DREAMS: Dynamic Resource Allocation for MapReduce with Data Skew

Size: px
Start display at page:

Download "DREAMS: Dynamic Resource Allocation for MapReduce with Data Skew"

Transcription

1 DREAMS: Dynamc Resource Allocaton for MapReduce wth Data Skew Zhhong Lu, Q Zhang, Mohamed Faten Zhan, Raouf Boutaba Yapng Lu and Zhenghu Gong College of Computer, Natonal Unversty of Defense Technology, Changsha, Hunan, Chna Emal:{zhlu,gzh}@nudt.edu.cn Scence and Technology on Parallel and Dstrbuted Processng Laboratory, Natonal Unversty of Defense Technology, Changsha, Hunan, Chna Emal:yplu@nudt.edu.cn Davd R. Cherton School of Computer Scence, Unversty of Waterloo, Waterloo, ON Canada Emal: {q8zhang,mfzhan,rboutaba}@uwaterloo.ca Abstract MapReduce has become a popular model for largescale data processng n recent years. However, exstng MapReduce schedulers stll suffer from an ssue known as parttonng skew, where the output of map tasks s unevenly dstrbuted among reduce tasks. In ths paper, we present DREAMS, a framework that provdes run-tme parttonng skew mtgaton. Unlke prevous approaches that try to balance the workload of reducers by reparttonng the ntermedate data assgned to each reduce task, n DREAMS we cope wth parttonng skew by adjustng task run-tme resource allocaton. We show that our approach allows DREAMS to elmnate the overhead of data reparttonng. Through experments usng both real and synthetc workloads runnng on a 11-node vrtual vrtualsed Hadoop cluster, we show that DREAMS can effectvely mtgate negatve mpact of parttonng skew, thereby mprovng job performance by up to 2.3%. I. INTRODUCTION In recent years, the exponental growth of data n many applcaton domans such as e-commerce, socal networkng and scentfc computng, has generated tremendous needs for large-scale data processng. In ths context, MapReduce [1] as a parallel computng framework has recently ganed sgnfcant popularty. In MapReduce, a job conssts of two types of tasks, namely Map and Reduce. Each map task takes a block of nput data and runs a user-specfed map functon to generate ntermedate key-value pars. Subsequently, each reduce task collects ntermedate key-value pars and apples a user-specfed reduce functon to produce the fnal output. Due to ts remarkable advantages n smplcty, robustness, and scalablty, MapReduce has been wdely used by companes such as Amazon, Facebook, and Yahoo! to process large volumes of data on a daly bass. Consequently, t has attracted consderable attenton from both ndustry and academa. Despte ts success, the current mplementatons of MapReduce stll suffer from several mportant lmtatons. In partcular, the most popular mplementaton of MapReduce, Apache Hadoop MapReduce [2], uses a hash functon Hash(HashCode(ntermedate key) mod ReduceNumber) to partton the ntermedate data among the reduce tasks. Whle the goal of usng the hash functon s to evenly dstrbute workload to each reduce tasks, n realty ths goal s rarely acheved [3] [5]. For example, Zachelas et. al. [3] have demonstrated the exstence of skewness n Youtube socal graph based on real workloads. The experments showed that, the bggest sze of parttons s larger than the smallest by more than a factor of fve. The skewed dstrbuton of reduce workload can have severe consequences. Frst, data skewness may lead to a large dfference n the runtme between the fastest and slowest tasks. As the completon tme of a MapReduce job s determned by the fnshng tme of the slowest reduce task, data skewness can cause certan tasks to run much slower than others, thereby severely delayng job completon. Second, Hadoop MapReduce allocates fxed-sze contaners to reduce tasks. However, due to data skewness, dfferent reduce tasks may have dfferent run-tme resource requrements. As a result, machnes that are runnng tasks wth heavy workload may experence resource contenton, whle machnes wth less data to process may experence resource dleness. There are several approaches recently proposed to handle parttonng skew n MapReduce [4], [6] [9]. Ibrahm et. al. proposed LEEN [6], a framework that balances reduce workload by assgnng ntermedate keys to reducers based on ther record szes. Whle ths approach can mtgate the negatve mpact of data skew, ts beneft s lmted snce the szes of records correspondng to each key can stll be unevenly dstrbuted. Furthermore, t does not perform well when dstrbuton of the records szes s severely skewed. Subsequently, Gufler et. al. [7] and Ramakrshnan et. al. [8] proposed technques to splt each key wth large record sze nto sub-keys to allow for more even dstrbuton of workload to reducers. However, most of these solutons have to wat untl all the map tasks completed to gather the partton sze nformaton before reduce tasks can be started. The authors of [5], [9] demonstrate that by startng the shuffle phase after all map tasks are completed, the overall job completon tme wll be prolonged. Whle the progressve samplng [8] and adaptve parttonng [4] can elmnate ths watng tme, the former approach requres an addtonal samplng phase to generate a parttonng plan before the job can be executed, whereas the latter approach ncurs an addtonal run-tme overhead (e.g. 3 seconds for certan jobs). In ether case, the overhead due to reparttonng can be qute large for small jobs that takes from 1 to 1 seconds to complete. These small jobs are qute common n today s producton clusters [1].

2 Fg. 1: MapReduce Programmng Model pars wth the same hash results are assgned to the same reduce task. In the reduce stage, each reduce task takes one partton (.e. the ntermedary key-value pars receved from all map tasks) as nput and performs the reduce functon on the partton to generate the fnal output. Ths s llustrated n Fgure 1. Typcally, the default hash functon can provde load balancng f the key frequences and the sze of key-value pars are unformly dstrbuted. Ths may fal wth skewed data. For example n the InvertedIndex applcaton, hash functon parttons the ntermedate data based on the words appeared n the fle. Therefore, reduce tasks processng more popular words wll be assgned a larger amount of data. As shown n Fgure 1, parttons are unevenly dstrbuted by the hash functon. P 1 s larger than P 2, whch causes workload mbalance between R1 and R2. [6] presents the causes of parttonng skew: Motvated by the lmtaton of the exstng solutons, n ths paper, we take a completely dfferent approach to address data skewness. Instead of subdvdng keys nto smaller subkeys to balance the reduce workload, our approach adjusts run-tme resource allocaton of each reducer to match ther correspondng data sze. Snce no reparttonng s nvolved, our approach completely elmnates the overhead due to reparttonng. To ths end, we present DREAMS, a Dynamc REsource Allocaton technque for MapReduce wth data Skew. DREAMS leverages hstorcal records to construct profles for each job type. Ths s reasonable because many producton jobs are executed repeatedly n today s producton clusters [11]. At run-tme, DREAMS can dynamcally detect data skewness and assgn more resources to reducers wth large parttons to make them fnsh faster. In DREAMS, we frst develop an onlne predcton model whch can estmate the partton szes of reduce tasks at runtme. We then establsh a performance model that correlates run-tme resource allocaton wth task completon tme. Usng ths performance model, the scheduler can make schedulng decsons that allocate the rght amount of resources to reduce tasks so as to equalze ther runnng tme. Through experments usng both real and synthetc workloads runnng on a 11-node vrtualzed Hadoop cluster, we show that DREAMS can effectvely mtgate negatve mpact of partton skew, thereby mprovng job performance by up to 2.3%. The rest of ths paper s organzed as follows. Secton II provdes the motvatons of our work. We descrbe the system archtecture of DREAMS n Secton III. Secton IV llustrates the desgn of DREAMS n detal. Secton V provdes the results of expermental evaluaton. Fnally, we summarze exstng work related to DREAMS n Secton VI, and draw our concluson n Secton VII. II. MOTIVATION In ths secton we provde an overvew of the parttonng skew problem and dscuss the resource allocaton ssues n current MapReduce mplementaton theren motvatng our study. In state of the art MapReduce systems, each map task processes one splt of nput data, and generates a sequence of key-value pars whch are called ntermedate data, on whch hash parttonng functon s performed. Snce all map tasks use the same hash parttonng functon, the key-value skewed key frequences: Some keys occur more frequently n the ntermedate data, causng those reduce tasks that process these popular keys become overloaded. skewed tuple szes: In applcatons where the szes of values n the key-value par vary sgnfcantly, uneven workload dstrbuton may arse. skewed executon tmes: Typcal n scenaros where processng a sngle, large key-value par may requre more tme than processng multple small pars. Even f the overall number of tuples per reduce task s the same, the executon tmes of reduce tasks may be dfferent. Due to many weaknesses and nadequaces experenced n the frst verson of Hadoop MapReduce (MRv1), the next generaton of Hadoop compute platform, YARN [2], has been proposed. Nevertheless, n both Hadoop MRv1 and MRv2 (a. k. a. YARN), the schedulers assume each reduce task has unform workload and resource consumpton, and therefore allocate dentcal resources to each reduce task. Specfcally, MRv1 adopts a slot-based allocaton scheme, where each machne s dvded nto dentcal slots that can be used to execute tasks. However, MRv1 does not provde resource solaton among co-located tasks, whch may cause performance degradaton at run-tme. On the other hand, YARN uses a contaner-based allocaton scheme, where each task s scheduled n an solated contaner wth guaranteed CPU ad memory resources that can be specfed n the request. But YARN stll allocates contaners of dentcal sze to all reduce tasks that belong to the same job. In the presence of parttonng skew, ths schedulng scheme can cause both varaton n task runnng tme and degradaton n resource utlzaton. For nstance, Kwon et. al. [4] demonstrated that n CloudBurst Applcaton, there s a factor of fve dfference n runtme between the fastest and the slowest reduce tasks. Snce the job completon tme depends on the slowest task, the runtme varaton of reduce tasks wll prolong the job executon. At the same tme, the reducers wth large parttons run slowly because the resources allocated to them are lmted by the contaner sze, whereas reducers wth lght workload tend to under-utlze the resources allocated to the contaner. In both cases, the resultng resource allocaton s neffcent. Most of the exstng approaches [4], [6] [9] tackle the parttonng skew problem by makng the workload assgn-

3 NodeManager Map1 Map2 Job Profle Job Profle Job Profle Partton Sze Montor Partton Stats Report Resource Request Resource Response Contaner Launch Applcaton Master Task Duraton Estmator Partton Sze Predctor NodeManager Partton Sze Montor Map3 Map4 Resource Manager Fne-graned Contaner Scheduler Resource Allocator NodeManager Partton Sze Montor Map5 Map6 Fg. 2: Archtecture of DREAMS ment unform among reduce tasks, thereby mtgatng the neffcences n both performance and utlzaton. However, achevng ths goal requres (sometmes heavy) modfcaton to the current Hadoop mplementaton, and often requres addtonal overhead n terms of samplng and adaptve parttonng. Therefore, n ths work we seek an alternatve soluton, consstng n adjustng contaner sze based on parttonng skew. Ths approach not only requres mnmal modfcaton to the exstng Hadoop mplementaton, but at the same tme can effectvely mtgate the negatve mpact of data skew. III. SYSTEM ARCHITECTURE Ths secton descrbes the desgn of our proposed resource allocaton framework called DREAMS. The archtecture of DREAMS s shown n Fgure 2. Specfcally, each Partton Sze Montor records the statstcs of ntermedate data that each map task generates at run-tme and sends them to the ApplcatonMaster though heartbeat messages. Partton Sze Predctor collects the partton sze reports from NodeManagers and predcts the partton szes for ths job at runtme. The Task Duraton Estmator constructs statstcal estmaton model of reduce task performance as a functon of ts partton sze and resource allocaton. The Resource Allocator determnes the amount of resources to be allocated to each reduce task based on the performance estmaton. Lastly, the Fnegraned Contaner Scheduler s responsble for schedulng task requests from ApplcatonMasters accordng to schedulng polces such as Far schedulng [12] and Domnant Resource Farness (DRF) [13]. The workflow of resource allocaton mechansm used by DREAMS conssts of 5 steps: (1) After the ApplcatonMaster s launched, t schedules all the map tasks frst and then ramps up the reduce task requests slowly accordng to the slowstart settng. Durng ther executon, each Partton Sze Montor records the sze of ntermedate data produced by each reduce task. It then sends the statstcs to the ApplcatonMaster through the RPC protocol used to montor the status of task n Hadoop. (2) Upon recevng the partton sze reports from the Partton Sze Montors, the Partton Sze Predctor performs sze predcton usng our proposed predcton model (see Secton IV-A). The task Duraton Estmator, whch uses the job profles (Secton IV-B), predcts the task duraton of each reduce task wth specfed amount of resources. Based on that, Resource Allocator determnes the amount of resources for each reduce task accordng to our proposed resource allocaton algorthm (Secton IV-C) to equalze the executon tme of all reduce tasks. (3) After that, the ResourceManager receves ApplcatonMaster s resource requests through the heartbeat messages, and schedule free contaners n the cluster to ApplcatonMaster. (4) Once the ApplcatonMaster obtans new contaners from ResourceManager, t assgns the correspondng contaner to ts pendng task, and fnally launches the task. IV. DREAMS DESIGN There are two man challenges that need to be addressed n DREAMS. Frst, to dentfy partton skew, t s necessary to develop a run-tme forecastng algorthm that predcts the partton sze of each reducer. Second, n order to determne the rght contaner sze for each reduce task, t s necessary to develop a task performance model that correlates task runnng tme wth resource allocaton. In the followng sectons, we shall descrbe our techncal solutons for each of the challenges. A. Predctng Partton Sze As mentoned prevously, the scheduler needs to know the partton sze of each reduce task n order to compute the correct contaner sze for that reduce task. Snce current Hadoop schedulers allow reduce tasks to be launched soon after a fracton (e.g. 5%) of map tasks are fnshed 1, t s necessary to predct the partton sze before the completon of all map tasks. To predct the partton sze of each reduce task to be scheduled, ( at ) run-tme the ApplcatonMaster collects two metrcs F j, S j, where F j s the percentage of map tasks that have been processed, (j ( [1, ) m] and m refers to the number of collected tuples F j, S j ) and S j s the sze of the partton generated by the completed map tasks for reduce task. In our mplementaton, we have modfed the reportng mechansm so that each map task reports ths nformaton to the ApplcatonMaster upon completon. Wth these metrcs, we use lnear regresson to determne the followng equaton for each reduce task [1, N]: a 1 + b 1 F j = S j j = 1, 2, m (1) We ntroduce an outer factor, δ, whch s the threshold to control our predcton model to stop the process of learnng, and fnalze the predcton. In practce, δ can be the map completon percentage at whch reduce tasks may be started to schedule (e.g. 5%). Every tme a new map task has fnshed, a 1 To mprove job runnng tme, exstng Hadoop schedulers overlap the executon of map tasks and reduce tasks by allowng reduce tasks to be launched before the completon of all map tasks

4 Duraton (Seconds) measured value regresson value Sze of Partton (MB) (a) InvertedIndex 1G Duraton (Seconds) measured value(1g) regresson value(1g) measured value(2g) regresson value(2g) Sze of Partton (MB) (b) InvertedIndex 1 and 2G Fg. 3: Relatonshp between task duraton and partton sze Duraton (Seconds) Memory Allocaton (a) a reduce task n Sort Duraton (Seconds) Memory Allocaton (b) a reduce task n InvertedIndex Fg. 5: Relatonshp between task duraton and mem. allocaton Duraton (Seconds) measured value regresson value regresson value(pece wse) CPU allocaton (vcores) (a) a reduce task n Sort Duraton (Seconds) measured value regresson value regresson value(pece wse) CPU allocaton (vcores) (b) a reduce task n InvertedIndex Fg. 4: Relatonshp between task duraton and CPU allocaton new tranng data s created. When the fracton of map tasks reaches δ, we calculate the scalng factors (a 1, b 1 ) and predct the sze of partton for each reduce task of the whole data set, even though not all of the map tasks are completed. We notced that predcton schemes such as progressve samplng [8] can also be used by DREAMS for partton sze predcton. However, the reparttonng mechansm used n [8] s based on a parttonng plan, and as result, t requres progressve samplng to be executed each tme before the job starts. In our case, snce we do not need to modfy the mplementaton of parttonng, our partton sze predcton can be done entrely onlne. Thus, we found our current predcton scheme s smple yet suffcent to produce hgh qualty predcton results. B. Reduce Phase Performance Model In ths secton, we desgn a task performance model that correlates the completon tme of ndvdual reduce tasks wth ther partton sze and resource allocaton. As Hadoop YARN only allows the CPU and memory sze of contaner to be specfed, n our mplementaton we focus on capturng the mpact of CPU and memory allocaton on task performance. In order to dentfy the relatonshp between task runnng tme, partton sze and resource allocaton, we run a set of benchmarks n our Testbed cluster by varyng resource allocaton. More specfcally, each benchmark s generated by varyng CPU allocaton = {1 vcore, 2 vcores,, 8 vcores}, memory allocaton Alloc mem = {1 GB, 2 GB,, 8 GB}, and nput dataset D set = {1 GB, 2 GB, 3 GB, 5 GB} for dfferent jobs. We run each benchmark 1 tmes, and collect the average result over the runs for each benchmark. In the frst set of experments, we fx the CPU and memory allocaton of each reduce task and focus on dentfyng the relatonshp between partton sze and task runnng tme. To llustrate, Fgure 3a shows the result of runnng the InvertedIndex job usng 1GB nput. It s evdent that there s a lnear relatonshp between partton sze and runnng tme. Furthermore, Fgure 3b shows the result when the nput sze of the job s changed from 1GB to 2GB. Agan, the runnng tme s lnearly correlated wth partton sze. However, at the same tme, we also found that the sze of total ntermedate data denoted as D (the sum of all parttons) has an mpact on task duraton whle varyng the nput dataset. Smlar observaton s also made n [14], where Zhang et. al. show the duraton of the shuffle phase can be approxmated wth a pece-wse lnear functon when the ntermedate data per reduce task s larger that 3.2 GB n ther Hadoop Cluster. Ths s consstent wth the phenomenon we observed. In the next set of experments, we fx the nput sze and vary ether the CPU or memory allocaton of each reduce task. Fgure 4 shows the typcal results for Sort and InvertedIndex job by varyng the CPU allocaton (memory allocaton s fxed to 1 GB). We found that task runnng tme s nversely proportonal to CPU allocaton. In partcular, the task runnng tme s approxmately halved when the CPU allocaton s ncreased from 1 vcore to 2 vcores. Whle ths relatonshp s accurate when the number of vcores s small, we also found ths model s no longer accurate when a large amount of CPU resource s allocated to a task. In these cases, the resource bottleneck may swtch from CPU to other resource dmensons lke dsk I/O, n whch case the benefts of ncreasng CPU allocaton would decrease. Thus, we can expect that the duraton of reduce tasks mght be approxmated wth a dfferent nversely proportonal functon when CPU allocaton exceeds a threshold ϕ. Ths threshold could be related to Job characterstcs and cluster confguraton. However, for a dfferent Job and Hadoop cluster, ϕ can be easly determned by comparng the change n task duraton whle ncreasng CPU allocaton. 2 We then repeat the same experment for memory;we vary the memory allocaton from 1 to 7 GB whle the CPU s fxed to 1 vcore. We found the same relatonshp does not apply to memory. Fgure 5 shows the task runnng tme as 2 We use the followng polcy n ths paper: we ncrease the CPU allocaton from 1 vcore to 8 vcores, and caluate the speedup of task runnng tme between current and prevous CPU allcoatons denoted as Speedup j (j [1, 7]). The frst CPU allocaton where Speedup j <.5 Speedup j 1 s consdered as the threshold ϕ.

5 a functon of memory allocaton. We found that even though memory allocaton s ncreased, no mprovement can be found. We beleve the reason s that memory s not the bottleneck resource for ths task. In ths case, the memory allocaton wll not affect task duraton as long as t s suffcent for ths task. Based on the above observatons, we now derve our task performance model. For each reduce task among N reduce tasks, let T denote the executon tme of reduce task, P denote the sze of partton for reduce task, D denote the sze of the ntermedate data for the job, and denote as the CPU allocaton for reduce task, the performance model can be stated as: W hen <=ϕ, T =α + βp +γd + W hen >ϕ, T =α +β P +γ D+ ζ ζ + ηp + η P + ξd + ξ D where α, β, γ, ζ, η, α, β, γ, ζ and η are the coeffcent factors to be solved usng nonlnear regresson [15]. In practce, we may leverage hstorcal records of job executon to provde nput to the regresson algorthm. Ths s reasonable n producton envronments as many jobs are executed routnely n today s producton data centers. Specfcally, we capture a trple (T, P, ) for each reduce task of the job. Usng the trples for all reduce tasks as tranng data, we can easly learn the coeffcent factors n the performance model for each job. In the end, we produce one performance model M j for each job j that can be used as nput for schedulng. Fnally, we would lke to menton that whle our performance model focuses on CPU allocaton, we beleve our model can be extended to handle the case where other resources becomes the performance bottleneck by havng addtonal terms (e.g. smlar to the second and thrd term n equaton 2) n our performance model. C. Schedulng Algorthm Once the performance model has been traned and the partton sze has been predcted, the scheduler can now decde how much resource to be allocated to each task. In order to mtgate the mpact of data skew, we adopt a smple strategy whch s to make all reduce tasks have smlar runnng tme. Algorthm 1 descrbes our resource allocaton polcy. After reachng the threshold δ, the partton sze of each reduce task can be predcted wth the predcton model. As to memory allocaton, t does not affect task duraton as long as t s suffcent for ths task, whch s dscussed n Secton IV-B. We adjust the memory allocaton to P Unt Unt mem, where Unt mem s the mnmum allocaton mem of memory. Wth respect to CPU allocaton, we obtan the amount of resources accordng to performance model M j, as descrbed from lne 5 to lne 12. Frst, we calculate the executon tme T md, whch represents the tme t takes to complete the task wth the medan partton sze P md, by performance model M j. After that, we set T md as target for each reduce task, and calculate the amount of resources that each reduce (2) Algorthm 1 Resource allocaton algorthm Input: δ - Threshold of stoppng tranng the Partton Sze Predcton Model; M j - Reduce Phase Performance Model of Job j; ϕ - Maxmum allocaton of CPU; Output: C - Set of resource allocatons for each reduce task, Alloc mem 1: Collect S and F, when a success completon event of map tasks s receved by ApplcatonMaster 2: When threshold δ s reached: 3: Stop tranng and fnalze Partton Sze Predcton Model 4: Predct Set < P > whle F = 1% 5: Calculate the medan value P md n Set < P > 6: Calculate T md, when P = P md, = 1vcore usng M j 7: for each reduce task [1, N] do 8: Alloc mem = P Unt Unt mem ; mem 9: Solve the Equaton 2 for 1: f ϕ then 11: = ϕ 12: end f 13: C = C {, Alloc mem } 14: end for 15: return C task needs. Because nodes have fnte resource capactes, should be less than the capactes. Besdes, from our experence, after CPU allocaton to a task reaches a threshold, ncreasng allocaton wll not mprove the executon tme, but nstead results n wastng CPU resource as shown n Secton IV-B. We consder should be less than threshold ϕ, whch s also an nput to our algorthm. V. EVALUATION We perform our experments on 11 vrtual machnes (VMs) n the SAVI Testbed [16], whch contans a large cluster wth many server machnes. Each VM has four 2 GHz cores, 8 GB RAM and 8 GB hard dsk. We deploy Hadoop YARN 2.4. wth one VM as Resource Manager and Name Node, and remanng 1 VMs as workers. Each worker s confgured wth 8 vrtual cores and 7GB RAM (leavng 1GB for other processes). The mnmum CPU and memory allocatons to a contaner are 1 vcore and 1 GB respectvely. The HDFS block sze s set to 128MB, and the replcaton level s set to 3. We chose two jobs to evaluate DREAMS: (1) Sort, whch s ncluded n a MapReduce benchmark n Hadoop dstrbuton. It takes sequence fles whch are generated by RandomWrter as nput, and outputs the sorted data, and (2) InvertedIndex, whch comes from PUMA benchmarks [17]. It takes a lst of documents as nput and generates an nverted ndex for these documents. We use Wkpeda data [17] for ths applcaton. A. Accuracy of predcton of partton sze In ths set of experments, we wanted to valdate the accuracy of the partton sze predcton model. To ths end, we execute MapReduce jobs on dfferent datasets wth dfferent thresholds δ, and compute the average relatve error () of all parttons n each scenaro. The s defned as follows. P pred P measrd = 1 N N =1 P measrd (3)

6 Applcaton TABLE I: Average relatve error of partton sze predcton Type Sze(GB) δ =.5 δ =.6 δ =.7 δ =.8 δ =.9 δ =.1 Sort Synthetc % 2.9% 1.94% 1.81% 1.71% 1.71% Sort Synthetc 2 1.6% 1.43% 1.32% 1.26% 1.17% 1.13% Sort Synthetc 5 1.1% 1.1%.94%.9%.84%.78% InvertedIndex Wkpeda % 7.63% 7.5% 7.5% 6.43% 5.87% InvertedIndex Wkpeda % 5.25% 5.8% 4.79% 4.53% 4.38% InvertedIndex Wkpeda % 4.43% 4.21% 4.7% 3.9% 3.7% where N s the number of reduce tasks n ths job, P pred and P measrd are the predcted value and measured value of partton sze of reduce task respectvely. Table I summarzes the average relatve errors n each scenaro. We run 1 experments for each scenaro and adopt the average. It can be seen that the s less than 8.2% n all cases. Furthermore, wth threshold δ ncreases, the predcton accuracy s mproved. B. Accuracy of reduce phase performance model In order to formally evaluate the accuracy and workload ndependency of the generated performance model, we compute the predcton error for Sort and InvertedIndex wth dfferent nput workloads. We perform two valdatons as follow: Test-on-tranng - evaluate the accuracy of preformance model based on the tranng dataset. That s, we compute the predcted reduce task duraton for each tuple (P, ) 3 by usng the performance model whch s learned from ths tranng dataset, then compute a predcton error; Test-on-unknown - evaluate the accuracy of performance model usng unknown dataset. That s, we compute the predcted reduce task duraton for each tuple (P, ) by usng the performance model whch s learned from 1 G workload (Ths derved model s consdered as a profle), then compute a predcton error. For both valdatons, we leverage the to evaluate the accuracy usng followng equaton: = 1 k T pred l T measrd l l=1 (4) k T measrd l where k s the number of tuples (P, ) for a nput dataset. Table II summarzes the average relatve error of reduce task performance model for Sort and InvertedIndex. More specfcally, wth regard to Test-on-tranng valdaton, the predcton error for Sort and InvertedIndex wth all of the workloads s less than 15%. For the Test-on-unknown group, the predcton error s slghtly hgher than the correspondng value n the Test-on-tranng, stll less than 2%. These results confrm the accuracy of our performance model. C. Performance Evaluaton We have mplemented DREAMS on Hadoop YARN 2.4. as an addtonal feature. Implementng ths approach requres 3 For example, there are N reduce tasks of a job, for each reduce task, there are one value of P and 8 values of {1, 2,, 8}. Therefore, there are 8N tuples for ths workload. TABLE II: Average relatve errors of reduce task performance model Applcaton Type Sze(GB) Test-on-tranng Test-on-unknown Sort Synthetc % 9.36% Sort Synthetc % 1.62% Sort Synthetc % 16.38% Sort Synthetc % 19.57% InvertedIndex Wkpeda % 13.97% InvertedIndex Wkpeda % 13.31% InvertedIndex Wkpeda % 16.44% InvertedIndex Wkpeda % 17.6% Job Completon Tme (Seconds) Task Task Natve DREAMS Sze of set (GB) (a) Sort Job Completon Tme (Seconds) Natve DREAMS Sze of set (GB) (b) InvertedIndex Fg. 6: Job completon tme of ndvdual jobs Map Stage Reduce Stage Tme (Seconds) 5 (a) Task Executon Tmelne Percentage CPU Memory Tme (Seconds) (b) CPU and Mem. Utl. Fg. 7: Sortng 1GB wth Natve Hadoop Map Stage Reduce Stage Tme (Seconds) (a) Task Executon Tmelne Percentage CPU Memory Tme (Seconds) (b) CPU and Mem. Utl. Fg. 8: Sortng 1GB wth DREAMS

7 Applcaton TABLE III: Workloads characterstcs Type Sze(GB) CV of Partton Szes #Map and Reduce tasks Sort Synthetc % 8,64 Sort Synthetc % 16,64 Sort Synthetc % 24,64 Sort Synthetc % 4,64 InvertedIndex Wkpeda % 73,8 InvertedIndex Wkpeda % 169,8 InvertedIndex Wkpeda % 252,8 InvertedIndex Wkpeda % 396,8 mnmal change to the exstng Hadoop archtecture. In ths secton, we compare the performance of DREAMS aganst natve Hadoop YARN 2.4. (called Natve n ths paper). The slowstart threshold s set to 1%, and the CgroupsLCEResourcesHandler s enabled. We frst evaluated DREAMS usng ndvdual Job (ether Sort or InvertedIndex) wth several nput data szes from 1 GB to 5 GB. Table III gves an overvew of these workloads. Note that tunng the number of reduce tasks for each workload can mprove job completon tme [18]. To solate ths effect, we fx the number of reduce tasks for each job. The CV (coeffcent of varaton) of partton szes represents the skewness of the reduce nput dstrbuton. We can see from the table that the CV values of all the workloads are less than 5% 4. The experment results are shown n Fgure 6. We can see from the fgure that DREAMS outperforms Natve for all cases. In partcular, DREAMS mproves job completon tme by 2.3% when sortng 5 GB data. To understand the reason behnd the performance gan, we plotted the tmelne and cluster CPU and memory usage of executng 1G Sort for Natve and DREAMS n Fgure 7 and Fgure 8. We found that DREAMS equalzes the duratons among reduce tasks, and acheves hgher CPU and Memory utlzaton than Natve n reduce stage. More specfcally, the utlzaton between DREAMS and Natve durng map stage s smlar; after map stage completes (around 15 seconds mark), both CPU and memory utlzaton of DREAMS become hgher than Natve. Furthermore, we have found that DREAMS generally acheves hgher reducton n job completon tme for Sort rather than InvertedIndex. That s because DREAMS only mproves the resource allocaton n reduce stage, but leavng map stage unchanged. And Sort s reduce ntensve, where reduce stage takes longer tme than map stage. As a result, DREAMS s able to provde hgher gan for the job runnng tme of reduce-ntensve jobs. We now present our evaluaton results usng multple jobs n parallel. Accordng to the cumulatve dstrbuton functon of job runnng tmes from a producton workload trace at Facebook [19], the job completon tmes follow a long-tal dstrbuton. More specfcally, most of the jobs (more than 5%) are less than 1 seconds long, and the dstrbuton of nterarrval tmes for ths workload trace s roughly exponental wth a mean of 14 seconds. Therefore, n ths evaluaton, we vary the number of jobs of 5G Sort and 5G InvertedIndex from 1 to 16 to create batch workloads, and submt the jobs wth an nter-arrval tme followng exponental dstrbuton wth a mean of 14 seconds. We run each of the batch workloads 5 tmes usng Natve and DREAMS. The results of average 4 The CV of each workload for Sort s the same, because these workloads are generated by the same RandomWrter. job completon tme are shown n Fgure 9a. It can be seen that DREAMS outperforms Natve Hadoop n all scenaros. Admttedly, the gan of DREAMS n experments wth multple jobs s less than the gan n sngle job experments. It s because reduce tasks of small jobs only last dozens of seconds, whch means the dfference between the longest and shortest task s only dozens of seconds. When the number of jobs s ncreasng, many short tasks are scheduled one after the other. As a result, there s a chance that these short tasks can ft nto resource vacancy that skewed tasks generate. Therefore, n some cases DREAMS can obtan only dozens of seconds gan for these small jobs (Note that dozens of seconds consttute a bg gan n sngle job scenaro). In the future, we ntend to evaluate DREAMS usng multple large jobs. Fgure 9b and 9c show the resource utlzaton of the cluster durng the executon of each batch for Natve and DREAMS respectvely. It can be seen from the dagrams that DREAMS acheves slghtly hgher CPU utlzaton than the Natve Hadoop, and the memory utlzatons of both methods are smlar. That s because the bggest partton sze of all the reduce tasks n ths workload s less than the mnmal memory allocaton, DREAMS does not adjust the memory. But wth respect to CPU allocaton, DREAMS makes an adjustment for dfferent reduce tasks, thereby achevng hgher CPU utlzaton. VI. RELATED WORK The data skew problem n MapReduce has been extensvely nvestgated recently. Kwon et.al. [2] present fve types of skews n MapReduce applcatons whch are caused by the characterstcs of the algorthm and dataset, and propose best practces to mtgate skew. On mtgatng the mpact of skewed data, several approaches have been proposed. The authors, n [7] and [6], defne a cost model for schedulng Reduce keys to reduce tasks so as to balance the load among reduce tasks. However, both approaches have to wat untl all the map tasks have completed. As shown n [5], ths would ncrease the job completon tme. In order to equally dstrbute the load to worker machnes whle overlappng the map and reduce phase, the proposal n [9] apples a Greedy-Balance approach of assgnng unassgned keys to the machne wth the least load. Ths soluton s based on the assumpton that the sze of each key-value par s dentcal, whch s not true n real workloads. Even though the results n ths paper show a reducton of maxmum load compared to default soluton, shuffle fnshng tme s worse than the default soluton. Also ths paper provdes no evaluaton about whether the job completon tme can be shortened. Unlke those later shufflng approaches, Ramakrshnan et.al. [8] propose a progressve sampler to estmate the ntermedate data dstrbuton and then partton the data to balance the load across all reduce tasks. However, ths soluton needs an addtonal samplng phase before jobs start, whch can be tme-consumng. Instead of choppng the large parttons to balance the load, SkewTune [4] reparttons heavly skewed parttons to acheve ths goal. However, t mposes an overhead whle reparttonng data and concatenatng orgnal output. Compared to SkewTune, our soluton dynamcally allocates the rght amount of resources to tasks to equalze the tasks completon tme, whch s smpler and ncurs no overhead. Fnally, Zachelas et al. propose DynamcShare [3], whch ams at schedulng MapReduce jobs n heterogeneous systems to meet ther real-tme response requrements, and achevng an

8 Average Job Completon Tme (Seconds) Natve DREAMS Number of Jobs (a) Job Completon Tme Utlzaton CPU Memory Number of Jobs (b) Resource Utl. of DREAMS Utlzaton CPU Memory Number of Jobs (c) Resource Utl. of Natve Fg. 9: Multple Jobs Benchmark even dstrbuton of the parttons by assgnng the parttons n such a way that puts more work on powerful nodes. Smlar to SkewTune, t mposes an overhead for the parttons assgnment procedure. Besdes, DynamcShare cannot start parttons assgnment untl all map tasks have completed. Resource-aware schedulng has receved consderable attenton n recent years. The orgnal Hadoop MapReduce mplements a slot-based resource allocaton scheme, whch does not take run-tme task resource consumpton nto consderaton. To address ths lmtaton, Hadoop YARN [2] represents a major endeavor towards resource-aware schedulng n MapReduce clusters. It offers the ablty to specfy the sze of contaner n terms of requrements for each type of resources. However, YARN assumes the resource consumpton for each Map (or Reduce) task n a job s dentcal, whch s not true for data skewed MapReduce jobs. Sharma et. al. propose MROrchestrator [21], a MapReduce resource framework that can dentfy the resource defct based on resource proflng, and dynamcally adjusts the resource allocaton. Compared wth our soluton, MROrchestrator cannot dentfy stragglers of workload mbalance before tasks launch, and t cannot judcously place tasks that need more resource on the machnes wth more free resources. In other words, f all CPUntensve tasks are launched n a machne, no matter how MROrchestrator adjusts the allocaton, resource defct cannot be mtgated. There are several other proposals that fall n another category of resource schedulng polces such as [11], [13], [22], [23]. The man focus of these approaches s on adjustng the resource allocaton n terms of the number of Map and Reduce slots for the jobs n order to acheve farness, maxmze resource utlzaton or meet job deadlne. These however do not address the data skew problem. VII. CONCLUSION MapReduce has become a predomnant model for largescale data processng n recent years. However, exstng MapReduce schedulers stll use a smple hash functon to assgn map outputs to reduce tasks. Ths smple data assgnment scheme may result n a phenomenon known as parttonng skew, where the output of map tasks s unevenly dstrbuted among reduce tasks. Whle many approaches have been proposed to address ths ssue, exstng solutons often ncur an addtonal overhead for run-tme partton sze predcton and data reparttonng. Motvated by ths lmtaton, n ths paper we present DREAMS, a framework for run-tme parttonng skew mtgaton. Unlke prevous approaches that try to balance the reduce workload by reparttonng the workload assgned to each reduce task, n DREAMS we cope wth parttonng skew by adjustng task run-tme resource allocaton. To do so, we frst develop an on-lne partton sze predcton model whch can estmate the partton szes of reduce tasks at run-tme. Our experments results show that the average relatve error s less than 8.2% n all cases. Second, we desgn a reduce task performance model that correlates task duraton wth run-tme resource allocaton and nput sze of reduce tasks. The valdaton results show that the worse predcton error s 19.57%. Thrd, we demonstrate the beneft of leveragng resource-awareness for run-tme skew mtgaton. Through experments usng real and synthetc workloads, we show that DREAMS can effectvely mtgate the negatve mpact of parttonng skew whle ncurrng neglgble overhead, thereby mprovng job runnng tme by up to 2.3%. ACKNOWLEDGEMENT Ths work s supported n part by the Natonal Natural Scence Foundaton of Chna (No ), and n part by the Smart Applcatons on Vrtual Infrastructure (SAVI) project funded under the Natonal Scences and Engneerng Research Councl of Canada (NSERC) Strategc Networks grant number NETGP REFERENCES [1] J. Dean and S. Ghemawat, Mapreduce: smplfed data processng on large clusters, Communcatons of the ACM, vol. 51, no. 1, pp , 28. [2] V. K. Vavlapall, A. C. Murthy, C. Douglas, S. Agarwal, M. Konar, R. Evans, T. Graves, J. Lowe, H. Shah, S. Seth et al., Apache hadoop yarn: Yet another resource negotator, n Proceedngs of the 4th annual Symposum on Cloud Computng. ACM, 213, p. 5. [3] A. Computng, N. Zachelas, and V. Kalogerak, Real-tme schedulng of skewed mapreduce jobs n heterogeneous envronments. [4] Y. Kwon, M. Balaznska, B. Howe, and J. Rola, Skewtune: mtgatng skew n mapreduce applcatons, n Proceedngs of the 212 ACM SIGMOD Internatonal Conference on Management of Data. ACM, 212, pp [5] M. Hammoud, M. S. Rehman, and M. F. Sakr, Center-of-gravty reduce task schedulng to lower mapreduce network traffc, n Cloud Computng (CLOUD), 212 IEEE 5th Internatonal Conference on. IEEE, 212, pp

9 [6] S. Ibrahm, H. Jn, L. Lu, B. He, G. Antonu, and S. Wu, Handlng parttonng skew n mapreduce usng leen, Peer-to-Peer Networkng and Applcatons, vol. 6, no. 4, pp , 213. [7] B. Gufler, N. Augsten, A. Reser, and A. Kemper, Handng data skew n mapreduce, n Proceedngs of the 1st Internatonal Conference on Cloud Computng and Servces Scence, vol. 146, 211, pp [8] S. R. Ramakrshnan, G. Swart, and A. Urmanov, Balancng reducer skew n mapreduce workloads usng progressve samplng, n Proceedngs of the Thrd ACM Symposum on Cloud Computng. ACM, 212, p. 16. [9] Y. Le, J. Lu, F. Ergun, and D. Wang, Onlne load balancng for mapreduce wth skewed data nput. [1] L. Cheng, Q. Zhang, and R. Boutaba, Mtgatng the negatve mpact of preempton on heterogeneous mapreduce workloads, n Proceedngs of the 7th Internatonal Conference on Network and Servces Management. Internatonal Federaton for Informaton Processng, 211, pp [11] A. Verma, L. Cherkasova, and R. H. Campbell, Ara: automatc resource nference and allocaton for mapreduce envronments, n Proceedngs of the 8th ACM nternatonal conference on Autonomc computng. ACM, 211, pp [12] Hadoop, Far scheduler, hadoop-yarn/hadoop-yarn-ste/farscheduler.html. [13] A. Ghods, M. Zahara, B. Hndman, A. Konwnsk, S. Shenker, and I. Stoca, Domnant resource farness: Far allocaton of multple resource types. n NSDI, vol. 11, 211, pp [14] Z. Zhang, L. Cherkasova, and B. T. Loo, Benchmarkng approach for desgnng a mapreduce performance model, n Proceedngs of the 4th ACM/SPEC Internatonal Conference on Performance Engneerng. ACM, 213, pp [15] D. M. Bates and D. G. Watts, Nonlnear regresson: teratve estmaton and lnear approxmatons. Wley Onlne Lbrary, [16] J.-M. Kang, H. Bannazadeh, and A. Leon-Garca, Sav testbed: Control and management of converged vrtual ct resources, n Integrated Network Management (IM 213), 213 IFIP/IEEE Internatonal Symposum on. IEEE, 213, pp [17] F. Ahmad, S. Lee, M. Thottethod, and T. Vjaykumar, Puma: Purdue mapreduce benchmarks sute, 212. [18] Z. Zhang, L. Cherkasova, and B. T. Loo, Autotune: Optmzng executon concurrency and resource usage n mapreduce workflows. n ICAC, 213, pp [19] M. Zahara, D. Borthakur, J. Sen Sarma, K. Elmeleegy, S. Shenker, and I. Stoca, Delay schedulng: a smple technque for achevng localty and farness n cluster schedulng, n Proceedngs of the 5th European conference on Computer systems. ACM, 21, pp [2] Y. Kwon, M. Balaznska, B. Howe, and J. Rola, A study of skew n mapreduce applcatons, Open Crrus Summt, 211. [21] B. Sharma, R. Prabhakar, S. Lm, M. T. Kandemr, and C. R. Das, Mrorchestrator: A fne-graned resource orchestraton framework for mapreduce clusters, n Cloud Computng (CLOUD), 212 IEEE 5th Internatonal Conference on. IEEE, 212, pp [22] J. Polo, D. Carrera, Y. Becerra, J. Torres, E. Ayguadé, M. Stender, and I. Whalley, Performance-drven task co-schedulng for mapreduce envronments, n Network Operatons and Management Symposum (NOMS), 21 IEEE. IEEE, 21, pp [23] J. Wolf, D. Rajan, K. Hldrum, R. Khandekar, V. Kumar, S. Parekh, K.- L. Wu, and A. Balmn, Flex: A slot allocaton schedulng optmzer for mapreduce workloads, n Mddleware 21. Sprnger, 21, pp. 1 2.

Dynamic Resource Allocation for MapReduce with Partitioning Skew

Dynamic Resource Allocation for MapReduce with Partitioning Skew Ths artcle has been accepted for publcaton n a future ssue of ths journal, but has not been fully edted. Content may change pror to fnal publcaton. Ctaton nformaton: DOI 1.119/TC.216.253286, IEEE Transactons

More information

Forecasting the Direction and Strength of Stock Market Movement

Forecasting the Direction and Strength of Stock Market Movement Forecastng the Drecton and Strength of Stock Market Movement Jngwe Chen Mng Chen Nan Ye cjngwe@stanford.edu mchen5@stanford.edu nanye@stanford.edu Abstract - Stock market s one of the most complcated systems

More information

The Development of Web Log Mining Based on Improve-K-Means Clustering Analysis

The Development of Web Log Mining Based on Improve-K-Means Clustering Analysis The Development of Web Log Mnng Based on Improve-K-Means Clusterng Analyss TngZhong Wang * College of Informaton Technology, Luoyang Normal Unversty, Luoyang, 471022, Chna wangtngzhong2@sna.cn Abstract.

More information

IWFMS: An Internal Workflow Management System/Optimizer for Hadoop

IWFMS: An Internal Workflow Management System/Optimizer for Hadoop IWFMS: An Internal Workflow Management System/Optmzer for Hadoop Lan Lu, Yao Shen Department of Computer Scence and Engneerng Shangha JaoTong Unversty Shangha, Chna lustrve@gmal.com, yshen@cs.sjtu.edu.cn

More information

Fault tolerance in cloud technologies presented as a service

Fault tolerance in cloud technologies presented as a service Internatonal Scentfc Conference Computer Scence 2015 Pavel Dzhunev, PhD student Fault tolerance n cloud technologes presented as a servce INTRODUCTION Improvements n technques for vrtualzaton and performance

More information

A Replication-Based and Fault Tolerant Allocation Algorithm for Cloud Computing

A Replication-Based and Fault Tolerant Allocation Algorithm for Cloud Computing A Replcaton-Based and Fault Tolerant Allocaton Algorthm for Cloud Computng Tork Altameem Dept of Computer Scence, RCC, Kng Saud Unversty, PO Box: 28095 11437 Ryadh-Saud Araba Abstract The very large nfrastructure

More information

An Alternative Way to Measure Private Equity Performance

An Alternative Way to Measure Private Equity Performance An Alternatve Way to Measure Prvate Equty Performance Peter Todd Parlux Investment Technology LLC Summary Internal Rate of Return (IRR) s probably the most common way to measure the performance of prvate

More information

A Programming Model for the Cloud Platform

A Programming Model for the Cloud Platform Internatonal Journal of Advanced Scence and Technology A Programmng Model for the Cloud Platform Xaodong Lu School of Computer Engneerng and Scence Shangha Unversty, Shangha 200072, Chna luxaodongxht@qq.com

More information

Methodology to Determine Relationships between Performance Factors in Hadoop Cloud Computing Applications

Methodology to Determine Relationships between Performance Factors in Hadoop Cloud Computing Applications Methodology to Determne Relatonshps between Performance Factors n Hadoop Cloud Computng Applcatons Lus Eduardo Bautsta Vllalpando 1,2, Alan Aprl 1 and Alan Abran 1 1 Department of Software Engneerng and

More information

What is Candidate Sampling

What is Candidate Sampling What s Canddate Samplng Say we have a multclass or mult label problem where each tranng example ( x, T ) conssts of a context x a small (mult)set of target classes T out of a large unverse L of possble

More information

Feature selection for intrusion detection. Slobodan Petrović NISlab, Gjøvik University College

Feature selection for intrusion detection. Slobodan Petrović NISlab, Gjøvik University College Feature selecton for ntruson detecton Slobodan Petrovć NISlab, Gjøvk Unversty College Contents The feature selecton problem Intruson detecton Traffc features relevant for IDS The CFS measure The mrmr measure

More information

A New Quality of Service Metric for Hard/Soft Real-Time Applications

A New Quality of Service Metric for Hard/Soft Real-Time Applications A New Qualty of Servce Metrc for Hard/Soft Real-Tme Applcatons Shaoxong Hua and Gang Qu Electrcal and Computer Engneerng Department and Insttute of Advanced Computer Study Unversty of Maryland, College

More information

The OC Curve of Attribute Acceptance Plans

The OC Curve of Attribute Acceptance Plans The OC Curve of Attrbute Acceptance Plans The Operatng Characterstc (OC) curve descrbes the probablty of acceptng a lot as a functon of the lot s qualty. Fgure 1 shows a typcal OC Curve. 10 8 6 4 1 3 4

More information

Frequency Selective IQ Phase and IQ Amplitude Imbalance Adjustments for OFDM Direct Conversion Transmitters

Frequency Selective IQ Phase and IQ Amplitude Imbalance Adjustments for OFDM Direct Conversion Transmitters Frequency Selectve IQ Phase and IQ Ampltude Imbalance Adjustments for OFDM Drect Converson ransmtters Edmund Coersmeer, Ernst Zelnsk Noka, Meesmannstrasse 103, 44807 Bochum, Germany edmund.coersmeer@noka.com,

More information

METHODOLOGY TO DETERMINE RELATIONSHIPS BETWEEN PERFORMANCE FACTORS IN HADOOP CLOUD COMPUTING APPLICATIONS

METHODOLOGY TO DETERMINE RELATIONSHIPS BETWEEN PERFORMANCE FACTORS IN HADOOP CLOUD COMPUTING APPLICATIONS METHODOLOGY TO DETERMINE RELATIONSHIPS BETWEEN PERFORMANCE FACTORS IN HADOOP CLOUD COMPUTING APPLICATIONS Lus Eduardo Bautsta Vllalpando 1,2, Alan Aprl 1 and Alan Abran 1 1 Department of Software Engneerng

More information

Module 2 LOSSLESS IMAGE COMPRESSION SYSTEMS. Version 2 ECE IIT, Kharagpur

Module 2 LOSSLESS IMAGE COMPRESSION SYSTEMS. Version 2 ECE IIT, Kharagpur Module LOSSLESS IMAGE COMPRESSION SYSTEMS Lesson 3 Lossless Compresson: Huffman Codng Instructonal Objectves At the end of ths lesson, the students should be able to:. Defne and measure source entropy..

More information

Forecasting the Demand of Emergency Supplies: Based on the CBR Theory and BP Neural Network

Forecasting the Demand of Emergency Supplies: Based on the CBR Theory and BP Neural Network 700 Proceedngs of the 8th Internatonal Conference on Innovaton & Management Forecastng the Demand of Emergency Supples: Based on the CBR Theory and BP Neural Network Fu Deqang, Lu Yun, L Changbng School

More information

CHOLESTEROL REFERENCE METHOD LABORATORY NETWORK. Sample Stability Protocol

CHOLESTEROL REFERENCE METHOD LABORATORY NETWORK. Sample Stability Protocol CHOLESTEROL REFERENCE METHOD LABORATORY NETWORK Sample Stablty Protocol Background The Cholesterol Reference Method Laboratory Network (CRMLN) developed certfcaton protocols for total cholesterol, HDL

More information

DEFINING %COMPLETE IN MICROSOFT PROJECT

DEFINING %COMPLETE IN MICROSOFT PROJECT CelersSystems DEFINING %COMPLETE IN MICROSOFT PROJECT PREPARED BY James E Aksel, PMP, PMI-SP, MVP For Addtonal Informaton about Earned Value Management Systems and reportng, please contact: CelersSystems,

More information

A Novel Methodology of Working Capital Management for Large. Public Constructions by Using Fuzzy S-curve Regression

A Novel Methodology of Working Capital Management for Large. Public Constructions by Using Fuzzy S-curve Regression Novel Methodology of Workng Captal Management for Large Publc Constructons by Usng Fuzzy S-curve Regresson Cheng-Wu Chen, Morrs H. L. Wang and Tng-Ya Hseh Department of Cvl Engneerng, Natonal Central Unversty,

More information

Study on Model of Risks Assessment of Standard Operation in Rural Power Network

Study on Model of Risks Assessment of Standard Operation in Rural Power Network Study on Model of Rsks Assessment of Standard Operaton n Rural Power Network Qngj L 1, Tao Yang 2 1 Qngj L, College of Informaton and Electrcal Engneerng, Shenyang Agrculture Unversty, Shenyang 110866,

More information

benefit is 2, paid if the policyholder dies within the year, and probability of death within the year is ).

benefit is 2, paid if the policyholder dies within the year, and probability of death within the year is ). REVIEW OF RISK MANAGEMENT CONCEPTS LOSS DISTRIBUTIONS AND INSURANCE Loss and nsurance: When someone s subject to the rsk of ncurrng a fnancal loss, the loss s generally modeled usng a random varable or

More information

Cloud Auto-Scaling with Deadline and Budget Constraints

Cloud Auto-Scaling with Deadline and Budget Constraints Prelmnary verson. Fnal verson appears In Proceedngs of 11th ACM/IEEE Internatonal Conference on Grd Computng (Grd 21). Oct 25-28, 21. Brussels, Belgum. Cloud Auto-Scalng wth Deadlne and Budget Constrants

More information

DBA-VM: Dynamic Bandwidth Allocator for Virtual Machines

DBA-VM: Dynamic Bandwidth Allocator for Virtual Machines DBA-VM: Dynamc Bandwdth Allocator for Vrtual Machnes Ahmed Amamou, Manel Bourguba, Kamel Haddadou and Guy Pujolle LIP6, Perre & Mare Cure Unversty, 4 Place Jusseu 755 Pars, France Gand SAS, 65 Boulevard

More information

Network Aware Load-Balancing via Parallel VM Migration for Data Centers

Network Aware Load-Balancing via Parallel VM Migration for Data Centers Network Aware Load-Balancng va Parallel VM Mgraton for Data Centers Kun-Tng Chen 2, Chen Chen 12, Po-Hsang Wang 2 1 Informaton Technology Servce Center, 2 Department of Computer Scence Natonal Chao Tung

More information

How To Calculate The Accountng Perod Of Nequalty

How To Calculate The Accountng Perod Of Nequalty Inequalty and The Accountng Perod Quentn Wodon and Shlomo Ytzha World Ban and Hebrew Unversty September Abstract Income nequalty typcally declnes wth the length of tme taen nto account for measurement.

More information

Open Access A Load Balancing Strategy with Bandwidth Constraint in Cloud Computing. Jing Deng 1,*, Ping Guo 2, Qi Li 3, Haizhu Chen 1

Open Access A Load Balancing Strategy with Bandwidth Constraint in Cloud Computing. Jing Deng 1,*, Ping Guo 2, Qi Li 3, Haizhu Chen 1 Send Orders for Reprnts to reprnts@benthamscence.ae The Open Cybernetcs & Systemcs Journal, 2014, 8, 115-121 115 Open Access A Load Balancng Strategy wth Bandwdth Constrant n Cloud Computng Jng Deng 1,*,

More information

An Evaluation of the Extended Logistic, Simple Logistic, and Gompertz Models for Forecasting Short Lifecycle Products and Services

An Evaluation of the Extended Logistic, Simple Logistic, and Gompertz Models for Forecasting Short Lifecycle Products and Services An Evaluaton of the Extended Logstc, Smple Logstc, and Gompertz Models for Forecastng Short Lfecycle Products and Servces Charles V. Trappey a,1, Hsn-yng Wu b a Professor (Management Scence), Natonal Chao

More information

Efficient Striping Techniques for Variable Bit Rate Continuous Media File Servers æ

Efficient Striping Techniques for Variable Bit Rate Continuous Media File Servers æ Effcent Strpng Technques for Varable Bt Rate Contnuous Meda Fle Servers æ Prashant J. Shenoy Harrck M. Vn Department of Computer Scence, Department of Computer Scences, Unversty of Massachusetts at Amherst

More information

Survey on Virtual Machine Placement Techniques in Cloud Computing Environment

Survey on Virtual Machine Placement Techniques in Cloud Computing Environment Survey on Vrtual Machne Placement Technques n Cloud Computng Envronment Rajeev Kumar Gupta and R. K. Paterya Department of Computer Scence & Engneerng, MANIT, Bhopal, Inda ABSTRACT In tradtonal data center

More information

Calculating the high frequency transmission line parameters of power cables

Calculating the high frequency transmission line parameters of power cables < ' Calculatng the hgh frequency transmsson lne parameters of power cables Authors: Dr. John Dcknson, Laboratory Servces Manager, N 0 RW E B Communcatons Mr. Peter J. Ncholson, Project Assgnment Manager,

More information

Answer: A). There is a flatter IS curve in the high MPC economy. Original LM LM after increase in M. IS curve for low MPC economy

Answer: A). There is a flatter IS curve in the high MPC economy. Original LM LM after increase in M. IS curve for low MPC economy 4.02 Quz Solutons Fall 2004 Multple-Choce Questons (30/00 ponts) Please, crcle the correct answer for each of the followng 0 multple-choce questons. For each queston, only one of the answers s correct.

More information

IMPACT ANALYSIS OF A CELLULAR PHONE

IMPACT ANALYSIS OF A CELLULAR PHONE 4 th ASA & μeta Internatonal Conference IMPACT AALYSIS OF A CELLULAR PHOE We Lu, 2 Hongy L Bejng FEAonlne Engneerng Co.,Ltd. Bejng, Chna ABSTRACT Drop test smulaton plays an mportant role n nvestgatng

More information

An Analysis of Central Processor Scheduling in Multiprogrammed Computer Systems

An Analysis of Central Processor Scheduling in Multiprogrammed Computer Systems STAN-CS-73-355 I SU-SE-73-013 An Analyss of Central Processor Schedulng n Multprogrammed Computer Systems (Dgest Edton) by Thomas G. Prce October 1972 Techncal Report No. 57 Reproducton n whole or n part

More information

A Load-Balancing Algorithm for Cluster-based Multi-core Web Servers

A Load-Balancing Algorithm for Cluster-based Multi-core Web Servers Journal of Computatonal Informaton Systems 7: 13 (2011) 4740-4747 Avalable at http://www.jofcs.com A Load-Balancng Algorthm for Cluster-based Mult-core Web Servers Guohua YOU, Yng ZHAO College of Informaton

More information

RESEARCH ON DUAL-SHAKER SINE VIBRATION CONTROL. Yaoqi FENG 1, Hanping QIU 1. China Academy of Space Technology (CAST) yaoqi.feng@yahoo.

RESEARCH ON DUAL-SHAKER SINE VIBRATION CONTROL. Yaoqi FENG 1, Hanping QIU 1. China Academy of Space Technology (CAST) yaoqi.feng@yahoo. ICSV4 Carns Australa 9- July, 007 RESEARCH ON DUAL-SHAKER SINE VIBRATION CONTROL Yaoq FENG, Hanpng QIU Dynamc Test Laboratory, BISEE Chna Academy of Space Technology (CAST) yaoq.feng@yahoo.com Abstract

More information

How To Understand The Results Of The German Meris Cloud And Water Vapour Product

How To Understand The Results Of The German Meris Cloud And Water Vapour Product Ttel: Project: Doc. No.: MERIS level 3 cloud and water vapour products MAPP MAPP-ATBD-ClWVL3 Issue: 1 Revson: 0 Date: 9.12.1998 Functon Name Organsaton Sgnature Date Author: Bennartz FUB Preusker FUB Schüller

More information

Performance Analysis of Energy Consumption of Smartphone Running Mobile Hotspot Application

Performance Analysis of Energy Consumption of Smartphone Running Mobile Hotspot Application Internatonal Journal of mart Grd and lean Energy Performance Analyss of Energy onsumpton of martphone Runnng Moble Hotspot Applcaton Yun on hung a chool of Electronc Engneerng, oongsl Unversty, 511 angdo-dong,

More information

A New Task Scheduling Algorithm Based on Improved Genetic Algorithm

A New Task Scheduling Algorithm Based on Improved Genetic Algorithm A New Task Schedulng Algorthm Based on Improved Genetc Algorthm n Cloud Computng Envronment Congcong Xong, Long Feng, Lxan Chen A New Task Schedulng Algorthm Based on Improved Genetc Algorthm n Cloud Computng

More information

1 Example 1: Axis-aligned rectangles

1 Example 1: Axis-aligned rectangles COS 511: Theoretcal Machne Learnng Lecturer: Rob Schapre Lecture # 6 Scrbe: Aaron Schld February 21, 2013 Last class, we dscussed an analogue for Occam s Razor for nfnte hypothess spaces that, n conjuncton

More information

ANALYZING THE RELATIONSHIPS BETWEEN QUALITY, TIME, AND COST IN PROJECT MANAGEMENT DECISION MAKING

ANALYZING THE RELATIONSHIPS BETWEEN QUALITY, TIME, AND COST IN PROJECT MANAGEMENT DECISION MAKING ANALYZING THE RELATIONSHIPS BETWEEN QUALITY, TIME, AND COST IN PROJECT MANAGEMENT DECISION MAKING Matthew J. Lberatore, Department of Management and Operatons, Vllanova Unversty, Vllanova, PA 19085, 610-519-4390,

More information

Profit-Aware DVFS Enabled Resource Management of IaaS Cloud

Profit-Aware DVFS Enabled Resource Management of IaaS Cloud IJCSI Internatonal Journal of Computer Scence Issues, Vol. 0, Issue, No, March 03 ISSN (Prnt): 694-084 ISSN (Onlne): 694-0784 www.ijcsi.org 37 Proft-Aware DVFS Enabled Resource Management of IaaS Cloud

More information

Network Security Situation Evaluation Method for Distributed Denial of Service

Network Security Situation Evaluation Method for Distributed Denial of Service Network Securty Stuaton Evaluaton Method for Dstrbuted Denal of Servce Jn Q,2, Cu YMn,2, Huang MnHuan,2, Kuang XaoHu,2, TangHong,2 ) Scence and Technology on Informaton System Securty Laboratory, Bejng,

More information

Self-Adaptive SLA-Driven Capacity Management for Internet Services

Self-Adaptive SLA-Driven Capacity Management for Internet Services Self-Adaptve SLA-Drven Capacty Management for Internet Servces Bruno Abrahao, Vrglo Almeda and Jussara Almeda Computer Scence Department Federal Unversty of Mnas Geras, Brazl Alex Zhang, Drk Beyer and

More information

Data Broadcast on a Multi-System Heterogeneous Overlayed Wireless Network *

Data Broadcast on a Multi-System Heterogeneous Overlayed Wireless Network * JOURNAL OF INFORMATION SCIENCE AND ENGINEERING 24, 819-840 (2008) Data Broadcast on a Mult-System Heterogeneous Overlayed Wreless Network * Department of Computer Scence Natonal Chao Tung Unversty Hsnchu,

More information

This article appeared in a journal published by Elsevier. The attached copy is furnished to the author for internal non-commercial research and

This article appeared in a journal published by Elsevier. The attached copy is furnished to the author for internal non-commercial research and Ths artcle appeared n a journal publshed by Elsever. The attached copy s furnshed to the author for nternal non-commercal research and educaton use, ncludng for nstructon at the authors nsttuton and sharng

More information

Politecnico di Torino. Porto Institutional Repository

Politecnico di Torino. Porto Institutional Repository Poltecnco d Torno Porto Insttutonal Repostory [Artcle] A cost-effectve cloud computng framework for acceleratng multmeda communcaton smulatons Orgnal Ctaton: D. Angel, E. Masala (2012). A cost-effectve

More information

An Interest-Oriented Network Evolution Mechanism for Online Communities

An Interest-Oriented Network Evolution Mechanism for Online Communities An Interest-Orented Network Evoluton Mechansm for Onlne Communtes Cahong Sun and Xaopng Yang School of Informaton, Renmn Unversty of Chna, Bejng 100872, P.R. Chna {chsun,yang}@ruc.edu.cn Abstract. Onlne

More information

Improved SVM in Cloud Computing Information Mining

Improved SVM in Cloud Computing Information Mining Internatonal Journal of Grd Dstrbuton Computng Vol.8, No.1 (015), pp.33-40 http://dx.do.org/10.1457/jgdc.015.8.1.04 Improved n Cloud Computng Informaton Mnng Lvshuhong (ZhengDe polytechnc college JangSu

More information

Cloud-based Social Application Deployment using Local Processing and Global Distribution

Cloud-based Social Application Deployment using Local Processing and Global Distribution Cloud-based Socal Applcaton Deployment usng Local Processng and Global Dstrbuton Zh Wang *, Baochun L, Lfeng Sun *, and Shqang Yang * * Bejng Key Laboratory of Networked Multmeda Department of Computer

More information

Minimal Coding Network With Combinatorial Structure For Instantaneous Recovery From Edge Failures

Minimal Coding Network With Combinatorial Structure For Instantaneous Recovery From Edge Failures Mnmal Codng Network Wth Combnatoral Structure For Instantaneous Recovery From Edge Falures Ashly Joseph 1, Mr.M.Sadsh Sendl 2, Dr.S.Karthk 3 1 Fnal Year ME CSE Student Department of Computer Scence Engneerng

More information

Enabling P2P One-view Multi-party Video Conferencing

Enabling P2P One-view Multi-party Video Conferencing Enablng P2P One-vew Mult-party Vdeo Conferencng Yongxang Zhao, Yong Lu, Changja Chen, and JanYn Zhang Abstract Mult-Party Vdeo Conferencng (MPVC) facltates realtme group nteracton between users. Whle P2P

More information

How To Solve An Onlne Control Polcy On A Vrtualzed Data Center

How To Solve An Onlne Control Polcy On A Vrtualzed Data Center Dynamc Resource Allocaton and Power Management n Vrtualzed Data Centers Rahul Urgaonkar, Ulas C. Kozat, Ken Igarash, Mchael J. Neely urgaonka@usc.edu, {kozat, garash}@docomolabs-usa.com, mjneely@usc.edu

More information

Fair Virtual Bandwidth Allocation Model in Virtual Data Centers

Fair Virtual Bandwidth Allocation Model in Virtual Data Centers Far Vrtual Bandwdth Allocaton Model n Vrtual Data Centers Yng Yuan, Cu-rong Wang, Cong Wang School of Informaton Scence and Engneerng ortheastern Unversty Shenyang, Chna School of Computer and Communcaton

More information

Statistical Methods to Develop Rating Models

Statistical Methods to Develop Rating Models Statstcal Methods to Develop Ratng Models [Evelyn Hayden and Danel Porath, Österrechsche Natonalbank and Unversty of Appled Scences at Manz] Source: The Basel II Rsk Parameters Estmaton, Valdaton, and

More information

SPEE Recommended Evaluation Practice #6 Definition of Decline Curve Parameters Background:

SPEE Recommended Evaluation Practice #6 Definition of Decline Curve Parameters Background: SPEE Recommended Evaluaton Practce #6 efnton of eclne Curve Parameters Background: The producton hstores of ol and gas wells can be analyzed to estmate reserves and future ol and gas producton rates and

More information

On the Optimal Control of a Cascade of Hydro-Electric Power Stations

On the Optimal Control of a Cascade of Hydro-Electric Power Stations On the Optmal Control of a Cascade of Hydro-Electrc Power Statons M.C.M. Guedes a, A.F. Rbero a, G.V. Smrnov b and S. Vlela c a Department of Mathematcs, School of Scences, Unversty of Porto, Portugal;

More information

Calculation of Sampling Weights

Calculation of Sampling Weights Perre Foy Statstcs Canada 4 Calculaton of Samplng Weghts 4.1 OVERVIEW The basc sample desgn used n TIMSS Populatons 1 and 2 was a two-stage stratfed cluster desgn. 1 The frst stage conssted of a sample

More information

Multiple-Period Attribution: Residuals and Compounding

Multiple-Period Attribution: Residuals and Compounding Multple-Perod Attrbuton: Resduals and Compoundng Our revewer gave these authors full marks for dealng wth an ssue that performance measurers and vendors often regard as propretary nformaton. In 1994, Dens

More information

Traffic-light a stress test for life insurance provisions

Traffic-light a stress test for life insurance provisions MEMORANDUM Date 006-09-7 Authors Bengt von Bahr, Göran Ronge Traffc-lght a stress test for lfe nsurance provsons Fnansnspetonen P.O. Box 6750 SE-113 85 Stocholm [Sveavägen 167] Tel +46 8 787 80 00 Fax

More information

A Hierarchical Anomaly Network Intrusion Detection System using Neural Network Classification

A Hierarchical Anomaly Network Intrusion Detection System using Neural Network Classification IDC IDC A Herarchcal Anomaly Network Intruson Detecton System usng Neural Network Classfcaton ZHENG ZHANG, JUN LI, C. N. MANIKOPOULOS, JAY JORGENSON and JOSE UCLES ECE Department, New Jersey Inst. of Tech.,

More information

Optimal Map Reduce Job Capacity Allocation in Cloud Systems

Optimal Map Reduce Job Capacity Allocation in Cloud Systems Optmal Map Reduce Job Capacty Allocaton n Cloud Systems Marzeh Malemajd Sharf Unversty of Technology, Iran malemajd@ce.sharf.edu Danlo Ardagna Poltecnco d Mlano, Italy danlo.ardagna@polm.t Mchele Cavotta

More information

A heuristic task deployment approach for load balancing

A heuristic task deployment approach for load balancing Xu Gaochao, Dong Yunmeng, Fu Xaodog, Dng Yan, Lu Peng, Zhao Ja Abstract A heurstc task deployment approach for load balancng Gaochao Xu, Yunmeng Dong, Xaodong Fu, Yan Dng, Peng Lu, Ja Zhao * College of

More information

J. Parallel Distrib. Comput. Environment-conscious scheduling of HPC applications on distributed Cloud-oriented data centers

J. Parallel Distrib. Comput. Environment-conscious scheduling of HPC applications on distributed Cloud-oriented data centers J. Parallel Dstrb. Comput. 71 (2011) 732 749 Contents lsts avalable at ScenceDrect J. Parallel Dstrb. Comput. ournal homepage: www.elsever.com/locate/pdc Envronment-conscous schedulng of HPC applcatons

More information

INVESTIGATION OF VEHICULAR USERS FAIRNESS IN CDMA-HDR NETWORKS

INVESTIGATION OF VEHICULAR USERS FAIRNESS IN CDMA-HDR NETWORKS 21 22 September 2007, BULGARIA 119 Proceedngs of the Internatonal Conference on Informaton Technologes (InfoTech-2007) 21 st 22 nd September 2007, Bulgara vol. 2 INVESTIGATION OF VEHICULAR USERS FAIRNESS

More information

Joint Scheduling of Processing and Shuffle Phases in MapReduce Systems

Joint Scheduling of Processing and Shuffle Phases in MapReduce Systems Jont Schedulng of Processng and Shuffle Phases n MapReduce Systems Fangfe Chen, Mural Kodalam, T. V. Lakshman Department of Computer Scence and Engneerng, The Penn State Unversty Bell Laboratores, Alcatel-Lucent

More information

Load Balancing By Max-Min Algorithm in Private Cloud Environment

Load Balancing By Max-Min Algorithm in Private Cloud Environment Internatonal Journal of Scence and Research (IJSR ISSN (Onlne: 2319-7064 Index Coperncus Value (2013: 6.14 Impact Factor (2013: 4.438 Load Balancng By Max-Mn Algorthm n Prvate Cloud Envronment S M S Suntharam

More information

A Performance Analysis of View Maintenance Techniques for Data Warehouses

A Performance Analysis of View Maintenance Techniques for Data Warehouses A Performance Analyss of Vew Mantenance Technques for Data Warehouses Xng Wang Dell Computer Corporaton Round Roc, Texas Le Gruenwald The nversty of Olahoma School of Computer Scence orman, OK 739 Guangtao

More information

An Integrated Dynamic Resource Scheduling Framework in On-Demand Clouds *

An Integrated Dynamic Resource Scheduling Framework in On-Demand Clouds * JOURNAL OF INFORMATION SCIENCE AND ENGINEERING 30, 1537-1552 (2014) An Integrated Dynamc Resource Schedulng Framework n On-Demand Clouds * College of Computer Scence and Technology Zhejang Unversty Hangzhou,

More information

Can Auto Liability Insurance Purchases Signal Risk Attitude?

Can Auto Liability Insurance Purchases Signal Risk Attitude? Internatonal Journal of Busness and Economcs, 2011, Vol. 10, No. 2, 159-164 Can Auto Lablty Insurance Purchases Sgnal Rsk Atttude? Chu-Shu L Department of Internatonal Busness, Asa Unversty, Tawan Sheng-Chang

More information

Multi-Resource Fair Allocation in Heterogeneous Cloud Computing Systems

Multi-Resource Fair Allocation in Heterogeneous Cloud Computing Systems 1 Mult-Resource Far Allocaton n Heterogeneous Cloud Computng Systems We Wang, Student Member, IEEE, Ben Lang, Senor Member, IEEE, Baochun L, Senor Member, IEEE Abstract We study the mult-resource allocaton

More information

Mining Multiple Large Data Sources

Mining Multiple Large Data Sources The Internatonal Arab Journal of Informaton Technology, Vol. 7, No. 3, July 2 24 Mnng Multple Large Data Sources Anmesh Adhkar, Pralhad Ramachandrarao 2, Bhanu Prasad 3, and Jhml Adhkar 4 Department of

More information

To manage leave, meeting institutional requirements and treating individual staff members fairly and consistently.

To manage leave, meeting institutional requirements and treating individual staff members fairly and consistently. Corporate Polces & Procedures Human Resources - Document CPP216 Leave Management Frst Produced: Current Verson: Past Revsons: Revew Cycle: Apples From: 09/09/09 26/10/12 09/09/09 3 years Immedately Authorsaton:

More information

On-Line Fault Detection in Wind Turbine Transmission System using Adaptive Filter and Robust Statistical Features

On-Line Fault Detection in Wind Turbine Transmission System using Adaptive Filter and Robust Statistical Features On-Lne Fault Detecton n Wnd Turbne Transmsson System usng Adaptve Flter and Robust Statstcal Features Ruoyu L Remote Dagnostcs Center SKF USA Inc. 3443 N. Sam Houston Pkwy., Houston TX 77086 Emal: ruoyu.l@skf.com

More information

Sketching Sampled Data Streams

Sketching Sampled Data Streams Sketchng Sampled Data Streams Florn Rusu, Aln Dobra CISE Department Unversty of Florda Ganesvlle, FL, USA frusu@cse.ufl.edu adobra@cse.ufl.edu Abstract Samplng s used as a unversal method to reduce the

More information

Self-Adaptive Capacity Management for Multi-Tier Virtualized Environments

Self-Adaptive Capacity Management for Multi-Tier Virtualized Environments Self-Adaptve Capacty Management for Mult-Ter Vrtualzed Envronments Ítalo Cunha, Jussara Almeda, Vrgílo Almeda, Marcos Santos Computer Scence Department Federal Unversty of Mnas Geras Belo Horzonte, Brazl,

More information

Luby s Alg. for Maximal Independent Sets using Pairwise Independence

Luby s Alg. for Maximal Independent Sets using Pairwise Independence Lecture Notes for Randomzed Algorthms Luby s Alg. for Maxmal Independent Sets usng Parwse Independence Last Updated by Erc Vgoda on February, 006 8. Maxmal Independent Sets For a graph G = (V, E), an ndependent

More information

1. Fundamentals of probability theory 2. Emergence of communication traffic 3. Stochastic & Markovian Processes (SP & MP)

1. Fundamentals of probability theory 2. Emergence of communication traffic 3. Stochastic & Markovian Processes (SP & MP) 6.3 / -- Communcaton Networks II (Görg) SS20 -- www.comnets.un-bremen.de Communcaton Networks II Contents. Fundamentals of probablty theory 2. Emergence of communcaton traffc 3. Stochastc & Markovan Processes

More information

RequIn, a tool for fast web traffic inference

RequIn, a tool for fast web traffic inference RequIn, a tool for fast web traffc nference Olver aul, Jean Etenne Kba GET/INT, LOR Department 9 rue Charles Fourer 90 Evry, France Olver.aul@nt-evry.fr, Jean-Etenne.Kba@nt-evry.fr Abstract As networked

More information

Lecture 2: Single Layer Perceptrons Kevin Swingler

Lecture 2: Single Layer Perceptrons Kevin Swingler Lecture 2: Sngle Layer Perceptrons Kevn Sngler kms@cs.str.ac.uk Recap: McCulloch-Ptts Neuron Ths vastly smplfed model of real neurons s also knon as a Threshold Logc Unt: W 2 A Y 3 n W n. A set of synapses

More information

Section 5.4 Annuities, Present Value, and Amortization

Section 5.4 Annuities, Present Value, and Amortization Secton 5.4 Annutes, Present Value, and Amortzaton Present Value In Secton 5.2, we saw that the present value of A dollars at nterest rate per perod for n perods s the amount that must be deposted today

More information

How To Know The Components Of Mean Squared Error Of Herarchcal Estmator S

How To Know The Components Of Mean Squared Error Of Herarchcal Estmator S S C H E D A E I N F O R M A T I C A E VOLUME 0 0 On Mean Squared Error of Herarchcal Estmator Stans law Brodowsk Faculty of Physcs, Astronomy, and Appled Computer Scence, Jagellonan Unversty, Reymonta

More information

J. Parallel Distrib. Comput.

J. Parallel Distrib. Comput. J. Parallel Dstrb. Comput. 71 (2011) 62 76 Contents lsts avalable at ScenceDrect J. Parallel Dstrb. Comput. journal homepage: www.elsever.com/locate/jpdc Optmzng server placement n dstrbuted systems n

More information

A hybrid global optimization algorithm based on parallel chaos optimization and outlook algorithm

A hybrid global optimization algorithm based on parallel chaos optimization and outlook algorithm Avalable onlne www.ocpr.com Journal of Chemcal and Pharmaceutcal Research, 2014, 6(7):1884-1889 Research Artcle ISSN : 0975-7384 CODEN(USA) : JCPRC5 A hybrd global optmzaton algorthm based on parallel

More information

Dominant Resource Fairness in Cloud Computing Systems with Heterogeneous Servers

Dominant Resource Fairness in Cloud Computing Systems with Heterogeneous Servers 1 Domnant Resource Farness n Cloud Computng Systems wth Heterogeneous Servers We Wang, Baochun L, Ben Lang Department of Electrcal and Computer Engneerng Unversty of Toronto arxv:138.83v1 [cs.dc] 1 Aug

More information

8 Algorithm for Binary Searching in Trees

8 Algorithm for Binary Searching in Trees 8 Algorthm for Bnary Searchng n Trees In ths secton we present our algorthm for bnary searchng n trees. A crucal observaton employed by the algorthm s that ths problem can be effcently solved when the

More information

A DYNAMIC CRASHING METHOD FOR PROJECT MANAGEMENT USING SIMULATION-BASED OPTIMIZATION. Michael E. Kuhl Radhamés A. Tolentino-Peña

A DYNAMIC CRASHING METHOD FOR PROJECT MANAGEMENT USING SIMULATION-BASED OPTIMIZATION. Michael E. Kuhl Radhamés A. Tolentino-Peña Proceedngs of the 2008 Wnter Smulaton Conference S. J. Mason, R. R. Hll, L. Mönch, O. Rose, T. Jefferson, J. W. Fowler eds. A DYNAMIC CRASHING METHOD FOR PROJECT MANAGEMENT USING SIMULATION-BASED OPTIMIZATION

More information

Statistical Scalability Analysis of Communication Operations in Distributed Applications

Statistical Scalability Analysis of Communication Operations in Distributed Applications Statstcal Scalablty Analyss of Communcaton Operatons n Dstrbuted Applcatons Jeffrey S. Vetter Mchael O. McCracken Center for Appled Scentfc Computng Lawrence Lvermore Natonal Laboratory Lvermore, Calforna,

More information

Causal, Explanatory Forecasting. Analysis. Regression Analysis. Simple Linear Regression. Which is Independent? Forecasting

Causal, Explanatory Forecasting. Analysis. Regression Analysis. Simple Linear Regression. Which is Independent? Forecasting Causal, Explanatory Forecastng Assumes cause-and-effect relatonshp between system nputs and ts output Forecastng wth Regresson Analyss Rchard S. Barr Inputs System Cause + Effect Relatonshp The job of

More information

1. Measuring association using correlation and regression

1. Measuring association using correlation and regression How to measure assocaton I: Correlaton. 1. Measurng assocaton usng correlaton and regresson We often would lke to know how one varable, such as a mother's weght, s related to another varable, such as a

More information

8.5 UNITARY AND HERMITIAN MATRICES. The conjugate transpose of a complex matrix A, denoted by A*, is given by

8.5 UNITARY AND HERMITIAN MATRICES. The conjugate transpose of a complex matrix A, denoted by A*, is given by 6 CHAPTER 8 COMPLEX VECTOR SPACES 5. Fnd the kernel of the lnear transformaton gven n Exercse 5. In Exercses 55 and 56, fnd the mage of v, for the ndcated composton, where and are gven by the followng

More information

PAS: A Packet Accounting System to Limit the Effects of DoS & DDoS. Debish Fesehaye & Klara Naherstedt University of Illinois-Urbana Champaign

PAS: A Packet Accounting System to Limit the Effects of DoS & DDoS. Debish Fesehaye & Klara Naherstedt University of Illinois-Urbana Champaign PAS: A Packet Accountng System to Lmt the Effects of DoS & DDoS Debsh Fesehaye & Klara Naherstedt Unversty of Illnos-Urbana Champagn DoS and DDoS DDoS attacks are ncreasng threats to our dgtal world. Exstng

More information

Multi-Source Video Multicast in Peer-to-Peer Networks

Multi-Source Video Multicast in Peer-to-Peer Networks ult-source Vdeo ultcast n Peer-to-Peer Networks Francsco de Asís López-Fuentes*, Eckehard Stenbach Technsche Unverstät ünchen Insttute of Communcaton Networks, eda Technology Group 80333 ünchen, Germany

More information

A Dynamic Load Balancing for Massive Multiplayer Online Game Server

A Dynamic Load Balancing for Massive Multiplayer Online Game Server A Dynamc Load Balancng for Massve Multplayer Onlne Game Server Jungyoul Lm, Jaeyong Chung, Jnryong Km and Kwanghyun Shm Dgtal Content Research Dvson Electroncs and Telecommuncatons Research Insttute Daejeon,

More information

TIGER:Thermal-Aware File Assignment in Storage Clusters

TIGER:Thermal-Aware File Assignment in Storage Clusters TIGER:Thermal-Aware Fle Assgnment n Storage Clusters Ajt Chavan, Xunfe Jang, Mohemmad I. Alghamd, Xao Qn,Mnghua Jang, and Jfu Zhang Department of Computer Scence and Software Engneerng, Auburn Unversty,

More information

Checkng and Testng in Nokia RMS Process

Checkng and Testng in Nokia RMS Process An Integrated Schedulng Mechansm for Fault-Tolerant Modular Avoncs Systems Yann-Hang Lee Mohamed Youns Jeff Zhou CISE Department Unversty of Florda Ganesvlle, FL 326 yhlee@cse.ufl.edu Advanced System Technology

More information

The Greedy Method. Introduction. 0/1 Knapsack Problem

The Greedy Method. Introduction. 0/1 Knapsack Problem The Greedy Method Introducton We have completed data structures. We now are gong to look at algorthm desgn methods. Often we are lookng at optmzaton problems whose performance s exponental. For an optmzaton

More information

SMART: Scalable, Bandwidth-Aware Monitoring of Continuous Aggregation Queries

SMART: Scalable, Bandwidth-Aware Monitoring of Continuous Aggregation Queries : Scalable, Bandwdth-Aware Montorng of Contnuous Aggregaton Queres Navendu Jan, Praveen Yalagandula, Mke Dahln, and Yn Zhang Unversty of Texas at Austn HP Labs ABSTRACT We present, a scalable, bandwdth-aware

More information

Bayesian Network Based Causal Relationship Identification and Funding Success Prediction in P2P Lending

Bayesian Network Based Causal Relationship Identification and Funding Success Prediction in P2P Lending Proceedngs of 2012 4th Internatonal Conference on Machne Learnng and Computng IPCSIT vol. 25 (2012) (2012) IACSIT Press, Sngapore Bayesan Network Based Causal Relatonshp Identfcaton and Fundng Success

More information

) of the Cell class is created containing information about events associated with the cell. Events are added to the Cell instance

) of the Cell class is created containing information about events associated with the cell. Events are added to the Cell instance Calbraton Method Instances of the Cell class (one nstance for each FMS cell) contan ADC raw data and methods assocated wth each partcular FMS cell. The calbraton method ncludes event selecton (Class Cell

More information