Adaptive Load Balancing of Parallel Applications with Multi-Agent Reinforcement Learning on Heterogeneous Systems

Size: px
Start display at page:

Download "Adaptive Load Balancing of Parallel Applications with Multi-Agent Reinforcement Learning on Heterogeneous Systems"

Transcription

1 Adatve Load Balancng of Parallel Alcatons wth Mult-Agent Renforcement Learnng on Heterogeneous Systems Johan PAREN, Kata VERBEECK, Ann NOWE, Krs SEENHAU COMO, VUB Brussels, Belgum Emal: {kaverbee, an.nowe, Jan LEMEIRE, Erk DIRKX PADX, VUB Brussels, Belgum Emal: Submtted for Scentfc Programmng ournal, secal ssue on Dstrbuted Comutng and Comutaton, IOS Press ABSRAC We reort on the mrovements that can be acheved by alyng machne learnng technques, n tcular renforcement learnng, for the dynamc load balancng of allel alcatons. he alcatons beng consdered here are coarse gran data ntensve alcatons. Such alcatons ut hgh ressure on the nterconnect of the hardware. Synchronzaton and load balancng n comlex, heterogeneous networks need fast, flexble, adatve load balancng algorthms. Vewng a allel alcaton as a one-state coordnaton game n the framework of mult-agent renforcement learnng, and by usng a recently ntroduced mult-agent exloraton technque, we are able to mrove uon the classc ob farmng aroach. he mrovements are acheved wth lmted comutaton and communcaton overhead. Keywords: Parallel rocessng, Adatve load balancng, renforcement learnng, heterogeneous network, ntellgent agents, data ntensve alcatons. 1. INRODUCION Load balancng s crucal for allel alcatons snce t ensures a good use of the caacty of the allel rocessng unts. We look at alcatons whch ut hgh demands on the allel nterconnect n terms of throughut. Examles are comresson alcatons whch both rocess mortant amounts of data and requre a lot of comutatons. Data ntensve alcatons [2] requre a lot of communcaton and are therefore dreaded for most allel archtectures. he roblem s exacerbated when workng wth heterogeneous allel hardware. hs s the case n our exerment usng a heterogeneous cluster of PC s to execute allel alcatons wth a master-slave software archtecture. Adatve load balancng s ndsensable f system erformance s unredctable and no ror knowledge s avalable [1]. In the mult-agent communty, adatve load balancng s an nterestng testbed for mult-agent learnng algorthms, lkewse for mult-agent renforcement algorthms as n [9][12]. However the nterretatons and models of load balancng there are not always n the vew of real allel alcatons. We reort on the results obtaned by adatve agents n the farmng scheme. he dea s to vew all slaves as ndeendent renforcement learnng agents who try to learn the amount of data to request from the master, so as to mnmze the total run tme of the allel alcaton and/or the total dle tme of the master. As the agents share a common goal, from the game theoretcal ont of vew [10], ths setu can be seen as a coordnaton game. Recently a new exloraton technque for ndvdual renforcement learners used n coordnaton games was ntroduced, see [16][17]. hs technque allows the slave to learn ndeendently and adatvely whch amount of data to request to the master; amng at an effcent use of the communcaton lnk by the grou of agents. Our results show that the mult-agent learnng technque mroves uon the sequental ob farmng scheme for allelzng data ntensve alcatons. Secton 2 defnes the roblem of load balancng and gves an overvew of the exstng load balancng strateges. Secton 3 dscusses erformance metrcs, whle secton 4 ntroduces renforcement learnng and outlnes the mult-agent renforcement learnng algorthm. Secton 5 and 6 gve the exermental setu and results acheved. Fnally some conclusons are drawn n secton LOAD BALANCING/JOB SCHEDULING Load balancng ams at assgnng to each rocessor an amount of work roortonal to ts erformance, mnmzng the executon tme of the rogram. However, rocessor heterogenety and erformance fluctuatons make statc load balancng nsuffcent [1]. We nvestgate dynamc, local, dstrbuted load balancng strateges [18], whch are based on heurstcs, snce fndng the otmal soluton has shown to be NP-comlete n general [11]. Followng the agent hlosohy, the request assgnment strategy s a recever-ntated algorthm [6], n whch the consumers of workload look for roducers [13]. he goal s a fast adatve system that otmzes comutaton and synchronzaton. Problem descrton In stuatons were the communcaton tme s not neglgble, as s the case for data ntensve alcatons, faster rocessng unts can ncur serous enaltes due to slower unts. A data request ssued

2 by a slow unt can stall a faster unt when usng farmng. hs of course results n a reducton of the allelsm. hs henomenon s bound to occur when slaves request dentcal amounts of data from the master. hs s ndeendently of the actual amount by neglectng the communcaton delay, accetable gven suffcently large requests. In order to mrove uon the ob-farmng scheme when workng wth heterogeneous hardware, the slaves have to request dfferent amount of data from the master (server). Indeed ther resectve consumton of communcaton bandwdth should be roortonal to ther rocessng ower. Slower rocessng unts should avod obstructng faster ones by requestng less data. Comutaton model he ntal comutaton model s ob farmng. In ths master-slave archtecture the slaves (one er rocessng unt) request a chunk of a certan sze from the master. As soon as the data has been rocessed the result s transferred to the master and the slave sends a new request (fgure 1). 1. request chunk 4. send result master 2. send chunk communcaton bottleneck slave 1 slave 2 slave 3. data heterogeneous crunchng rocessors Fg 1: Model. hs scheme has the advantage of beng both smle and effcent. Indeed, n the case of heterogeneous hardware, the load (amount of rocessed data) wll deend on the rocessng seed of the dfferent rocessng unts. Faster rocessng unts wll more frequently request data and thus be able to rocess more data. he bottleneck of data ntensve alcatons wth a master-slave archtecture s the lnk connectng the slaves to the master. In the resented exerments all the slaves share a sngle lnk to ther master (through an ethernet swtch). In ths scenaro the alcaton s erformance wll be nfluenced by the effcent use of the shared lnk to the master. Indeed, the fact that the alcaton has a coarse granularty only ensures that the comutaton communcaton rato s larger than one. But t does not reclude a low allel effcency even when usng ob farmng. 3. PERFORMANCE MERICS We quantfy the beneft of allel rocessng by the seedu S= seq /, how much faster the allel rogram runs wth resect to the runtme of the sequental verson. he orton of that s not used for useful comutaton s therefore as consdered lost rocessor cycles [4], or overhead: Overhead =. s. ( 1) Hence, the choce of seedu as the erformance measure mles means that each rocessor has tme allocated to erform ts t of the ob: 1 2 = = =... =. ( 2) Wth +, =. ( 3) com com overhead the tme that rocessor erforms ts t, w of the total useful work W and overhead the tme not overlaed wth the comutaton of the overhead of tye. o study the overall mact of the overhead, we can rewrte Eq. (2) as =. ( 4) We wll develo the overhead equatons for homogeneous rocessors (wth equal comutng ower) whch mles that seq = com. ( 5) ogether wth Eqs. (3) and (4), the seedu can then be rewrtten as: S = ( 6), overhead 1+ seq he mact of the overhead on the seedu s thus reflected by ts rato wth the sequental runtme. Let us therefore defne ths rato for each overhead tye : overheads Ovh =. ( 7) seq hese ratos quantfy the cost of the overheads. Note that overhead s the summaton of that tye of overhead over all rocessors. Heterogenety Let us have a look how system heterogenety affects the erformance. For dfferent comutng owers, we ntroduce the relatve seed, measured wth resect to a reference machne: ( W ) com = ( 8) com( W ) Unlke most works on heterogeneous allel systems [19], we exress the seeds relatvely to a fxed machne [3], and not relatvely to the fastest rocessor. We thnk that choosng a fxed reference, wth whch the sequental runtme seq s measured, allows a clearer erformance analyss. Equaton (5) s based on the assumton that the runtme does not deend on a tcular orton of the work. Smlarly, we assume that the varous rocessors essentally dffer n ther clock seed [3] and not n the sze of the task. he total relatve rocessng ower of the allel system s then:.

3 PP = ( 9) For a heterogeneous system, the effcency s the erformance comed wth the deal erformance, namely PP: S E = ( 10) PP Wth the average relatve seed PP = we defne the degree of heterogenety H of a allel system by the standard devaton of [3]: 2 H = ( ) ( 11) o contnue the erformance analyss of the revous agrah, Eq. (5) should be relaced by s. com ( 12) = Also for the calculaton of the overhead rato Ovh (Eq. 7),, overhead should be scaled wth,. overhead ( 13) Ovh = seq hus, we can conclude that all tme measurements on a rocessor should be scaled wth. Slower rocessors wll carry out less useful work n a gven tme and the allel overheads wll have less mact on the seedu. hs result corresonds wth the aroach roosed n [4], whch essentally uses rocessor cycles as tme unt. Granularty In load-balancng roblems, the overheads are manly communcaton and dle tme, that we call blockng overhead. In our settng, the learnng algorthm can only mrove the erformance by mnmzng blockng, snce the communcaton tme reresents the lower bound. Moreover, the communcaton overhead s roortonal wth the total data sze of the work. For data-ntensve alcatons wth large messages, all other contrbutons to the communcaton tme, lke the latency or delay, can be neglected. he total work conssts of W quantums, where each quantum rocesses an amount of q data bytes, the communcaton tme becomes comm = β. qdata. W ( 14) where β s the tme to transmt a byte from the master to any slave. We assume t to be constant, as we do not consder heterogeneous communcaton networks. In the same way, the comutaton tme can be exressed as a functon of the work W by a frst order equaton. In ths equaton the lnear term s the man contrbuton, esecally for lnear data-ntensve alcatons that we consder. he comutaton tme becomes com = τ. qo. W ( 15) where q o s the number of oeratons er quantum work and τ reresents the atomc comutng tme er oeraton' on the reference machne. However, a ece of code cannot be dvded n equal oeratons, τ should be consdered a reference comutng tme. It s the roduct q o..τ, the length of a run-tme quantum, that we wll use and that we consdered to be constant. We then defne the granularty as a relatve measure of the amount of comutaton wth resect to the amount of communcaton wthn a allel algorthm mlementaton [14]: 1 com o gran = = =. ( 16) comm Ovhcomm β qdata It deends on hardware and software, so τ s β called the hardware granularty and q o the q data software granularty. he erformance s affected by the overall granularty, ndeendent of how the granularty s sread over software and hardware. We defne gran as the granularty measured on the reference machne and gran the granularty of rocessor. Snce the communcaton tme s fxed, the granularty of each rocessor s gran = gran. ( 17) Load balancng equlbrum (2a) (2b) Fg 2: Load balancng wth a clear comutaton bottleneck. Executon rofle(a) and overhead dstrbuton (b) τ q

4 Fg 3: Load balancng wth a clear communcaton bottleneck. Executon rofle(a) and overhead dstrbuton (b) We wll nvestgate the non-trval case of load balancng where the total comutaton ower of the slaves matches the communcaton bandwdth of the master for a gven total granularty. In other cases, there wll be ether a clear communcaton or comutaton bottleneck, what makes adatve load balancng unnecessary. Indeed, wth a total comutaton ower that s lower than the master s communcaton bandwdth, there s no bottleneck, the master wll be able to serve the slaves constantly and these wll work at 100% effcency. hs can be seen n the exerment of fgure 2, where the total comutaton ower s lower than the communcaton bandwdth. In that case, more rocessors can be added to ncrease the seedu. On the other hand, when the total comutaton ower ncreases, the master wll get requests at a hgher rate and the communcaton becomes the bottleneck. hen, the slaves wll start to block, watng to be served and ther effcency dros. hs can be seen n fgure 3, where the total comutaton ower s hgher than the communcaton bandwdth. In that case, the surlus of slave rocessors should better be used for other work. In both cases, the dstrbuton of the work can hardly be otmzed. We wll nvestgate the non-trval case, when comutaton and communcaton seed match. In that case, workload assgnment and synchronzaton becomes necessary. At the equlbrum ont, an deal load dstrbuton exsts, so that the master can serve the slaves constantly and these are fully busy. here wll be no blockng on master or slaves, so ther allel runtme s 1 = BW. ( 18) = + = Bw. + gran. Bw. ( > 1) comm com ( 19) Where gran s the granularty of rocessor. Snce all allel runtmes are equal (Eq. (2)), Eq. (18) equals Eq. (19) and w can be calculated for each slave w = BW. W gran ( 20) B.(1 + gran ) 1+ We know that W s the summaton of all w, hence = W 1+ gran w = = W ( 21) hs results n the load balancng equlbrum condton, whch ndcates a match between the systems communcaton bandwdth and rocessng ower 1 1+ gran = 1 ( 22) We ntroduce the load balancng equlbrum factor lbe to quantfy ths match 1 lbe = 1 + gran ( 23) Where lbe=1 reflects an equlbrum, lbe < 1 s for lower rocessng ower (Fg 2, where lbe=0.5) and lbe > 1 ndcates a communcaton bottleneck (Fg 3, where lbe=2) For a homogeneous system, gran s constant for all rocessors, and accordng to Eq.(22), the equlbrum condton becomes = gran + 2 and the seedu s then S = 2. Note that should be greater than 2, we need at least 2 slaves, so that one can comute when the other communcates. he load balancng equlbrum factor s then 1 lbe = 1 + gran ( 24) 4. MULI AGEN REINFORCEMEN LEARNING Renforcement learnng (RL) s the roblem faced by an agent that learns behavor through tral-and-error nteractons wth a dynamc envronment. A model of renforcement learnng conssts of a dscrete set of envronment states, a dscrete set of agent actons and a set of scalar renforcement sgnals. For each nteracton the agent receves renforcement and the next state of the envronment, and chooses an acton. he agent's ob s to fnd a olcy,.e. a mang from states to actons, whch maxmzes some long-turn measure of renforcement. hese rewards can take lace arbtrarly far n the future. o obtan a hgh overall reward, an agent has to refer actons that t has learned n the ast and found to be good,.e. exlotaton, however dscoverng such actons s only ossble by tryng out alternatve actons,.e. exloraton. Nether exlotaton, nor exloraton can be ursued exclusvely. In our load-balancng roblem settng rocessors are of the recever-ntated tye and can thus be vewed as agents, whch we wll gve extra learnng abltes. Each rocessor wll be an ndeendent RL agent, whch tres to learn an otmal chunk sze of data to ask the master, so that the blockng tme for others s mnmzed and therefore also the total comutaton tme. he agents actons are the ossble amounts of data or block szes avalable. Because we want to restrct ourselves to renforcement learners wth a dscrete set of actons, we secfy at the begnnng a set of ossble block szes the agents can choose

5 from. Snce there are multle slaves exstng together, whch nfluence each other, we have to use a mult-agent renforcement learnng scheme. In our alcaton, the slaves can only be n one state and they share a common goal. As such ths mult-agent roblem can be vewed as a coordnaton game, n whch the agents should learn to converge to the Pareto Otmal Nash equlbrum 1 of the game. In [17] an algorthm for learnng n a mult-agent coordnaton game s gven. We descrbe t n short n the next subsecton. Exloraton n a mult-agent coordnaton game he learnng scheme starts wth a number of exloraton erods and s followed by an exlotaton hase. At the begnnng of each exloraton erod, agents behave selfsh and nave;.e. they gnore the other agents n the envronment and use some tradtonal renforcement learnng technque to maxmze ther ayoff. he exloraton erod ends when the agents have found a Nash equlbrum. hs wll haen because t s assured by results from Learnng Automata theory [8] and also holds for other renforcement learnng algorthms such as Q-learnng [7][15]. Whch Nash equlbrum the agents fnd s not known n advance, t deends on the ntal condtons and the basn of attracton of the dfferent equlbrums. he goal s to converge to the best Nash equlbrum,.e. the Pareto Otmal equlbrum. After havng converged, every agent wll exclude ts last acton layed, so that the ont acton sace becomes consderable smaller. In a new erod of lay, a new Nash equlbrum wll be found. Before a new erod starts, the agent kees some statstcs for the acton t has converged too. When the agent has no actons left, ts orgnal acton sace s restored. New exloraton erods are layed untl the user decdes to sto the exloraton hase and decdes to exlot the best Nash equlbrum found. hs s acheved wthout any communcaton between the agents, because they sense the same ayoff for each ont acton layed. he ont acton they choose after exloraton s the best they have seen so far. Under some assumtons concernng erod-length and number of erods layed, convergence to the Pareto Otmal equlbrum s assured, even for stochastc rewards/ayoffs see [17]. In the next subsecton we exlan how to use ths mult-agent learnng algorthm n the load balancng exerment, whch can be vewed as an asynchronous verson 2 of a coordnaton game. 5. EXPERIMENAL SEUP 1 An outcome of a game s sad to be Pareto Otmal f there exsts no other outcome, for whch all the layers smultaneously do better. A Nash equlbrum of a game s a soluton for whch no agent can do better by changng hs olcy, as all the other kee layng ther Nash equlbrum olcy [10]. 2 Asynchronous n the sense that layers do not take ther actons synchronously, rewards may be delayed etc. o assess the resented algorthm for coarse gran data ntensve alcatons on heterogeneous allel hardware a synthetcs aroach has been used. An alcaton has been wrtten usng the PVM [5] message-assng lbrary to exerment wth the dfferent dmensons of the roblem. he alcaton has been desgned not to erform any real comutaton, but nstead t relaces the comutaton and communcaton hases by delays wth equvalent duraton 3. For the exermental setu, H and lbe are 2 exerment ameters. H and lbe are set to 1 n all exerments. he granularty of the rocessors wll then be chosen randomly accordng to a normal 1 dstrbuton wth 1, derved average = cce from the homogeneous case granularty, Eq. (24), and stddev = H. hs guarantees the exerment to be n an nterestng load balancng equlbrum. Learnng to request data In our load balancng settng, the slaves use a reward-nacton learnng automata scheme for selfsh lay durng the exloraton erods [8]. A learnng automaton descrbes the nternal state of an agent as a robablty dstrbuton accordng to whch actons should be chosen. hese robabltes are adusted accordng to the success or falure of the actons taken. he udate scheme we wll use here s the reward-nacton udate scheme, gven n equaton (25). A constant reward ameter a between 0 and 1 s used to renforce good actons. he feedback r gves the renforcement from the envronment. For ths scheme t s roven that layers wll converge to a Nash equlbrum when t s used n a one-state coordnaton game, see [8]. ( n + 1) = ( n) + a. r( n).(1 ( n)) f acton was chosen at tme n ( n + 1) = ( n) a. r( n). ( n) for all actons dfferent from ( 25) he feedback r rovded to the learner s the nverse blockng tme, whch s the tme a slave has to wat before the master acknowledges ts request for data. It s used to udate the acton robabltes by Eq.(25). Less nterestng actons wll ncur hgher blockng tmes and thus wll have lower robablty assocated wth them. In our exerment, the slaves wll learn to request a chunk sze that mnmzes ts blockng tme. o that end each slave has a stochastc learnng automaton whch uses Eq.(25), as shown n Fg 4. In the resented results the block sze s a multle of a gven ntal block sze. Here the multles are 1, 2 and 3 tmes the ntal block sze 3 he exermental alcaton can easly be turned nto a real alcaton by relacng the delay roducng code wth real code.

6 Chunk sze S 1 S 2 S 3 P P 1 P 2 P 3 random choce of S ~ P feedback: 1/blocktme request S blockng comm & com of S Fg 4: Renforcement learnng durng an exloraton erod. At the end of an exloraton erod the slaves have converged to a Nash equlbrum,.e. each slave has learnt a block sze to ask the master. hey send to the master ther average erformance durng ths erod, whch s the average amount of data each was able to comute. When the master has receved ths message from each slave, the overall reward, s sent back to each slave, whch uses ths to udate ts statstcs for the acton last layed. he overall reward the agents receve for the ont acton layed s gven by the total average amount of data the agents comuted n allel durng the erod of lay. hs ayoff can easly be attached to the data ackages the master sends to ts slaves, no extra communcaton s generated. he same argument goes for the slave sendng erod nformaton to the master. he number of exloraton erods s chosen n advance. After all erods have been layed, the slaves chose the acton whch was rewarded best by the master. We wll come our adatve algorthm wth a statc load-balancng scheme where fxed amounts of data are requested. 6. EXPERIMENS Fgure 5 shows the tme course of a tycal exerment n ts exlotaton erod wth the comutaton, communcaton and blockng hases. (data sze = 10GB, communcaton seed = 10MB/s, #slaves=3, average chunk sze =1MB, H=1, lbe=1, the length of the exloraton erod = 100s and the number of erods = 10) maxmally exloted. Performance results As shown n table 1, we have run several exerments wth varyng ont acton saces (JAS). We use u to 7 slaves wth each 5 ossble actons, whch results n a ont acton sace wth 5 7 ossbltes. he 4 th column gves the absolute gan n total comutaton tme for the adatve mult-agent algorthm comed to the total comutaton tme for the farmng algorthm. he 5 th column gves the gan n total comutaton tme whch s maxmal ossble for the gven settng. he total comutaton tme can only decrease when the dle tme of the master s decreased. No gan s ossble on the total communcaton tme of the data. herefore the last column gves the absolute gan comed to the gan whch was maxmal ossble. In all exerments consderable mrovements were made. Agents are able to fnd settngs n whch the faster slave s blocked less than the others, see also Fg. 5. he adatve characterstc of the agents makes them able to take ther heterogenety nto account. In larger ont acton saces the gan dros, but by adustng the ameters of the learnng algorthm,.e. makng sure the agents can exlore more by enlargng the exloraton erods length and ncreasng the number of exloraton erods, erformance gets better agan n the last exerments. he data sze ncreased from 10 to 30 GB n larger ont acton saces, the erod length from 100 to 200 tme stes and the number of erods from 10 to 20. All the other settngs were taken as above. he fgures shown are averaged results for 10 runs of each exerment. Slaves Actons #JAS abs. gan max. gan rel. gan able 1: Average gan n total comutaton tme for adatve load balancng comed to statc load balancng (farmng). Fg 5: Executon rofle of the exlotaton hase of a tycal exerment wth 3 slaves We observe that the slaves have learned to dstrbute the requests ncely and hence use the lnk wth the master effcently. Slave 3, whch s the fastest rocessor wth lowest granularty, s served constantly and has no blockng tme. Slave 2 has some blockng tme. However, t s the slowest rocessor, so the rocessng ower of the system s able 2 shows the results for unform load balancng. In the unform verson, nstead of each agent learnng the best amount of chunk sze to ask out of a gven set of ossble szes, the agents wll now choose each ossble chunk sze wth a unform dstrbuton. Results show that unform lay s on average comable to statc load balancng. So, our learnng aroach erforms better than both.

7 Slaves Actons #JAS abs. gan max. gan rel. gan able 2: Average gan n total comutaton tme for unform load balancng comed to statc load balancng (farmng). Overhead Analyss able 3 gves an overvew of the gan n blockng overhead. he blockng tme reresents the cumulated blockng tme of all the slaves durng the allel run, the dle tme s the dle tme of the master. he numbers resented here are averaged over 10 ndeendent runs. Intally a allel run (usng ob farmng) wth 3 slaves whch can use 3 dfferent block szes resulted n an average of seconds of dle tme for the master. However, when learnng s used the dle tme of master reduces to seconds, a gan of 30.37%. he gan n blockng tme follows the same nterretaton. Slaves Actons Idle (gan) Blockng (gan) % 50.7% % 51.3% % 38.7% % 20.7% % 20.7% % -3.4% % 3.20% % 3.00% % 3.00% able 3: Average gan for adatve load balancng n dle tme of the master and total blockng tme of the slaves. As shown the task of reducng the blockng tme becomes harder as the number of slaves ncreases. hs s a drect result of the exonental ncrease n sze of the search sace. 7. CONCLUSIONS Comlex, heterogeneous system controlled for otmal use by mult-agent renforcement learnng s romsng. We mlemented renforcement learners for dstrbuted load balancng of data ntensve alcatons. ogether they use a novel technque of coordnated exloraton, whch was already tested and roven to convergence to the Pareto Otmal Nash equlbrum n stochastc coordnaton games from game theory. he master-slave setu we used for our load balancng exerments was vewed as an asynchronous coordnaton game. Results show that the mult-agent algorthm used stll works for asynchronous games. he frst erformance results show consderable mrovements uon a statc load balancer. he agents are able to fnd settngs n whch the faster slave s less blocked durng communcaton wth the master than the others. he adatve characterstc of the agents makes them able to take ther heterogenety nto account. Even more when a slave s removed from the set-u, the remanng agents can adust themselves tot the new stuaton. he algorthm works locally on the slaves (recever-ntated), thus actng lke ntellgent agents. he agents use both local and global nformaton n ther learnng rocess, however no extra communcaton s needed n-between the slaves to acheve ths global nformaton. Every agent communcates ts erod-erformances to the master, whch n turn wll send the global nformaton back to every slave. hs extra nformaton can be attached to the global data ackages and requests, wthout causng extra overload. We are lannng to use ths mult-agent aroach for data ntensve alcatons such as comresson alcatons. 8. REFERENCES [1] I. Bancescu and V. Velusamy, Load Balancng Hghly Irregular Comutatons wth the Adatve Factorng, Proceedngs of the 16 th Internatonal Parallel & Dstrbuted Processng Symosum, IEEE, Los Alamtos, Calforna, [2] M. D. Beynon,. Kurc et al. Effcent Manulaton of Large Datasets on Heterogeneous Storage Systems, Proceedngs of the 16 th Internatonal Parallel & Dstrbuted Processng Symosum, IEEE, Los Alamtos, Calforna, [3] Clemats, A. and Corana, A., Modelng erformance of heterogeneous allel comutng systems, In Parallel Comutng 25, Elsever Scence, [4] Crovella, M. E. and Leblanc,.J., Parallel Performance Predcton usng Lost Cycles Analyss, n Proc. of Suercomutng 94, IEEE Comuter Socety, [5] A. Gest, A. Begueln et al., PVM: Parallel Vrtual Machne, the MI ress, [6] D. Guta and P. Be, "Load sharng n dstrbuted systems", In Proceedngs of the Natonal Worksho on Dstrbuted Comutng, January [7] Kaelblng L.P., Ltmann M.L., Moore A.W.,: Renforcement Learnng: A Survey. Journal of Artfcal Intellgence Research 4 (1996) [8] Narendra K., hathachar M., Learnng Automata: An Introducton, Prentce-Hall (1989). [9] Nowé, A., Verbeeck, K., Dstrbuted Renforcement learnng, Loadbased Routng a case study, Proceedngs of the Neural, Symbolc and Renforcement Methods for sequence Learnng

8 Worksho at ca99, [10] Osborne J.O.,Rubnsten A., A course n game theory, Cambrdge, MA: MI Press (1994). [11] C.C. Prce and S. Krshnarasad, Software allocaton models for dstrbuted systems, n Proceedngs of the 5 th Internatonal Conference on Dstrbuted Comutng, ages 40-47, [12] Schaerf A., Shoham Y., ennenholtz M., Adatve Load Balancng: A Study n Mult-Agent Learnng, Journal of Artfcal Intellgence Research (1995) [13]. Schnekenburger and G. Rackl, Imlementng Dynamc Load Dstrbuton Strateges wth Orbx, Internatonal Conference on Parallel and Dstrbuted Processng echnques and Alcatons (PDPA'97), Las Vegas, Nevada, [14] Stone H.S., Hgh-Performance Comuter Archtecture, Addson-Wesley, Massachusetts, [15] Sutton, R.S., Barto, A.G., Renforcement Learnng: An ntroducton, Cambrdge, MA: MI Press (1998). [16] Verbeeck, K., Nowé, A., Lenaerts., Parent, J., Learnng to reach the Pareto Otmal Nash Equlbrum as a eam, LNAI Proceedngs of the 15th Australan Jont Conference on Artfcal Intellgence. P (2002). [17] Verbeeck, K., Nowé, uyls, K., Coordnated Exloraton n Stochastc Common Interest Games, hrd symosum on Adatve Agents and Mult-agents Systems, AAMAS-3. (2003) [18] M.J. Zak, We L; S. Parthasarathy, Customzed dynamc load balancng for a network of workstatons, Proceedngs of the Hgh Performance Dstrbuted Comutng (HPDC'96), IEEE, [19] X. Zhang, Y. Yan, Modelng and characterzng allel comutng erformance on heterogeneous networks of workstatons, n Proc. of the 7 th IEEE Symosum on Par. And Dstr. Proc., IEEE, 1995.

Dynamic Load Balancing of Parallel Computational Iterative Routines on Platforms with Memory Heterogeneity

Dynamic Load Balancing of Parallel Computational Iterative Routines on Platforms with Memory Heterogeneity Dynamc Load Balancng of Parallel Comutatonal Iteratve Routnes on Platforms wth Memory Heterogenety Davd Clare, Alexey Lastovetsy, Vladmr Rychov School of Comuter Scence and Informatcs, Unversty College

More information

Load Balancing of Parallelized Information Filters

Load Balancing of Parallelized Information Filters IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING, VOL. XXX, NO. XX, XXXXXXX 2001 1 Load Balancng of Parallelzed Informaton Flters Nel C. Rowe, Member, IEEE Comuter Socety, and Amr Zaky, Member, IEEE

More information

A Study on Secure Data Storage Strategy in Cloud Computing

A Study on Secure Data Storage Strategy in Cloud Computing Journal of Convergence Informaton Technology Volume 5, Number 7, Setember 00 A Study on Secure Data Storage Strategy n Cloud Comutng Danwe Chen, Yanjun He, Frst Author College of Comuter Technology, Nanjng

More information

Optimal maintenance of a production-inventory system with continuous repair times and idle periods

Optimal maintenance of a production-inventory system with continuous repair times and idle periods Proceedngs o the 3 Internatonal Conerence on Aled Mathematcs and Comutatonal Methods Otmal mantenance o a roducton-nventory system wth contnuous rear tmes and dle erods T. D. Dmtrakos* Deartment o Mathematcs

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

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

Chapter 3: Dual-bandwidth Data Path and BOCP Design

Chapter 3: Dual-bandwidth Data Path and BOCP Design Chater 3: Dual-bandwdth Data Path and BOCP Desgn 3. Introducton The focus of ths thess s on the 4G wreless moble Internet networks to rovde data servces wthn the overlang areas of CDA2000-WLA networks.

More information

A NEW ACTIVE QUEUE MANAGEMENT ALGORITHM BASED ON NEURAL NETWORKS PI. M. Yaghoubi Waskasi MYaghoubi@ece.ut.ac.ir. M. J. Yazdanpanah Yazdan@ut.ac.

A NEW ACTIVE QUEUE MANAGEMENT ALGORITHM BASED ON NEURAL NETWORKS PI. M. Yaghoubi Waskasi MYaghoubi@ece.ut.ac.ir. M. J. Yazdanpanah Yazdan@ut.ac. A NEW ACTIVE QUEUE MANAGEMENT ALGORITHM BASED ON NEURAL NETWORKS M. Yaghoub Waskas MYaghoub@ece.ut.ac.r M. J. Yazdananah Yazdan@ut.ac.r N. Yazdan Yazdan@ut.ac.r Control and Intellgent Processng Center

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

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

Feasibility of Using Discriminate Pricing Schemes for Energy Trading in Smart Grid

Feasibility of Using Discriminate Pricing Schemes for Energy Trading in Smart Grid Feasblty of Usng Dscrmnate Prcng Schemes for Energy Tradng n Smart Grd Wayes Tushar, Chau Yuen, Bo Cha, Davd B. Smth, and H. Vncent Poor Sngapore Unversty of Technology and Desgn, Sngapore 138682. Emal:

More information

An Analytical Model for Multi-tier Internet Services and Its Applications

An Analytical Model for Multi-tier Internet Services and Its Applications An Analytcal Model for Mult-ter Internet Servces and Its Alcatons Bhuvan Urgaonkar, Govann Pacfc, Prashant Shenoy, Mke Sretzer, and Asser Tantaw Det. of Comuter Scence, Servce Management Mddleware Det.,

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

Evaluation of the information servicing in a distributed learning environment by using monitoring and stochastic modeling

Evaluation of the information servicing in a distributed learning environment by using monitoring and stochastic modeling MultCraft Internatonal Journal of Engneerng, Scence and Technology Vol, o, 9, -4 ITERATIOAL JOURAL OF EGIEERIG, SCIECE AD TECHOLOGY wwwest-ngcom 9 MultCraft Lmted All rghts reserved Evaluaton of the nformaton

More information

Conferencing protocols and Petri net analysis

Conferencing protocols and Petri net analysis Conferencng protocols and Petr net analyss E. ANTONIDAKIS Department of Electroncs, Technologcal Educatonal Insttute of Crete, GREECE ena@chana.tecrete.gr Abstract: Durng a computer conference, users desre

More information

A Comprehensive Analysis of Bandwidth Request Mechanisms in IEEE 802.16 Networks

A Comprehensive Analysis of Bandwidth Request Mechanisms in IEEE 802.16 Networks A Comrehensve Analyss of Bandwdth Reuest Mechansms n IEEE 802.6 Networks Davd Chuck, Kuan-Yu Chen and J. Morrs Chang Deartment of Electrcal and Comuter Engneerng Iowa State Unversty, Ames, Iowa 500, USA

More information

Applied Research Laboratory. Decision Theory and Receiver Design

Applied Research Laboratory. Decision Theory and Receiver Design Decson Theor and Recever Desgn Sgnal Detecton and Performance Estmaton Sgnal Processor Decde Sgnal s resent or Sgnal s not resent Nose Nose Sgnal? Problem: How should receved sgnals be rocessed n order

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

Project Networks With Mixed-Time Constraints

Project Networks With Mixed-Time Constraints Project Networs Wth Mxed-Tme Constrants L Caccetta and B Wattananon Western Australan Centre of Excellence n Industral Optmsaton (WACEIO) Curtn Unversty of Technology GPO Box U1987 Perth Western Australa

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

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

A Prediction System Based on Fuzzy Logic

A Prediction System Based on Fuzzy Logic Proceedngs of the World Congress on Engneerng and Comuter Scence 2008 WCECS 2008, October 22-24, 2008, San Francsco, USA A Predcton System Based on Fuzzy Logc Vadeh.V,Monca.S, Mohamed Shek Safeer.S, Deeka.M

More information

Portfolio Loss Distribution

Portfolio Loss Distribution Portfolo Loss Dstrbuton Rsky assets n loan ortfolo hghly llqud assets hold-to-maturty n the bank s balance sheet Outstandngs The orton of the bank asset that has already been extended to borrowers. Commtment

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

Efficient Computation of Optimal, Physically Valid Motion

Efficient Computation of Optimal, Physically Valid Motion Vol. xx No. xx,.1 5, 200x 1 Effcent Comutaton of Otmal, Physcally Vald Moton Anthony C. Fang 1 and Nancy S. Pollard 2 1 Deartment of Comuter Scence, Natonal Unversty of Sngaore 2 Robotcs Insttute, Carnege

More information

Dynamic Fleet Management for Cybercars

Dynamic Fleet Management for Cybercars Proceedngs of the IEEE ITSC 2006 2006 IEEE Intellgent Transportaton Systems Conference Toronto, Canada, September 17-20, 2006 TC7.5 Dynamc Fleet Management for Cybercars Fenghu. Wang, Mng. Yang, Ruqng.

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

PERFORMANCE ANALYSIS OF PARALLEL ALGORITHMS

PERFORMANCE ANALYSIS OF PARALLEL ALGORITHMS Software Analye PERFORMANCE ANALYSIS OF PARALLEL ALGORIHMS Felcan ALECU PhD, Unverty Lecturer, Economc Informatc Deartment, Academy of Economc Stude, Bucharet, Romana E-mal: alecu.felcan@e.ae.ro Abtract:

More information

NON-CONSTANT SUM RED-AND-BLACK GAMES WITH BET-DEPENDENT WIN PROBABILITY FUNCTION LAURA PONTIGGIA, University of the Sciences in Philadelphia

NON-CONSTANT SUM RED-AND-BLACK GAMES WITH BET-DEPENDENT WIN PROBABILITY FUNCTION LAURA PONTIGGIA, University of the Sciences in Philadelphia To appear n Journal o Appled Probablty June 2007 O-COSTAT SUM RED-AD-BLACK GAMES WITH BET-DEPEDET WI PROBABILITY FUCTIO LAURA POTIGGIA, Unversty o the Scences n Phladelpha Abstract In ths paper we nvestgate

More information

A New Technique for Vehicle Tracking on the Assumption of Stratospheric Platforms. Department of Civil Engineering, University of Tokyo **

A New Technique for Vehicle Tracking on the Assumption of Stratospheric Platforms. Department of Civil Engineering, University of Tokyo ** Fuse, Taash A New Technque for Vehcle Tracng on the Assumton of Stratosherc Platforms Taash FUSE * and Ehan SHIMIZU ** * Deartment of Cvl Engneerng, Unversty of Toyo ** Professor, Deartment of Cvl Engneerng,

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

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

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

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

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

A Ratio-Based Control Algorithm for Defense of DDoS Attacks

A Ratio-Based Control Algorithm for Defense of DDoS Attacks A Rato-Based ontrol Algorthm for Defense of DDoS Attacks Sheng-Ya n Yong Xong Jyh-harn u Deartment of omuter Scence eas A&M Unversty ollege Staton X 7784- {shengyayonglu}@cs.tamu.edu echncal Reort 005--4

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 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

How Sets of Coherent Probabilities May Serve as Models for Degrees of Incoherence

How Sets of Coherent Probabilities May Serve as Models for Degrees of Incoherence 1 st Internatonal Symposum on Imprecse Probabltes and Ther Applcatons, Ghent, Belgum, 29 June 2 July 1999 How Sets of Coherent Probabltes May Serve as Models for Degrees of Incoherence Mar J. Schervsh

More information

Institute of Informatics, Faculty of Business and Management, Brno University of Technology,Czech Republic

Institute of Informatics, Faculty of Business and Management, Brno University of Technology,Czech Republic Lagrange Multplers as Quanttatve Indcators n Economcs Ivan Mezník Insttute of Informatcs, Faculty of Busness and Management, Brno Unversty of TechnologCzech Republc Abstract The quanttatve role of Lagrange

More information

Vision Mouse. Saurabh Sarkar a* University of Cincinnati, Cincinnati, USA ABSTRACT 1. INTRODUCTION

Vision Mouse. Saurabh Sarkar a* University of Cincinnati, Cincinnati, USA ABSTRACT 1. INTRODUCTION Vson Mouse Saurabh Sarkar a* a Unversty of Cncnnat, Cncnnat, USA ABSTRACT The report dscusses a vson based approach towards trackng of eyes and fngers. The report descrbes the process of locatng the possble

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

A Structure Preserving Database Encryption Scheme

A Structure Preserving Database Encryption Scheme A Structure Preservng Database Encryton Scheme Yuval Elovc, Ronen Wasenberg, Erez Shmuel, Ehud Gudes Ben-Guron Unversty of the Negev, Faculty of Engneerng, Deartment of Informaton Systems Engneerng, Postfach

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

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

Distributed Multi-Target Tracking In A Self-Configuring Camera Network

Distributed Multi-Target Tracking In A Self-Configuring Camera Network Dstrbuted Mult-Target Trackng In A Self-Confgurng Camera Network Crstan Soto, B Song, Amt K. Roy-Chowdhury Department of Electrcal Engneerng Unversty of Calforna, Rversde {cwlder,bsong,amtrc}@ee.ucr.edu

More information

An Enhanced Super-Resolution System with Improved Image Registration, Automatic Image Selection, and Image Enhancement

An Enhanced Super-Resolution System with Improved Image Registration, Automatic Image Selection, and Image Enhancement An Enhanced Super-Resoluton System wth Improved Image Regstraton, Automatc Image Selecton, and Image Enhancement Yu-Chuan Kuo ( ), Chen-Yu Chen ( ), and Chou-Shann Fuh ( ) Department of Computer Scence

More information

Rate Monotonic (RM) Disadvantages of cyclic. TDDB47 Real Time Systems. Lecture 2: RM & EDF. Priority-based scheduling. States of a process

Rate Monotonic (RM) Disadvantages of cyclic. TDDB47 Real Time Systems. Lecture 2: RM & EDF. Priority-based scheduling. States of a process Dsadvantages of cyclc TDDB47 Real Tme Systems Manual scheduler constructon Cannot deal wth any runtme changes What happens f we add a task to the set? Real-Tme Systems Laboratory Department of Computer

More information

Chapter 4 ECONOMIC DISPATCH AND UNIT COMMITMENT

Chapter 4 ECONOMIC DISPATCH AND UNIT COMMITMENT Chapter 4 ECOOMIC DISATCH AD UIT COMMITMET ITRODUCTIO A power system has several power plants. Each power plant has several generatng unts. At any pont of tme, the total load n the system s met by the

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

A Secure Password-Authenticated Key Agreement Using Smart Cards

A Secure Password-Authenticated Key Agreement Using Smart Cards A Secure Password-Authentcated Key Agreement Usng Smart Cards Ka Chan 1, Wen-Chung Kuo 2 and Jn-Chou Cheng 3 1 Department of Computer and Informaton Scence, R.O.C. Mltary Academy, Kaohsung 83059, Tawan,

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

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

A Lyapunov Optimization Approach to Repeated Stochastic Games

A Lyapunov Optimization Approach to Repeated Stochastic Games PROC. ALLERTON CONFERENCE ON COMMUNICATION, CONTROL, AND COMPUTING, OCT. 2013 1 A Lyapunov Optmzaton Approach to Repeated Stochastc Games Mchael J. Neely Unversty of Southern Calforna http://www-bcf.usc.edu/

More information

Research Article Competition and Integration in Closed-Loop Supply Chain Network with Variational Inequality

Research Article Competition and Integration in Closed-Loop Supply Chain Network with Variational Inequality Hndaw Publshng Cororaton Mathematcal Problems n Engneerng Volume 2012, Artcle ID 524809, 21 ages do:10.1155/2012/524809 Research Artcle Cometton and Integraton n Closed-Loo Suly Chan Network wth Varatonal

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

Analysis and Modeling of Buck Converter in Discontinuous-Output-Inductor-Current Mode Operation *

Analysis and Modeling of Buck Converter in Discontinuous-Output-Inductor-Current Mode Operation * Energy and Power Engneerng, 3, 5, 85-856 do:.436/ee.3.54b63 Publshed Onlne July 3 (htt://www.scr.org/journal/ee) Analyss and Modelng of Buck Converter n Dscontnuous-Outut-Inductor-Current Mode Oeraton

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

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

Financial Mathemetics

Financial Mathemetics Fnancal Mathemetcs 15 Mathematcs Grade 12 Teacher Gude Fnancal Maths Seres Overvew In ths seres we am to show how Mathematcs can be used to support personal fnancal decsons. In ths seres we jon Tebogo,

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

Learning User's Scheduling Criteria in a Personal Calendar Agent!

Learning User's Scheduling Criteria in a Personal Calendar Agent! Learnng User's Schedulng Crtera n a Personal Calendar Agent! Shh-ju Ln and Jane Yung-jen Hsu Deartment of Comuter Scence and Informaton Engneerng Natonal Tawan Unversty 1 Sec 4 Roosevelt Road, Tae, 106

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

On the Interaction between Load Balancing and Speed Scaling

On the Interaction between Load Balancing and Speed Scaling On the Interacton between Load Balancng and Speed Scalng Ljun Chen and Na L Abstract Speed scalng has been wdely adopted n computer and communcaton systems, n partcular, to reduce energy consumpton. An

More information

Load Balancng n the L Norm Baruch Awerbuch æ Yoss Azar y Edward F. Grove z Det. of Comuter Scence Det. of Comuter Scence Det. of Comuter Scence Johns Hokns Unversty Tel-Avv Unversty Duke Unversty Baltmore,

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

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

Efficient Bandwidth Management in Broadband Wireless Access Systems Using CAC-based Dynamic Pricing

Efficient Bandwidth Management in Broadband Wireless Access Systems Using CAC-based Dynamic Pricing Effcent Bandwdth Management n Broadband Wreless Access Systems Usng CAC-based Dynamc Prcng Bader Al-Manthar, Ndal Nasser 2, Najah Abu Al 3, Hossam Hassanen Telecommuncatons Research Laboratory School of

More information

Simple Interest Loans (Section 5.1) :

Simple Interest Loans (Section 5.1) : Chapter 5 Fnance The frst part of ths revew wll explan the dfferent nterest and nvestment equatons you learned n secton 5.1 through 5.4 of your textbook and go through several examples. The second part

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

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

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

Application of Multi-Agents for Fault Detection and Reconfiguration of Power Distribution Systems

Application of Multi-Agents for Fault Detection and Reconfiguration of Power Distribution Systems 1 Applcaton of Mult-Agents for Fault Detecton and Reconfguraton of Power Dstrbuton Systems K. Nareshkumar, Member, IEEE, M. A. Choudhry, Senor Member, IEEE, J. La, A. Felach, Senor Member, IEEE Abstract--The

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

On the Interaction between Load Balancing and Speed Scaling

On the Interaction between Load Balancing and Speed Scaling On the Interacton between Load Balancng and Speed Scalng Ljun Chen, Na L and Steven H. Low Engneerng & Appled Scence Dvson, Calforna Insttute of Technology, USA Abstract Speed scalng has been wdely adopted

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

Energy Conserving Routing in Wireless Ad-hoc Networks

Energy Conserving Routing in Wireless Ad-hoc Networks Energy Conservng Routng n Wreless Ad-hoc Networks Jae-Hwan Chang and Leandros Tassulas Department of Electrcal and Computer Engneerng & Insttute for Systems Research Unversty of Maryland at College ark

More information

AN APPOINTMENT ORDER OUTPATIENT SCHEDULING SYSTEM THAT IMPROVES OUTPATIENT EXPERIENCE

AN APPOINTMENT ORDER OUTPATIENT SCHEDULING SYSTEM THAT IMPROVES OUTPATIENT EXPERIENCE AN APPOINTMENT ORDER OUTPATIENT SCHEDULING SYSTEM THAT IMPROVES OUTPATIENT EXPERIENCE Yu-L Huang Industral Engneerng Department New Mexco State Unversty Las Cruces, New Mexco 88003, U.S.A. Abstract Patent

More information

Recurrence. 1 Definitions and main statements

Recurrence. 1 Definitions and main statements Recurrence 1 Defntons and man statements Let X n, n = 0, 1, 2,... be a MC wth the state space S = (1, 2,...), transton probabltes p j = P {X n+1 = j X n = }, and the transton matrx P = (p j ),j S def.

More information

Dynamic Pricing for Smart Grid with Reinforcement Learning

Dynamic Pricing for Smart Grid with Reinforcement Learning Dynamc Prcng for Smart Grd wth Renforcement Learnng Byung-Gook Km, Yu Zhang, Mhaela van der Schaar, and Jang-Won Lee Samsung Electroncs, Suwon, Korea Department of Electrcal Engneerng, UCLA, Los Angeles,

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

Section 5.3 Annuities, Future Value, and Sinking Funds

Section 5.3 Annuities, Future Value, and Sinking Funds Secton 5.3 Annutes, Future Value, and Snkng Funds Ordnary Annutes A sequence of equal payments made at equal perods of tme s called an annuty. The tme between payments s the payment perod, and the tme

More information

Towards an Effective Personalized Information Filter for P2P Based Focused Web Crawling

Towards an Effective Personalized Information Filter for P2P Based Focused Web Crawling Journal of Comuter Scence (1): 97-103, 006 ISS 1549-3636 006 Scence Publcatons Towards an Effectve Personalzed Informaton Flter for PP Based Focused Web Crawlng Fu Xang-hua and Feng Bo-qn Deartment of

More information

INVENTORY MANAGEMENT REVISED

INVENTORY MANAGEMENT REVISED Scence & Mltary 2/2011 INVENTORY MANAGEMENT REVISED Analyss of behavoral asects of decson makng wthn Sales & Oeratons Plannng rocess Peter JUREČKA Abstract: The urose of ths artcle s to extend the standard

More information

VRT012 User s guide V0.1. Address: Žirmūnų g. 27, Vilnius LT-09105, Phone: (370-5) 2127472, Fax: (370-5) 276 1380, Email: info@teltonika.

VRT012 User s guide V0.1. Address: Žirmūnų g. 27, Vilnius LT-09105, Phone: (370-5) 2127472, Fax: (370-5) 276 1380, Email: info@teltonika. VRT012 User s gude V0.1 Thank you for purchasng our product. We hope ths user-frendly devce wll be helpful n realsng your deas and brngng comfort to your lfe. Please take few mnutes to read ths manual

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

A role based access in a hierarchical sensor network architecture to provide multilevel security

A role based access in a hierarchical sensor network architecture to provide multilevel security 1 A role based access n a herarchcal sensor network archtecture to provde multlevel securty Bswajt Panja a Sanjay Kumar Madra b and Bharat Bhargava c a Department of Computer Scenc Morehead State Unversty

More information

Master s Thesis. Configuring robust virtual wireless sensor networks for Internet of Things inspired by brain functional networks

Master s Thesis. Configuring robust virtual wireless sensor networks for Internet of Things inspired by brain functional networks Master s Thess Ttle Confgurng robust vrtual wreless sensor networks for Internet of Thngs nspred by bran functonal networks Supervsor Professor Masayuk Murata Author Shnya Toyonaga February 10th, 2014

More information

www.engineerspress.com Neural Network Solutions for Forward Kinematics Problem of Hybrid Serial-Parallel Manipulator

www.engineerspress.com Neural Network Solutions for Forward Kinematics Problem of Hybrid Serial-Parallel Manipulator www.engneersress.com World of Scences Journal ISSN: 307-307 Year: 03 Volume: Issue: 8 Pages: 48-58 Aahmad Ghanbar,, Arash ahman Deartment of Mechancal Engneerng, Unversty of Tabrz, Tabrz, Iran School of

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

Chosen Public Key and Ciphertext Secure Proxy Re-encryption Schemes

Chosen Public Key and Ciphertext Secure Proxy Re-encryption Schemes Internatonal Journal of Dgtal ontent Technology and ts Alcatons Volume 4 Number 9 December 00 hosen Publc Key and hertext Secure Proxy Re-encryton Schemes Lmng Fang Wlly Suslo Yongun Ren huneng Ge and

More information

JCM_VN_AM003_ver01.0 Sectoral scope: 03

JCM_VN_AM003_ver01.0 Sectoral scope: 03 Sectoral scoe: 03 Jont Credtng Mechansm Aroved Methodology VN_AM003 Imrovng the energy effcency of commercal buldngs by utlzaton of hgh effcency equment A. Ttle of the methodology Imrovng the energy effcency

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

2008/8. An integrated model for warehouse and inventory planning. Géraldine Strack and Yves Pochet

2008/8. An integrated model for warehouse and inventory planning. Géraldine Strack and Yves Pochet 2008/8 An ntegrated model for warehouse and nventory plannng Géraldne Strack and Yves Pochet CORE Voe du Roman Pays 34 B-1348 Louvan-la-Neuve, Belgum. Tel (32 10) 47 43 04 Fax (32 10) 47 43 01 E-mal: corestat-lbrary@uclouvan.be

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

行 政 院 國 家 科 學 委 員 會 補 助 專 題 研 究 計 畫 成 果 報 告 期 中 進 度 報 告

行 政 院 國 家 科 學 委 員 會 補 助 專 題 研 究 計 畫 成 果 報 告 期 中 進 度 報 告 行 政 院 國 家 科 學 委 員 會 補 助 專 題 研 究 計 畫 成 果 報 告 期 中 進 度 報 告 畫 類 別 : 個 別 型 計 畫 半 導 體 產 業 大 型 廠 房 之 設 施 規 劃 計 畫 編 號 :NSC 96-2628-E-009-026-MY3 執 行 期 間 : 2007 年 8 月 1 日 至 2010 年 7 月 31 日 計 畫 主 持 人 : 巫 木 誠 共 同

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 Article QoS and Energy Aware Cooperative Routing Protocol for Wildfire Monitoring Wireless Sensor Networks

Research Article QoS and Energy Aware Cooperative Routing Protocol for Wildfire Monitoring Wireless Sensor Networks The Scentfc World Journal Volume 3, Artcle ID 43796, pages http://dx.do.org/.55/3/43796 Research Artcle QoS and Energy Aware Cooperatve Routng Protocol for Wldfre Montorng Wreless Sensor Networks Mohamed

More information

An ILP Formulation for Task Mapping and Scheduling on Multi-core Architectures

An ILP Formulation for Task Mapping and Scheduling on Multi-core Architectures An ILP Formulaton for Task Mappng and Schedulng on Mult-core Archtectures Yng Y, We Han, Xn Zhao, Ahmet T. Erdogan and Tughrul Arslan Unversty of Ednburgh, The Kng's Buldngs, Mayfeld Road, Ednburgh, EH9

More information

A DATA MINING APPLICATION IN A STUDENT DATABASE

A DATA MINING APPLICATION IN A STUDENT DATABASE JOURNAL OF AERONAUTICS AND SPACE TECHNOLOGIES JULY 005 VOLUME NUMBER (53-57) A DATA MINING APPLICATION IN A STUDENT DATABASE Şenol Zafer ERDOĞAN Maltepe Ünversty Faculty of Engneerng Büyükbakkalköy-Istanbul

More information