent effectvely. Currently, sngle-objectve resource allocaton and deployent ethod are usually adopted n the feld of vrtualzed server technology applcaton. Wth the goal of axzng utlty, the lterature [3] utlzes NUM odel n coputer network. One or ore physcal achnes resources, lke bandwdth of network lnk, dstrbutes one or ore job by vrtualzaton technology, reachng hgher level of allocaton of coputng resources of physcal achnes and optzng t by algorth. Wth the goal of energy-savng, the lterature [4] dynacally deploys vrtual achnes applcaton by energy-aware heurstc algorth. The lterature [5] put forward an proved preferental cooperaton descendng ethod to solve the proble of node bn packng. The ethod erely nvolves node ntegraton durng peak load stuaton, wthout consderaton of constrant that the goods ay be ncopatble wth the box. The lterature [6] suggests an adaptable anageent frae for vrtual achnes placeent, and studes on the soluton of genetc algorth to the overall placeent of vrtual achnes, effectvely reduce the nuber of physcal achnes and graton tes, but not consderng the ntegraton of physcal achnes resources by vrtual achnes n the solvng process. Nowadays, the optal ethod of ost vrtual achnes placeent s transforng ult-objectve optzaton proble to several sngle-objectve optzaton probles to be solved n stages. It rarely happens that ult-objectve s optzed at the sae te. In ost te, only partal optal soluton rather than global optal soluton s ganed. For the ssue of server ntegraton and resource allocaton n cloud coputng, a vrtual achnes resource allocaton algorth based on vrtual achnes ult-objectve genetc algorth s proposed. Wth the a of reduce the nuber of physcal achnes and resource allocaton, a best soluton s searched by genetc algorth to savng energy as uch as possble. 2 esource allocaton algorth n cloud coputng The defnton of bn packng proble [7] s that a set S of M n sze and a set P of N n sze gven, how all Correspondng author e-al: 12921509@qq.co 279 Inforaton and Coputer Technologes
COMPUTE MOELLIN & NEW TECHNOLOIES 2014 1(11) 279-24 eleents of S are packed n eleents of P wth least eleents of P used. BPP proble, a dffcult NP proble, cannot be done by a known optal algorth n polynoal te. The proble of vrtual achnes deployent s actually bn packng proble. In cloud coputng, how to reasonably deploy vrtual achnes to relevant nodes shall be consdered, realzng optal usage of resources whle eetng servce objectves of dfferent applcatons. The vrtual achnes placeent ay be regarded as vector bn packng proble. The goods beng packed are the vrtual achne under operaton, and the resources of vrtual achne are the changeable sze of goods. The box s the physcal node, and the capacty of the box s the usage threshold of node resource. The nuber of types of resources s the nuber of densons of vector bn packng proble. Assung that the nuber of physcal nodes s M and the nuber of vrtual achnes s N, the soluton space fro the vrtual N achnes to the physcal nodes s M. It s a NP proble slar wth bn packng proble that requres an approxate optal soluton. 2.1 ESCIPTION OF ISSUES IN MULTI- OBJECTIVE VITUAL MACHINES EPLOYMENT ALOITHM The resource asked by users to the cloud platfor s equal to s vrtual achne requrng specfc resource, and the applcatons package of each user operates on ther own vrtual achnes. It s an acadec research hotspot how to save energy and utlze cloud coputng resource as uch as possble to deploy ult-objectve vrtual achnes. eployng ult-objectve vrtual achne proble s a ultple cobnaton optzaton proble, as well as ult-objectve optzaton proble. The avalable resource of each physcal achne s ultdensonal vector, wth each denson as one of all resources of physcal achnes, and the resource needed by each physcal achne s also ult-densonal vector. The objectve s to allocate several vrtual achnes to several physcal achnes, axzng each resource utlzaton rate of physcal achnes and nzng the nuber of vrtual achne graton. The ult-objectve deployent proble s descrbed as follows: Make coputng, N as the physcal achne set n cloud PM N as vrtual achne set n cloud VM coputng, N as the total nuber of vrtual achnes, N as the avalable allocated resources set n cloud coputng and K as the total nuber of avalable allocated resources. M K M The objectve: ax U and n,k P. 1k 1 1 n N N, aong the, VM, PM,k Le Zheng U s usage rate of the K type of resource by physcal achne, P s the nuber of nodes of physcal achnes. Constrant: P 0,1. (1) U If P 1, t eans usng new physcal achne. C, (2),k,k U C. (3) n,k,k n Aong the, C eans the threshold value of the,k K type of resource of physcal achne. U s the n,k usage rate of the K type of resource by the n vrtual achne under physcal achne. The operaton of each type of resource of each physcal achne shall be less than the threshold value of each type of resource durng the vrtual achne deployent process. When there are several vrtual achnes under deployent n physcal achne, the total usage rate of resources of vrtual achne under physcal achne shall be less than the threshold value of each type of resource. 2.2 ESIN AN EALIZATION OF MULTI- OBJECTIVE VITUAL MACHINES EPLOYMENT ALOITHM For the huge cloud coputng centre, cobnatonal exploson ay occur n cobnatoral optzaton. enetc algorth s one of the ethods for solvng cobnaton proble now, snce t can concurrently handle wth all objectves and avod prorty orderng aong objectves. Therefore, genetc algorth s very sutable for solvng ult-objectve optal ssues []. A vrtual achne resource allocaton algorth based on vrtual achnes ult-objectve genetc algorth s proposed, on the bass of ult-objectve vrtual achnes deployent proble n cloud coputng centre. 2.2.1 Codng In the vrtual achnes deployent proble, there are three types of genetc codng ethods: (1) the representaton based on box; (2) the representaton based on goods; (3) the representaton based on group. Snce the objectve functon of bn packagng proble reles on the goods group, the forer two codng ethods face sngle goods, wth shortcong of unclear groupng nforaton. The shortcong of the thrd codng ethod s relyng on goods group, neglectng the dfference of utlzaton of physcal achne resources by each vrtual achne n the crossover and utaton process. In the paper, cobnng wth the codng ethod based on group 20 Inforaton and Coputer Technologes
COMPUTE MOELLIN & NEW TECHNOLOIES 2014 1(11) 279-24 and goods, dynac allocaton genetc algorth of cloud coputng resources s proposed. The codng based on goods n the paper anly adopts the codng based on the resource need of vrtual achne to the physcal achne. The resource need of vrtual achne to the physcal achne takng CPU, dsc and network as exaple, by N nuber of saplngs of vrtual achne n a whle T, calculates the nuber of operatons of CPU, dsc and network accordng to the saplng ponts. Then the nuber of operatons s coded, and the nuber of operatons of CPU, dsc and I/O are showed as forulas (4), (5), and (6). In order to understand the change of deand of vrtual achne for resources, the author of the paper adopts the energy effcency odels n the lterature [9] for data saplng. N c f u c fn t1 L T C t C t C C. (4) Aong the, f C t s the CPU frequency of vrtual achne at t te; C t s the usage rate of CPU u of vrtual achne at t te; C s the nuber of CPU c cores of vrtual achne; C s the nuber of fn calculaton of floatng pont n each perod. N d r w t1 L T t t. (5) Aong the, t s the data aount read fro r dsc of vrtual achne per second at t te; and t w s the data aount wrtten to dsc of vrtual achne per second at t te. 1 L T N t N t. (6) N n r w 2 t1 Aong the, N t s the data aount receved by r network card of vrtual achne per second at t te; and N t s the data aount sent by network card of w vrtual achne per second at t te. Then calculate the probablty of the calculaton aount of three types of resources needed by each vrtual achne n the calculaton aount of physcal achne recourses and conduct noralzaton processng. The forula of probablty of the calculaton aount s as the followng forula (7). P μ Z μ 1. (7) Aong the, μ s the calculaton aount of CPU, dsc and I/O of vrtual achne; and Z s the nuber of vrtual achnes n physcal achnes of vrtual Le Zheng achne. P ay be the CPU calculaton probablty of CPU, dsc and I/O of vrtual achne. Then noralzaton processng s conducted accordng to the rato of probablty H of n the entre set of all types of resources n current physcal achne. The value s n the range of [0, 10]. and the probablty dstrbuton s shown n fgure 1. P1 P2 P 1 2 2 Fgure 1. Probablty dstrbuton graph FIUE 1 Probablty dstrbuton graph Make c,, n represent the codng of the resource need by CPU, dsc and network, and ake the proportonalty dstrbuton of all types of resources of vrtual achne after noralzaton processng as the resource codng of vrtual achne. Constrant: c 0 9,c N, 0 9, N, n 0 9,n N. Make T as the threshold value of CPU resource of c physcal achne, T as the threshold value of dsc resource of physcal achne and T as threshold value I / O of CPU resource of physcal achne. In order to better llustrate the algorth rose n the paper, the values of T, c T and I / O T n the genetc operaton process s. The threshold less than 10 s for reservng part of resource space for graton of vrtual achnes. The codng ethod s shown as fgure 2. It anly takes group as chroosoe that has uneven length for the nconsstent nuber of genes n chroosoe. There are three types of deployent of 9 vrtual achnes: the length of EBA and FCQ are the sae but the types of deployent are dfferent; both the length of chroosoe of EBA and FCQ as well as the types of deployent are dfferent. 21 Inforaton and Coputer Technologes
COMPUTE MOELLIN & NEW TECHNOLOIES 2014 1(11) 279-24 Le Zheng Crossover process based on group codng s shown n fgure 3 wth steps as follows: 1. andoly select two father nodes, cross part and cross dot. 2. Insert a selected vrtual achne to father node to for new deployed physcal achne setoff vrtual achnes. 3. elete the repettve vrtual achnes n the new physcal achne set. Cross part 2.2.2 Evaluaton of ftness FIUE 2 Vrtual achne codng enetc algorth evaluates the pros and cons of ndvduals accordng to ftness. Ftness functon eans the correspondng relevance between the whole subjects and ther ftness. Evaluaton of ftness conducts evaluaton of each ndvdual and prepare for the next genetc operaton. Snce the objectve functon adopts physcal achnes as less as possble to place vrtual achnes as ore as possble, the ftness functon s shown as the followng forula (), accordng to the objectve functon and constrant condton: P K j j,k j 1 k 1 Ftness U / P. () Aong the, j eans the nuber of father node, P j as the total nuber of physcal achnes used by father node, and U as the utlzaton rate of k type of resource,k of vrtual achne. 2.2.3 Crossover The an functon of the crossover process n genetc algorth s lettng the next generaton nhert the excellent genes fro the parents and have chance to produce ore excellent generatons. There are two parts of the crossover process: one s crossover process based on group codng ang to nze the nuber of physcal achnes, and the other one s crossover process based on resource codng ang to axzng the resource of physcal achnes. 315 132 112 E 3 2 1 434 312 132 B Cross dott 4 132 6 542 (a)selecton of cross part and cross dot based on group 3 2 1 434 312 132 3 1 434 132 312 132 112 E (b) Insert the cross part 312 132 112 E 4 132 (c) elete repettve physcal achne Fgure FIUE 3. Codng 3 Codng cross cross process process based based on on group group Based on the resource codng crossover process and group codng crossover process, the genetc operaton of resource codng can fasten the convergence speed of the group codng genetc process, and ntegrate the resource of physcal achne occuped by the vrtual achnes. In order to reserve the group crossover result, the resource codng crossover process wll reserve the nserted vrtual achne group of group codng, not conductng resource codng crossover to t. Crossover process based on resource codng s shown n fgure 4 wth steps as follows: 1. Select the reanng vrtual achne of the frst father node and the second father node (excludng the cross part) of the group codng crossover result as the father nodes. Select the cross part and cross dots. Now the cross part s the vrtual achne but not the vrtual achne group. 2. Insert the selected vrtual achne to vrtual achne group. 3. Cobne ndependent vrtual achnes and delete repettve physcal achne as well as physcal achne wthout any vrtual achne. 4. Cobne the results of group codng crossover process and resource codng crossover process to get the crossover process result. A 6 7 5 542 221 112 6 7 5 542 221 112 6 7 542 221 22 Inforaton and Coputer Technologes
COMPUTE MOELLIN & NEW TECHNOLOIES 2014 1(11) 279-24 B Cross dott 3 1 (a)selecton of cross part and cross dot Cross part 434 132 542 6 A 6 542 6 7 542 221 542 (c) Integrate ndvdual vrtual achne and delete repettve and redundant physcal achne. 6 (b) Insert the cross part 312 132 112 E 542 (d)cobne the cross part of cross results between group codng and resource codng FIUE Fgure 4. 4 Cross Cross process process based based on on resource resource codng codng 3 Experental analyss 6 Le Zheng proportonalty and utaton proportonalty of ultobjectve vrtual achne deployent algorth s 0.7 and 0.5 respectvely, and the genetc algebra s set as 10. There are load paraeter and change of vrtual achne resource need durng the experent, takng 10 nute as a te unt to record the change of nuber of physcal achnes n 10 te unts by three types of algorth. The experent result s shown as fgure 5: FIUE 5 Coparsons of nuber of physcal achnes The experent shows that as te goes on, wth the dynac change of vrtual achne s need of resource, the nuber of physcal achnes n MOA algorth s less than that of BFA and FFA. It s because that n a dynac process, MOA algorth searches the least nuber of physcal achnes n generc operaton that eets the constrant condton. It shows that MOA algorth can effectvely reduce the nuber of physcal achnes. In order to verfy the proposed algorth, the author conducts sulaton experent n CloudS [10]. Wth the purpose of verfyng the effectveness and deployent schee, we select the followng two classcal vrtual achne deployent algorth ( Mult-object vrtual achnes resource allocaton Algorth, MOA ) to copare wth the ult-objectve vrtual achne resources dstrbuton algorth. Best Ft Algorth (BFA) eans to select the physcal achne that eets the resource need of vrtual achne wth least reanng resource durng the vrtual achne deployent process, akng the physcal achne least reanng resource. Frst Ft Algorth (FFA) eans to search physcal achnes n order durng the vrtual achne deployent process, lettng vrtual achne drectly deployed n the physcal achne that eets the resource need of vrtual achne. Experent 1 Calculaton of nuber of physcal achnes eploy 100 vrtual achnes n 50 physcal achnes usng three types of algorth ndependently, wth the sae nature of physcal achnes and vrtual achne tasks exceptng the deployent ethod and resource threshold value. Aong the, the crossover Experent 2 Calculaton of resource utlzaton rate Calculate the average resource utlzaton rate of physcal achnes by three types of algorth, takng 10 nute as a te unt to record the change of usage rate of CPU and nner storage of physcal achnes n 10 te unts, and calculate the average resource utlzaton rate. The experent results of average usage rate of CPU and nner storage by three types of algorth are shown n fgure 6 and 7. FIUE 6 Coparsons of the average usage rate of CPU 23 Inforaton and Coputer Technologes
COMPUTE MOELLIN & NEW TECHNOLOIES 2014 1(11) 279-24 Le Zheng whle BFA algorth try to deploy physcal achnes as less as possble but not consderng provng the resource usage rate, and there are randoness n FFA algorth that does not consder the resource usage rate. It shows that MOA algorth can prove the resource usage rate and save energy to soe extent. 4 Concluson FIUE 7 Coparsons of the average usage rate of nternal storage It shows fro the copared experental results that the average usage rate of CPU and nternal storage of physcal achnes deployed by MOA algorth s evdently hgher than that of BFA and FFA algorth. It s because MOA try to prove the resource usage rate as uch as possble by usng genetc algorth to adjust vrtual achne group durng the deployent process, In the paper, after analysng the research stuaton of vrtual achne deployent schee n cloud coputng, the author put forward proved genetc algorth, conduct group codng and resource need codng of vrtual achne, and prove the crossover and utaton operaton to resolve the proble of energy waste n cloud coputng. Under experental condton, t shows that the algorth can not only reduce the nuber of physcal achnes but also prove the resource utlzaton rate. In further research, the ssue whether the perforances aong vrtual achnes are related wll be ntroduced. eferences [1] Jnhua Hu, Janhua u, uofe Sun, et al. 2010 A schedulng strategy on load balancng of vrtual achne resources n cloud coputng envronent Thrd nternatonal syposu on parallel archtectures, algorths and prograng 9(96) 1-20 [2] Cherkasova L, upta, Vahdat A 2007 When vrtual s harder than real: esource allocaton challenges n vrtual achne based t envronents Techncal eport HPL-2007-25 [3] Truong H L, ustdar S 2010 Coposable cost estaton and ontorng for coputatonal applcatons n cloud coputng envronents Proceda coputer scence 1(1) 2175-4 [4] Bo L, Janxn L, Jnpeng Hua, et al. 2009 EnaCloud: an energysavng applcaton lve placeent approach for cloud coputng envronents IEEE nternatonal conference on cloud coputng 17(24) 21-5 [5] Ajro Y, Tanaka A 2007 Iprovng packng algorths for server consoldaton Proceedngs of the 33rd Internatonal Coputer Measureent roup Conference San ego 399-406 [6] Q, J Q, Pan J Z, u J 2011 Extendng descrpton logcs wth uncertanty reasonng n possblstc logc Internatonal journal of ntellgent systes 26 353-1 [7] Aktas H, Cagan N 2007 Soft sets and soft groups Inforaton scences 177 2726-35 [] Sun Y L, Perrott, Harer T, Cunnngha C, Wrght P 2010 An SLA focused fnancal servces nfrastructure Proceedngs of the 1st Internatonal Conference on Cloud Coputng Vrtualzaton (CCV 2010), Sngapore, 2010 [9] udolph S 2011 Foundatons of descrpton logcs In: Polleres, A., Aato, C., Arenas, M., Handschuh, S., Kroner,P., Ossowsk, S. and PatelSchneder, P.F., Eds. easonng Web. Seantc Technologes for the Web of ata, Lecture Notes n Coputer Scence, Sprnger, Berln, Hedelberg 76-136 [10] Calheros N, anjan, e ose C A F, et al. 2009 Cloud-S: A novel fraework for odelng and sulaton of cloud coputng nfrastructures and servces Parkvlle VIC: The Unversty of Melbourne Australa, rd Coputng and strbuted Systes Laboratory. Authors Le Zheng, born on August 3, 190, Chna Current poston, grades: researcher at Shandong Youth Unversty of Poltcal Scence, Chna. Unversty studes: aster s degree n Coputer Software and Theory fro Shandong Noral Unversty, Chna n 2006. Scentfc nterests: cloud coputng and dstrbuted coputng. 24 Inforaton and Coputer Technologes