Stochastic Models of Load Balancing and Scheduling in Cloud Computing Clusters

Size: px
Start display at page:

Download "Stochastic Models of Load Balancing and Scheduling in Cloud Computing Clusters"

Transcription

1 01 Proceedngs IEEE INFOCOM Stochastc Models of Load Balancng and Schedulng n Cloud Coputng Clusters Sva heja Magulur and R. Srkant Departent of ECE and CSL Unversty of Illnos at Urbana-Chapagn sva.theja@gal.co; rsrkant@llnos.edu Le Yng Departent of ECE Iowa State Unversty leyng@astate.edu Abstract Cloud coputng servces are becong ubqutous, and are startng to serve as the prary source of coputng power for both enterprses and personal coputng applcatons. We consder a stochastc odel of a cloud coputng cluster, where jobs arrve accordng to a stochastc process and request vrtual achnes (VMs, whch are specfed n ters of resources such as CPU, eory and storage space. Whle there are any desgn ssues assocated wth such systes, here we focus only on resource allocaton probles, such as the desgn of algorths for load balancng aong servers, and algorths for schedulng VM confguratons. Gven our odel of a cloud, we frst defne ts capacty,.e., the axu rates at whch jobs can be processed n such a syste. hen, we show that the wdely-used Best- Ft schedulng algorth s not throughput-optal, and present alternatves whch acheve any arbtrary fracton of the capacty regon of the cloud. We then study the delay perforance of these alternatve algorths through sulatons. I. INRODUCION Cloud coputng servces are becong the prary source of coputng power for both enterprses and personal coputng applcatons. A cloud coputng platfor can provde a varety of resources, ncludng nfrastructure, software, and servces, to users n an on-deand fashon. o access these resources, a cloud user subts a request for resources. he cloud provder then provdes the requested resources fro a coon resource pool (e.g., a cluster of servers, and allows the user to use these resources for a requred te perod. Copared to tradtonal own-and-use approaches, cloud coputng servces elnate the costs of purchasng and antanng the nfrastructures for cloud users, and allow the users to dynacally scale up and down coputng resources n real te based on ther needs. Several cloud coputng systes are now coercally avalable, ncludng Aazon EC syste [1], Google s AppEngne [], and Mcrosoft s Azure [3]. We refer to [4], [5], [6] for coprehensve surveys on cloud coputng. Whle cloud coputng servces n practce provde any dfferent servces, n ths paper, we consder cloud coputng platfors that provde nfrastructure as servce, n the for of Vrtual Machnes (VMs, to users. We assue cloud users request vrtual achnes (VMs, whch are specfed n ters of resources such as CPU, eory and storage space. Each request s called a job. he type of a job specfes the type of VM the user wants and the sze of the job specfes the aount of te requred. After recevng these requests, the cloud provder wll schedule the VMs on physcal achnes, called servers. here are any desgn ssues assocated wth such systes [7], [8], [9], [10], [11], [1]. In ths paper, we focus only on resource allocaton probles, such as the desgn of algorths for load balancng aong servers, and algorths for schedulng VM confguratons. We consder a stochastc odel of a cloud coputng cluster. We assue that jobs wth varable szes arrve accordng to a stochastc process, and are assgned to the servers accordng to a resource allocaton algorth. A job departs fro the syste after the VM s hosted for the requred aount of te. We assue jobs are queued n the syste when all servers are busy. We are nterested n the axu rates at whch jobs can be processed n such a syste, and resource allocaton algorths that can support the axu rates. he an contrbutons of ths paper are suarzed below. (1 We characterze the capacty regon of a cloud syste by establshng ts connecton to the capacty regon of a wreless network. he capacty of a cloud syste s defned to be the set of traffc loads under whch the queues n the syste can be stablzed. ( We then consder the wdely-used Best-Ft schedulng algorth and provde a sple exaple to show that t s not throughput-optal. Next, we pont out that the well-known MaxWeght algorth s throughput-optal n an deal scenaro, where jobs can be preepted and can grate aong servers, and servers can be reconfgured at each te nstant. In practce, preepton and VM graton are costly. herefore, otvated by the MaxWeght algorth, we present a non-preeptve algorth whch yopcally allocates a new job to a server usng current queue length nforaton whenever a departure occurs. We characterze the throughput of ths yopc algorth, and show that t can acheve any arbtrary fracton of the capacty regon f the algorth paraeters are chosen approprately. (3 he algorths entoned above requre central queues. In practce, a ore scalable approach s to route jobs to servers rght after ther arrvals. We consder the Jonthe-Shortest-Queue (JSQ algorth whch routes a job to the server wth the shortest queue. We prove that ths entals no loss n throughput copared to antanng a /1/$ IEEE 70

2 sngle central queue. (4 JSQ needs to keep track of queue lengths at all servers, whch ay becoe prohbtve when we have a large nuber of servers and the arrval rates of jobs are large. o address ths ssue, we propose the power-of-twochoces routng for the case of dentcal servers, and pck-and-copare routng for the case of non-dentcal servers. II. MODEL DESCRIPION A cloud syste conssts of a nuber of networked servers. Each of the servers ay host ultple Vrtual Machnes (VMs. Each VM requres a set of resources, ncludng CPU, eory, and storage space. VMs are classfed accordng to the resources they request. As an exaple, able I lsts three types of VMs (called nstances avalable n Aazon EC. Instance ype Meory CPU Storage Standard Extra Large 15 GB 8 EC unts 1,690 GB Hgh-Meory Extra Large 17.1 GB 6.5 EC unts 40 GB Hgh-CPU Extra Large 7 GB 0 EC unts 1,690 GB ABLE I HREE REPRESENAIVE INSANCES IN AMAZON EC We assue there are M dstnct VM confguratons and that each VM confguraton s specfed n ters of ts requreents for K dfferent resources. Let R k be the aount of type-k resource (e.g., eory requred by a type- VM (e.g., a standard extra large VM. Further, we assue that the cloud syste conssts of L dfferent servers. Let C k denote the aount of type-k resource at server. Gven a server, an M- densonal vector N s sad to be a feasble VM-confguraton f the gven server can sultaneously host N 1 type-1 VMs, N type- VMs,..., and N M type-m VMs. In other words, N s feasble at server f and only f M N R k C k =1 for all k. We let N ax denote the axu nuber of VMs of any type that can be served on any server. Exaple 1: Consder a server wth 30 GB eory, 30 EC coputng unts and 4, 000 GB storage space. hen N = (, 0, 0 and N = (0, 1, 1 are two feasble VMconfguratons on the server, where N 1 s the nuber of standard extra large VMs, N s the nuber of hgh-eory extra large VMs, and N 3 s the nuber of hgh-cpu extra large VMs. N = (0,, 1 s not a feasble VM confguraton on ths server because t does not have enough eory and coputng unts. In ths paper, we consder a cloud syste whch hosts VMs for clents. A VM request fro a clent specfes the type of VM the clent needs, and the aount of te requested. We call a VM request a job. A job s sad to be a type- job f a type- VM s requested. We consder a te-slotted syste n ths paper, and we say that the sze of the job s S f the VM needs to be hosted for S te slots. Gven our odel of a cloud syste, we next defne the concept of capacty for a cloud. III. CAPACIY OF A CLOUD What s the capacty of a cloud? Frst, as an exaple, consder the three servers defned n Exaple 1. Clearly ths syste has an aggregate capacty of 90 GB of eory, 90 EC copute unts and 1, 000 GB of storage space. However, such a crude defnton of capacty fals to reflect the syste s ablty to host VMs. For exaple, whle = , = 86 90, = , t s easy to verfy that the syste cannot host 4 hgh-eory extra large VMs and 3 hgh-cpu extra large VMs at the sae te. herefore, we have to ntroduce a VM-centrc defnton of capacty. Let A (t denote the set of type- jobs that arrve at the begnnng of te slot t, and let A (t = A (t,.e., the nuber of type- jobs that arrve at the begnnng of te slot t. We let W (t = j A S (t j be the total nuber of te slots requested by the jobs. We assue that W (t s a stochastc process whch s..d. across te slots, E[W (t] = λ and Pr(W (t = 0 > ɛ W for soe ɛ W > 0 for all and t. Many of these assuptons can be relaxed, but we consder the splest odel for ease of exposton. Let D (t denote the nuber of type- jobs that are served by the cloud at te slot t. Note that the job sze of each of these D (t jobs reduces by one at the end of te slot t. he workload due to type- jobs s defned to be the su of the reanng job szes of all jobs of type- n the syste. We let Q (t denote the workload of type- jobs n the network at the begnnng of te slot t, before any other job arrvals. hen the dynacs of Q (t can be descrbed as Q (t + 1 = (Q (t + W (t D (t. (1 We say that the cloud syste s stable f l sup t E[ Q (t] <,.e., the expected total workload n steady-state s bounded. A vector of arrvng loads λ s sad to be supportable f there exsts a resource allocaton echans under whch the cloud s stable. In the followng, we frst dentfy the set of supportable λs. Let N be the set of feasble VM-confguratons on a server. We defne a set C such that } L C = λ : λ = λ ( and λ ( Conv(N., ( =1 where Conv denotes the convex hull. We next use a sple exaple to llustrate the defnton of C. Exaple : Consder a sple cloud syste consstng of three servers. Servers 1 and are of the sae type (.e., they have the sae aount of resources, and server 3 s of a dfferent type. Assue there are two types of VMs. he set of feasble VM confguratons on servers 1 and s assued to be 703

3 1 0 1 Fg. 1. Regons Conv(N 1 and Conv(N Fg.. 3 (, he capacty regon C N 1 = N = (0, 0, (1, 0, (0, 1},.e., each of these servers can at ost host ether one type-1 VM or one type- VM. he set of feasble confguratons on server 3 s assued to be N 3 = (0, 0, (1, 0, (, 0, (0, 1},.e., the server can at ost host ether two type-1 VMs or one type- VM. he regons Conv(N 1 and Conv(N 3 are plotted n Fgure 1. Note that vector (0.75, 0.5 s n the regon Conv(N 1. Whle a type-1 server cannot host 0.75 type-1 VMs and 0.5 type- VM, we can host a type-1 VM on server 1 for 3/4 of the te, and a type- VM on the server for 1/4 of the te to support load (0.75, 0.5. he capacty regon C for ths sple cloud syste s plotted n Fgure. We call C the capacty regon of the cloud. hs defnton of the capacty of a cloud s otvated by slar defntons n [13]. We ntroduce the followng notaton: the servers are ndexed by. Let N ( (t denote the VM-confguraton on server at te slot t. Further defne D(t = N ( (t, so D (t s the total nuber of type- VMs hosted n the cloud at te t. As n [13], t s easy to show the followng results. Lea 1: D(t C for any t. heore 1: For any λ C, l E Q (t =. t IV. HROUGHPU OPIMAL SCHEDULING: CENRALIZED 0 APPROACHES In ths secton, we study centralzed approaches for job schedulng. We assue that jobs arrve at a central job scheduler, and are queued at the job scheduler. he scheduler dspatches a job to a server when the server has enough resources to host the VM requested by the job. In ths settng, servers do not have queues, and do not ake schedulng decsons. We call a job schedulng algorth throughput optal f the algorth can support any λ such that (1 + ɛλ C for soe ɛ > 0. 1 A. Best Ft s not hroughput Optal: A Sple Exaple A schedulng polcy that s used n practce s so called best-ft polcy [14], [15],.e., the job whch uses the ost aount of resources, aong all jobs that can be served, s selected for servce whenever resources becoe avalable. Such a defnton has to be ade ore precse when a VM requests ultple types of ultple resources. In the case of ultple types of resources, we can select one type of resource as reference resource, and defne best ft wth respect to ths resource. If there s a te, then best ft wth respect to another resource s consdered, and so on. Alternatvely, one can consder a partcular lnear or nonlnear cobnaton of the resources as a eta-resource and defne best ft wth respect to the eta-resource. We now show that best ft s not throughput optal. Consder a sple exaple where we have two servers, one type of resource and two types of jobs. A type-1 job requests half of the resource and four te slots of servce, and a type- job requests the whole resource and one te slot of servce. Now assue that ntally, the server 1 hosts one type-1 job and server s epty; two type-1 jobs arrve once every three te slots startng fro te slot 3, and type- jobs arrve accordng to soe arrval process wth arrval rate ɛ startng at te slot 5. Under the best-ft polcy, type-1 jobs are scheduled forever snce type- jobs cannot be scheduled when a type-1 job s n a server. So the workload due to type- jobs wll blow up to nfnty for any ɛ > 0. he syste, however, s clearly stablzable for ɛ < /3. Suppose we schedule type-1 jobs only n te slots 1, 7, 13, 19,...,.e., once every sx te slots. hen te slots 5, 6, 11, 1, 17, 18,... are avalable for type- jobs. So f ɛ < /3, both queues can be stablzed under ths perodc scheduler. he specfc arrval process we constructed s not key to the nstablty of best-ft. Assue type-1 and type- jobs arrve accordng to ndependent Posson processes wth rates λ 1 and λ, respectvely. Fgure 3 s a sulaton result whch shows that the total nuber of backlogged jobs blows up under best-ft wth λ 1 = 0.7 and λ = 0.1, but s stable under a MaxWeght-based polcy wth λ 1 = 0.7 and λ = 0.5. hs exaple rases the queston as to whether there are throughput-optal polces whch stablze the queues for all arrval rates whch le wthn the capacty regon, wthout requrng knowledge of the actual arrval rates. In the next subsecton, we answer ths queston affratvely by relatng the proble to a well-known schedulng proble n wreless networks. However, such a schedulng algorth requres job preepton. In the later sectons, we dscuss non-preeptve polces and the loss of capacty (whch can be ade arbtrarly sall due to non-preepton. B. Preeptve Algorths In ths subsecton, we assue that all servers can be reconfgured at the begnnng of each te slot, and a job can be nterrupted at the begnnng of each te and put back n the queue. We wll study the schees that do not nterrupt job servce n the next subsecton. We further assue the job 704

4 Fg. 3. he nuber of backlogged jobs under the best-ft polcy and a MaxWeght polcy scheduler antans a separate queue for each type of job, and szes of all jobs are bounded by S ax. Recall that Q (t s the workload of type- jobs at the begnnng of te slot t. We consder the followng server-by-server MaxWeght allocaton schee. Server-by-server MaxWeght allocaton: At the begnnng of te slot t, consder the th server. If the set of jobs on the server are not fnshed, ove the back to the central queue. Fnd a VM-confguraton N (t such that N ( (t arg ax N N Q (tn. At server, we create upto N ( (t type- VMs dependng on the nuber of jobs that are backlogged. Let N ( (t be the actual nuber of VMs that were created. hen, we set ( Q (t + 1 = Q (t + W (t. he fact that the proposed algorth s throughput optal follows fro [13] and s stated as a theore below. heore : Assue that a server can serve at ost N ax jobs at the sae te, and E[W(t] σ for any. he server-by-server MaxWeght allocaton s throughput optal,.e., l E Q (t < t f there exsts ɛ > 0 such that (1 + ɛλ C. C. Non-preeptve Algorths N ( he algorth presented n the prevous subsecton requres us to reconfgure the servers and re-allocate jobs at the begnnng of each te slot. In practce, a job ay not be nterruptable or nterruptng a job can be very costly (the syste needs to store a snapshot of the VM to be able to restart the VM later. In ths subsecton, we ntroduce a nonpreeptve algorth, whch s nearly throughput optal. Before we present the algorth, we outlne the basc deas frst. We group te slots nto a super te slot, where > S ax. At the begnnng of a super te slot, a confguraton s chosen accordng to the MaxWeght algorth. When jobs depart a server, the reanng resources n the server are flled agan usng the MaxWeght algorth; however, we pose the constrant that only jobs that can be copleted wthn the super slot can be served. So the algorth yopcally (wthout consderaton of the future uses resources, but s queue-length aware snce t uses the MaxWeght algorth. We now descrbe the algorth ore precsely. Myopc MaxWeght allocaton: We group te slots nto a super te slot. At te slot t, consder the th server. Let N ( (t be the set of VMs that are hosted on server at the begnnng of te slot t,.e., these correspond to the jobs that were scheduled n the prevous te slot but are stll n the syste. hese VMs cannot be reconfgured due to our nonpreepton requreent. he central controller fnds a new vector of confguratons Ñ ( (t to fll up the resources not used by N ( (t,.e., Ñ ( (t arg ax Q (tn, N:N+N ( (t N he central controller selects as any jobs as avalable n the queue, up to a axu of Ñ ( (t type- jobs at server, and subject to the constrant that a type- job can only be ( served f ts sze S j (t od. Let N (t denote the actual nuber of type- jobs selected. Server then serves the N (t ( new jobs of type, and the set of jobs N ( (t left over fro the prevous te slot. he queue length s updated as follows: ( N ( (t ( + N (t. Q (t + 1 = Q (t + W (t Note that ths yopc MaxWeght allocaton algorth dffers fro the server-by-server MaxWeght allocaton n two aspects: ( jobs are not nterrupted when served and ( when a job departs fro a server, new jobs are accepted wthout reconfgurng the server. We next characterze the throughput acheved by the yopc MaxWeght allocaton under the followng assuptons: ( job szes are unforly bounded by S ax, and ( W (t W ax for all and t. heore 3: Any job load that satsfes (1+ɛ S ax λ C for soe ɛ > 0 s supportable under the yopc MaxWeght allocaton. We skp the proof of ths theore because the proof s very slar to the proof of heore 4 n the next secton. It s portant to note that, unlke best ft, the yopc MaxWeght algorth can be ade to acheve any arbtrary fracton of the capacty regon by choosng suffcently large. V. RESOURCE ALLOCAION WIH LOAD BALANCING In the prevous secton, we consdered the case when there was a sngle queue for jobs of sae type, beng served at dfferent servers. hs requres a central authorty to antan a sngle queue for all servers n the syste. A ore dstrbuted 705

5 soluton s to antan queues at each server and route jobs as soon as they arrve. o the best of our knowledge, ths proble does not ft nto the schedulng/routng odel n [13]. However, we show that one can stll show use MaxWeght-type schedulng f the servers are load-balanced usng a jon-theshortest-queue (JSQ routng rule. In our odel, we assue that each server antans M dfferent queues for dfferent types of jobs. It then uses ths queue length nforaton n akng schedulng decsons. Let Q denote the vector of these queue lengths where Q s the queue length of type jobs at server. Routng and schedulng are perfored as descrbed n Algorth 1. Algorth 1 JSQ Routng and Myopc Maxweght Schedulng 1 Routng Algorth (JSQ Routng: All the type jobs that arrve n te slot t are routed to the server wth the shortest queue for type jobs.e., the server (t = arg n Q (t. herefore, the arrvals to Q n te 1,,,,L} slot t are gven by W (t = W (t f = (t 0 otherwse Schedulng Algorth (Myopc MaxWeght Schedulng for each server : te slots are grouped nto a super te slot. A MaxWeght confguraton s chosen at the begnnng of a super te slot. So, for t = n, confguraton Ñ ( (t s chosen accordng to Ñ ( (t arg ax N N Q (tn For all other t, at the begnnng of the te slot, a new confguraton s chosen as follows: Ñ ( (t arg ax Q (tn N:N+N ( (t N where N ( (t s the confguraton of jobs at server that are stll n servce at the end of the prevous te slot. As any jobs as avalable are selected for servce fro the queue, up to a axu of Ñ ( (t jobs of type, and subject to the constrant that a new type job s served only f t can fnsh ts servce by the end of the super te slot,.e., only f S j (t od. Let N ( (t denote the actual nuber of type jobs selected at server and defne N ( (t = N ( (t +N ( (t. he queue lengths are updated as follows: (3 Q (t + 1 = Q (t + W (t N ( (t. (4 he followng theore characterzes the throughput perforance of the algorth. (1+ɛ heore 4: Any job load vector that satsfes S ax λ C for soe ɛ > 0 s supportable under the JSQ routng and yopc MaxWeght allocaton as descrbed n Algorth 1 Proof: Let Y (t denote the state of the queue for type jobs, where Y j (t s the reanng job sze of the jth type- job at server. Frst, t s easy to see that Y(t = Y (t}, s a Markov chan under the yopc MaxWeght schedulng. Further defne S = y : Pr(Y(t = y Y(0 = 0 for soe t}, then Y(t s an rreducble Markov chan on state space S assung Y(0 = 0. hs cla holds because ( any state n S s reachable fro 0 and ( snce Pr(W (t = 0 ɛ W for all and t, the Markov chan can ove fro Y(t to 0 n fnte te wth a postve probablty. Further Q (t = j Yj, (t,.e., Q (t s a functon of Y (t. We wll frst show that the ncrease of Q (tn ( (t s bounded wthn a super te slot. For any t such that 1 (t od S ax, for each server, Q (tn ( (t 1 = Q (tn ( (t + ( Q (t N ( (t 1 N ( (t a Q (tn ( (t + Q (tñ ( (t ( Q (tn ( (t + Q (tñ ( (t I Q(t S axn ax = + ( Q (tn ( (t + Q (tñ ( (t I Q(t<S axn ax (b Q (tn ( (t + MS ax Nax where the nequalty (a follows fro the defnton Ñ ( (t; and nequalty (b holds because when Q (t S ax N ax, there are enough nuber of type- jobs to be allocated to the servers, and when 1 (t od S ax, all backlogged jobs are elgble to be served n ters of job szes. Now snce Q (t Q (t 1 = W (t N ( (t W ax + N ax, we have Q (t 1N ( (t 1 β + Q (tn ( (t (5 where β = MN ax (W ax + N ax + MS ax N ax. Let V (t = Q(t be the Lyapunov functon. Let t = n + τ for 0 τ <. hen, E[V (n + τ + 1 V (n + τ Q(n = q] [ ( =E Q (t + W (t N ( (t ] Q (t [ =E ( Q (t W (t N ( (t + ] ( W (t N (t ( Q(n = q (6 (7 706

6 [ K + E Q (tw (t ] Q (tn ( (t =K + E[Q (t(tw (t Q(n = q] E Q (tn ( (t =K + λ E[Q (t(t Q(n = q] E Q (tn ( (t K + λ W ax τ + λ E[Q (n (n Q(n = q] E Q (tn ( (t =K + λ W ax τ + λ q E Q (tn ( (t (8 (9 (10 (11 (1 where K = ML(S ax + N ax and = (n = arg n q. Equaton (9 follows fro the defnton of 1,,,,L} W n the routng algorth n (3. Equaton (10 follows fro the ndependence of the arrval process fro the queue length process. Inequalty (11 coes fro the fact that Q (t(t Q (n (t Q (n + W ax τ. Now, applyng (5 repeatedly for t [n, (n+1 S ax ], and sung over, we get Q (tn ( (t L(t n β Q (n N ( (n. (13 (1+ɛ Snce, S ax λ C, there exsts } λ such that (1+ɛ S ax λ Conv(N for all and λ = λ. Accordng to the schedulng algorth, for each, we have that (1 + ɛ Q (n λ S ax Q (n N ( (n. (14 hus, we get, Q (tn ( (t L(t n β Q (n N ( (n (15 L(t n β (1 + ɛ Q (n λ (16 S ax L(t n β (1 + ɛ Q S (n λ ax = L(t n β (1 + ɛ Q S (n λ. (17 ax Substtutng ths n (1, we get, for t [n, (n + 1 S ax ], E[V (n + τ + 1 V (n + τ Q(n = q] K + λ W ax τ + L(t n β + λ q (1 + ɛ q S λ. ax (18 Sung the drft for τ fro 0 to 1 usng (18 for τ [0, S ax ], and (1 for the reanng τ, we get E[V ((n + 1 V (n Q(n = q] K + 1 S ax 1 λ W ax τ + Lβ τ + (1 + ɛ λ q q S λ ( S ax ax K + 1 S ax 1 λ W ax τ + Lβ τ ɛ q λ. he theore then follows fro the Foster-Lyapunov theore [16], [17]. VI. SIMPLER LOAD BALANCING ALGORIHMS hough JSQ routng algorth s throughput optal, the job scheduler needs the queue length nforaton fro all the servers. hs poses a consderable councaton overhead as the arrval rates of jobs and nuber of servers ncrease. In ths secton, we present two alternatves whch have uch lower routng coplexty. A. Power-of-two-choces Routng and Myopc MaxWeght Schedulng An alternate to JSQ routng s the power-of-two-choces algorth [18], [19], [0], whch s uch spler to pleent. When a job arrves, two servers are sapled at rando, and the job s routed to the server wth the saller queue for that job type. In our algorth, n each te slot t, for each type of job, two servers 1 (t and (t are chosen unforly at rando. he job scheduler then routes all the type job arrvals n ths te slot to the server wth shorter queue length aong these two,.e., (t = arg n Q (t and so 1 (t, (t} W (t f = W (t = (t 0 otherwse. 707

7 Otherwse, the algorth s dentcal to the JSQ-Myopc MaxWeght algorth consdered earler. In ths secton, we wll provde a lower bound on the throughput of ths powerof-two-choces algorth n the non-preeptve case when all the servers have dentcal resource constrants. heore 5: When all the severs are dentcal, any job load that satsfes (1 + ɛ S ax λ C for soe ɛ > 0 s supportable under the power-of-two-choces routng and yopc MaxWeght allocaton algorth. Proof: Agan, we use V (t = Q(t as the Lyapunov functon. For fxed, let X (t be the rando varable whch denotes the two servers that were chosen by the routng algorth at te t for jobs of type. X (t s then unforly dstrbuted over all sets of two servers. Now, usng the tower property of condtonal expectaton, we have, E Q (tw (t [ =E X [E Q (tw (t ]] Q(n = q, X (t =, j } =E X [E [Q (tw (t + Q j (tw j (t Q(n = q, X(t =, j }]] =E X [E [n (Q (t, Q j (t W (t Q(n = q, X(t =, j }]] (19 [ Q (t + Q j (t E X [E W (t ]] Q(n = q, X(t =, j } q + q j =E X λ L 1 =λ ( L 1 q (0 =λ q L. (1 Equaton (19 follows fro the routng algorth and (0 follows fro the fact that X (t s unforly dstrbuted. Snce the schedulng algorth s dentcal to Algorth 1, (13 stll holds for any t such that 1 (t od S ax. hus, we have, Q (tn ( (t L(t n β Q (n N ( (n. ( We assue that all the servers are dentcal. So, C s obtaned by sung L copes of Conv(N. hus, snce (1+ɛ S ax λ C, (1+ɛ λ we have that S ax L Conv(N = Conv(N for all. Accordng to the schedulng algorth, for each, we have that (1 + ɛ Q (n λ S ax L hus, we get, Q (tn ( (t L(t n β Q (n N ( (n. (3 (1 + ɛ S ax Q (n λ L (4 L(t n β (1 + ɛ λ Q (n. (5 S ax L Now, substtutng (1 and (16 n (8 (whch also holds for power-of-two-choces routng and sung over t [n, (n + 1 1], we get E[V ((n + 1 V (n Q(n = q] K + λ q L (1 + ɛ S ax S ax 1 K + Lβ + Lβ λ S ax 1 q L τ ɛ τ ( S ax λ q L. hs proof can be copleted by applyng the Foster-Lyapunov theore [16], [17]. B. Pck-and-Copare Routng and Myopc MaxWeght Schedulng One drawback of the power-of-two-choces schedulng s that t s throughput optal only when all servers are dentcal. In the case of nondentcal servers, one can use pck-andcopare routng algorth nstead of power-of-two-choces. he algorth s otvated by the pck-and-copare algorth for wreless schedulng and swtch schedulng [1], and s as sple to pleent as power-of-two-choces, and can be shown to be optal even f the servers are not dentcal. We descrbe ths next. he schedulng algorth s dentcal to the prevous case. Pck-and-copare routng works as follows. In each te slot t, for each type of job, a server (t s chosen unforly at rando and copared wth the server to whch jobs were routed n the prevous te slot. he server wth the shorter queue length aong the two s chosen and all the type job arrvals n ths te slot are routed to that server. Let (t be the server to whch jobs wll be routed n te slot t. hen, (t = arg n Q (t and so (t, (t 1} W (t f = W (t = (t 0 otherwse. 708

8 (1+ɛ heore 6: Any job load vector that satsfes S ax λ C for soe ɛ > 0 s supportable under the pck-and-copare routng and yopc MaxWeght allocaton algorth. Proof: Consder the rreducble Markov chan Y(t = (Y(t, (t} and the Lyapunov functon V (t = Q(t. hen, as n (8 for t n, we have hs s possble because f λ > 0 and λ s not on the boundary of C, one can always fnd λ } so that λ > 0. Snce the schedulng part of the algorth s dentcal to Algorth 1, (16 stll holds for t [n, (n + 1 S ax ]. hus, we have Q (tn ( (t L(t n β (1 + ɛ Q (n λ S. (7 ax We also need a bound on the ncrease n Q (tn ( (t over ultple super te slots. It s not dffcult to show that for any t such that 1 (t od S ax, (see [] for detals Q (tn ( (t (8 L(t n β (1 + ɛ Q (n λ S. (9 ax Fx. Let n = arg n Q (n. Note that 1,,,,L} Q (t Q (t 1 = W (t N ( (t W ax + N ax. herefore, once there s a t 0 n such that (t 0 satsfes Q (t 0(t 0 Q n (t 0, (30 then, for all t t 0, we have Q (t(t Q n (n +(t n (W ax + N ax. Probablty that (30 does not happen s at ost ( 1 L 1 (t0 n. Choose t0 so that ths probablty s less than p = ɛ/4κ. hen, (1 + κp = 1 + ɛ/4. Choose k so that k > (t 0 n and ((n + k t 0 + κ(t 0 n k (1 + ɛ/4. hen (see [] for detals, [ (n+k 1 E Q (tw (t, (n = t=n K 1 + k (1 + 3ɛ/4 q λ (31 where K 1 = k τ (W ax + N ax LW ax. Now, substtutng (31 and (8 n (6 and sung over all t [n, (n + 1 1], we get E[V ((n + k V (n Q(n = q, (n = ] E[V (t + 1 V (t Q(n = q, (n = ] K 1 k ɛ q λ K + E Q (tw (t, (n = where K = k K +MK 1 +Lβ k S ax 1 τ. he result follows fro the Foster-Lyapunov theore [16], [17]. E Q (tn ( (t, (n =. VII. SIMULAIONS (6 Snce, (1 + ɛ S ax λ C, there exsts λ } In ths secton, we use sulatons to copare the perforance of the centralzed yopc MaxWeght schedulng algo- such that (1 + ɛ S ax λ Conv(N for all and λ = λ. hs rth, and the jont routng and schedulng algorth based } λ can be chosen so that there s a κ so that λ κλ on the power-of-two-choces and MaxWeght schedulng. We. consder a cloud coputng cluster wth 100 dentcal servers, and each server has the hardware confguraton specfed n Exaple 1. We assue jobs beng served n ths cloud belong to one of the three types specfed n able I. So VM confguratons (, 0, 0, (1, 0, 1, and (0, 1, 1 are the three axal VM confguratons for each server. It s easy to verfy that the load vector λ = (1, 1 3, 3 s on the boundary of the capacty regon of a server. o odel the large varablty n jobs szes, we assue job szes are dstrbuted as follows: when a new job s generated, wth probablty 0.7, the sze s an nteger that s unforly dstrbuted n the nterval [1, 50], wth probablty 0.15, t s an nteger that s unforly dstrbuted between 51 and 300, and wth probablty 0.15, t s unforly dstrbuted between 451 and 500. herefore, the average job sze s and the axu job sze s 500. We further assue the nuber of type- jobs arrvng at each te slot follows a Bnoal dstrbuton wth paraeter (α λ 130.5, 100. We vared the paraeter α fro 0.5 to 1 n our sulatons, whch vared the traffc ntensty of the cloud syste fro 0.5 to 1, where traffc ntensty s the factor by whch the load vector has to be dvded so that t les on the boundary of the capacty regon. Each sulaton was run for 500, 000 te slots. Frst we study the dfference between power-of-two-choce routng and JSQ routng by coparng the ean delays of the two algorths at varous traffc ntenstes for dfferent choces of frae szes. Our sulaton results ndcate that the delay perforance of the two algorths was not very dfferent. Due to page ltatons, we only provde a representatve saple of our sulatons here for the case where the frae sze s 4000 n Fgure 4. Next, we show the perforance of our algorths for varous values of the frae sze n Fgure 5. Agan, we ] have only shown a representatve saple for the power-of-twochoces routng (wth yopc MaxWeght schedulng. Fro heores 3 and 5, we know that any load less than Sax s supportable. he sulatons ndcate that the syste s stable even for the loads greater than ths value. hs s to be expected snce our proofs of heores 3 and 5 essentally gnore the 709

9 also see to provde good delay perforance. IX. ACKNOWLEDGEMENS Research supported n part by AFOSR MURI FA , ARO MURI W911NF , and NSF Grants CNS and CNS Fg. 4. Coparson of the ean delays n the cloud coputng cluster n the case wth a coon queue and n the case wth power-of-two-choces routng when frae sze s 4000 Fg. 5. Coparson of power-of-two-choces routng algorth for varous frae lengths jobs that are scheduled n the last S ax te slots of a frae. However, the fact that the stablty regon s larger for larger values of s confred by the sulatons. It s even ore nterestng to observe the delay perforance of our algorths as ncreases. Fgure 5 ndcates that the delay perforance does not degrade as ncreases and the throughput ncreases wth. So the use of queue-length nforaton sees to be the key ngredent of the algorth whle the optal pleentaton of the MaxWeght algorth sees to be secondary. VIII. CONCLUSIONS We consdered a stochastc odel for load balancng and schedulng n cloud coputng clusters. A prary contrbuton s the developent of frae-based non-preeptve VM confguraton polces. hese polces can be ade nearly throughput-optal by choosng suffcently long frae duratons, whereas the wdely used best ft polcy was shown to be not throughput optal. Sulatons ndcate that long frae duratons are not only good fro a throughput perspectve but REFERENCES [1] EC, [] AppEngne, [3] Azure, [4] I. Foster, Y. Zhao, I. Racu, and S. Lu, Cloud coputng and grd coputng 360-degree copared, n Grd Coputng Envronents Workshop, 008. GCE 08, 008, pp [5] M. Arbrust, A. Fox, R. Grffth, A. Joseph, R. Katz, A. Konwnsk, G. Lee, D. Patterson, A. Rabkn, I. Stoca et al., Above the clouds: A berkeley vew of cloud coputng, 009, tech. Rep. UCb/eeCs-009-8, EECS departent, U.C. berkeley. [6] D. A. Menasce and P. Ngo, Understandng cloud coputng: Experentaton and capacty plannng, n Proc. 009 Coputer Measureent Group Conf., 009. [7] X. Meng, V. Pappas, and L. Zhang, Iprovng the scalablty of data center networks wth traffc-aware vrtual achne placeent, n Proc. IEEE Infoco., 010, pp [8] Y. Yazr, C. Matthews, R. Farahbod, S. Nevlle, A. Gutoun, S. Gant, and Y. Coady, Dynac resource allocaton n coputng clouds usng dstrbuted ultple crtera decson analyss, n 010 IEEE 3rd Internatonal Conference on Cloud Coputng, 010, pp [9] K. sakalozos, H. Kllap, E. Stard, M. Roussopoulos, D. Paparas, and A. Dels, Flexble use of cloud resources through proft axzaton and prce dscrnaton, n Data Engneerng (ICDE, 011 IEEE 7th Internatonal Conference on, 011, pp [10] M. Ln, A. Weran, L. L. H. Andrew, and E. hereska, Dynac rghtszng for power-proportonal data centers, n Proc. IEEE Infoco., 011, pp [11] M. Wang, X. Meng, and L. Zhang, Consoldatng vrtual achnes wth dynac bandwdth deand n data centers, n Proc. IEEE Infoco., 011, pp [1] U. Shara, P. Shenoy, S. Sahu, and A. Shakh, Kngfsher: Cost-aware elastcty n the cloud, n Proc. IEEE Infoco., 011, pp [13] L. assulas and A. Ephredes, Stablty propertes of constraned queueng systes and schedulng polces for axu throughput n ulthop rado networks, IEEE rans. Autoat. Contr., vol. 4, pp , Deceber 199. [14] B. Spetkap and M. Bchler, A atheatcal prograng approach for server consoldaton probles n vrtualzed data centers, IEEE ransactons on Servces Coputng, pp , 010. [15] A. Beloglazov and R. Buyya, Energy effcent allocaton of vrtual achnes n cloud data centers, n th IEEE/ACM Internatonal Conference on Cluster, Cloud and Grd Coputng, 010, pp [16] S. Asussen, Appled Probablty and Queues. New York: Sprnger- Verlag, 003. [17] S. Meyn and R. L. weede, Markov chans and stochastc stablty. Cabrdge Unversty Press, 009. [18] M. Mtzenacher, he power of two choces n randozed load balancng, Ph.D. dssertaton, Unversty of Calforna at Berkeley, [19] Y.. He and D. G. Down, Lted choce and localty consderatons for load balancng, Perforance Evaluaton, vol. 65, no. 9, 008. [0] H. Chen and H. Q. Ye, Asyptotc optalty of balanced routng, 010, lgtyehq/papers/chenye11or.pdf. [1] L. assulas, Lnear coplexty algorths for axu throughput n radonetworks and nput queued swtches, n Proc. IEEE Infoco., [] S. Magulur, R. Srkant, and L. Yng, Stochastc odels of load balancng and schedulng n cloud coputng clusters, echncal Report, 710

Stochastic Models of Load Balancing and Scheduling in Cloud Computing Clusters

Stochastic Models of Load Balancing and Scheduling in Cloud Computing Clusters Stochastc Models of Load Balancng and Schedulng n Cloud Coputng Clusters Sva Theja Magulur and R. Srkant Departent of ECE and CSL Unversty of Illnos at Urbana-Chapagn sva.theja@gal.co; rsrkant@llnos.edu

More information

Stochastic Models of Load Balancing and Scheduling in Cloud Computing Clusters

Stochastic Models of Load Balancing and Scheduling in Cloud Computing Clusters Stochastc Models of Load Balancng and Schedulng n Cloud Coputng Clusters Sva Theja Magulur and R. Srkant Departent of ECE and CSL Unversty of Illnos at Urbana-Chapagn sva.theja@gal.co; rsrkant@llnos.edu

More information

Basic Queueing Theory M/M/* Queues. Introduction

Basic Queueing Theory M/M/* Queues. Introduction Basc Queueng Theory M/M/* Queues These sldes are created by Dr. Yh Huang of George Mason Unversty. Students regstered n Dr. Huang's courses at GMU can ake a sngle achne-readable copy and prnt a sngle copy

More information

An Electricity Trade Model for Microgrid Communities in Smart Grid

An Electricity Trade Model for Microgrid Communities in Smart Grid An Electrcty Trade Model for Mcrogrd Countes n Sart Grd Tansong Cu, Yanzh Wang, Shahn Nazaran and Massoud Pedra Unversty of Southern Calforna Departent of Electrcal Engneerng Los Angeles, CA, USA {tcu,

More information

BANDWIDTH ALLOCATION AND PRICING PROBLEM FOR A DUOPOLY MARKET

BANDWIDTH ALLOCATION AND PRICING PROBLEM FOR A DUOPOLY MARKET Yugoslav Journal of Operatons Research (0), Nuber, 65-78 DOI: 0.98/YJOR0065Y BANDWIDTH ALLOCATION AND PRICING PROBLEM FOR A DUOPOLY MARKET Peng-Sheng YOU Graduate Insttute of Marketng and Logstcs/Transportaton,

More information

How Much to Bet on Video Poker

How Much to Bet on Video Poker How Much to Bet on Vdeo Poker Trstan Barnett A queston that arses whenever a gae s favorable to the player s how uch to wager on each event? Whle conservatve play (or nu bet nzes large fluctuatons, t lacks

More information

TheHow and Why of Having a Successful Home Office

TheHow and Why of Having a Successful Home Office Near Optal Onlne Algorths and Fast Approxaton Algorths for Resource Allocaton Probles Nkhl R Devanur Kaal Jan Balasubraanan Svan Chrstopher A Wlkens Abstract We present algorths for a class of resource

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

Online Algorithms for Uploading Deferrable Big Data to The Cloud

Online Algorithms for Uploading Deferrable Big Data to The Cloud Onlne lgorths for Uploadng Deferrable Bg Data to The Cloud Lnquan Zhang, Zongpeng L, Chuan Wu, Mnghua Chen Unversty of Calgary, {lnqzhan,zongpeng}@ucalgary.ca The Unversty of Hong Kong, cwu@cs.hku.hk The

More information

Inventory Control in a Multi-Supplier System

Inventory Control in a Multi-Supplier System 3th Intl Workng Senar on Producton Econocs (WSPE), Igls, Autrche, pp.5-6 Inventory Control n a Mult-Suppler Syste Yasen Arda and Jean-Claude Hennet LAAS-CRS, 7 Avenue du Colonel Roche, 3077 Toulouse Cedex

More information

A R T I C L E S DYNAMIC VEHICLE DISPATCHING: OPTIMAL HEAVY TRAFFIC PERFORMANCE AND PRACTICAL INSIGHTS

A R T I C L E S DYNAMIC VEHICLE DISPATCHING: OPTIMAL HEAVY TRAFFIC PERFORMANCE AND PRACTICAL INSIGHTS A R T I C L E S DYAMIC VEHICLE DISPATCHIG: OPTIMAL HEAVY TRAFFIC PERFORMACE AD PRACTICAL ISIGHTS OAH GAS OPIM Departent, The Wharton School, Unversty of Pennsylvana, Phladelpha, Pennsylvana 19104-6366

More information

Dynamic Resource Allocation in Clouds: Smart Placement with Live Migration

Dynamic Resource Allocation in Clouds: Smart Placement with Live Migration Dynac Resource Allocaton n Clouds: Sart Placeent wth Lve Mgraton Mahlouf Had Ingéneur de Recherche ahlouf.had@rt-systex.fr Avec : Daal Zeghlache (TSP) daal.zeghlache@teleco-sudpars.eu FONDATION DE COOPERATION

More information

Scan Detection in High-Speed Networks Based on Optimal Dynamic Bit Sharing

Scan Detection in High-Speed Networks Based on Optimal Dynamic Bit Sharing Scan Detecton n Hgh-Speed Networks Based on Optal Dynac Bt Sharng Tao L Shgang Chen Wen Luo Mng Zhang Departent of Coputer & Inforaton Scence & Engneerng, Unversty of Florda Abstract Scan detecton s one

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

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

Power-of-Two Policies for Single- Warehouse Multi-Retailer Inventory Systems with Order Frequency Discounts

Power-of-Two Policies for Single- Warehouse Multi-Retailer Inventory Systems with Order Frequency Discounts Power-of-wo Polces for Sngle- Warehouse Mult-Retaler Inventory Systems wth Order Frequency Dscounts José A. Ventura Pennsylvana State Unversty (USA) Yale. Herer echnon Israel Insttute of echnology (Israel)

More information

Research Article Load Balancing for Future Internet: An Approach Based on Game Theory

Research Article Load Balancing for Future Internet: An Approach Based on Game Theory Appled Matheatcs, Artcle ID 959782, 11 pages http://dx.do.org/10.1155/2014/959782 Research Artcle Load Balancng for Future Internet: An Approach Based on Gae Theory Shaoy Song, Tngje Lv, and Xa Chen School

More information

Revenue Maximization Using Adaptive Resource Provisioning in Cloud Computing Environments

Revenue Maximization Using Adaptive Resource Provisioning in Cloud Computing Environments 202 ACM/EEE 3th nternatonal Conference on Grd Coputng evenue Maxzaton sng Adaptve esource Provsonng n Cloud Coputng Envronents Guofu Feng School of nforaton Scence, Nanng Audt nversty, Nanng, Chna nufgf@gal.co

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

Capacity Planning for Virtualized Servers

Capacity Planning for Virtualized Servers Capacty Plannng for Vrtualzed Servers Martn Bchler, Thoas Setzer, Benjan Spetkap Departent of Inforatcs, TU München 85748 Garchng/Munch, Gerany (bchler setzer benjan.spetkap)@n.tu.de Abstract Today's data

More information

An Analytical Model of Web Server Load Distribution by Applying a Minimum Entropy Strategy

An Analytical Model of Web Server Load Distribution by Applying a Minimum Entropy Strategy Internatonal Journal of Coputer and Councaton Engneerng, Vol. 2, No. 4, July 203 An Analytcal odel of Web Server Load Dstrbuton by Applyng a nu Entropy Strategy Teeranan Nandhakwang, Settapong alsuwan,

More information

A Fuzzy Optimization Framework for COTS Products Selection of Modular Software Systems

A Fuzzy Optimization Framework for COTS Products Selection of Modular Software Systems Internatonal Journal of Fuy Systes, Vol. 5, No., June 0 9 A Fuy Optaton Fraework for COTS Products Selecton of Modular Software Systes Pankaj Gupta, Hoang Pha, Mukesh Kuar Mehlawat, and Shlp Vera Abstract

More information

Ganesh Subramaniam. American Solutions Inc., 100 Commerce Dr Suite # 103, Newark, DE 19713, USA

Ganesh Subramaniam. American Solutions Inc., 100 Commerce Dr Suite # 103, Newark, DE 19713, USA 238 Int. J. Sulaton and Process Modellng, Vol. 3, No. 4, 2007 Sulaton-based optsaton for ateral dspatchng n Vendor-Managed Inventory systes Ganesh Subraana Aercan Solutons Inc., 100 Coerce Dr Sute # 103,

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

Virtual machine resource allocation algorithm in cloud environment

Virtual machine resource allocation algorithm in cloud environment COMPUTE MOELLIN & NEW TECHNOLOIES 2014 1(11) 279-24 Le Zheng Vrtual achne resource allocaton algorth n cloud envronent 1, 2 Le Zheng 1 School of Inforaton Engneerng, Shandong Youth Unversty of Poltcal

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

Quality of Service Analysis and Control for Wireless Sensor Networks

Quality of Service Analysis and Control for Wireless Sensor Networks Qualty of ervce Analyss and Control for Wreless ensor Networs Jaes Kay and Jeff Frol Unversty of Veront ay@uv.edu, frol@eba.uv.edu Abstract hs paper nvestgates wreless sensor networ spatal resoluton as

More information

CONSTRUCTION OF A COLLABORATIVE VALUE CHAIN IN CLOUD COMPUTING ENVIRONMENT

CONSTRUCTION OF A COLLABORATIVE VALUE CHAIN IN CLOUD COMPUTING ENVIRONMENT CONSTRUCTION OF A COLLAORATIVE VALUE CHAIN IN CLOUD COMPUTING ENVIRONMENT Png Wang, School of Econoy and Manageent, Jangsu Unversty of Scence and Technology, Zhenjang Jangsu Chna, sdwangp1975@163.co Zhyng

More information

Analysis of Energy-Conserving Access Protocols for Wireless Identification Networks

Analysis of Energy-Conserving Access Protocols for Wireless Identification Networks From the Proceedngs of Internatonal Conference on Telecommuncaton Systems (ITC-97), March 2-23, 1997. 1 Analyss of Energy-Conservng Access Protocols for Wreless Identfcaton etworks Imrch Chlamtac a, Chara

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

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

Support Vector Machines

Support Vector Machines Support Vector Machnes Max Wellng Department of Computer Scence Unversty of Toronto 10 Kng s College Road Toronto, M5S 3G5 Canada wellng@cs.toronto.edu Abstract Ths s a note to explan support vector machnes.

More information

The Packing Server for Real-Time Scheduling of MapReduce Workflows

The Packing Server for Real-Time Scheduling of MapReduce Workflows The Packng Server for Real-Te Schedulng of MapReduce Workflows Shen L, Shaohan Hu, Tarek Abdelzaher Unversty of Illnos at Urbana-Chapagn {shenl3, shu7, zaher}@llnos.edu Abstract Ths paper develops new

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

Maximizing profit using recommender systems

Maximizing profit using recommender systems Maxzng proft usng recoender systes Aparna Das Brown Unversty rovdence, RI aparna@cs.brown.edu Clare Matheu Brown Unversty rovdence, RI clare@cs.brown.edu Danel Rcketts Brown Unversty rovdence, RI danel.bore.rcketts@gal.co

More information

II. THE QUALITY AND REGULATION OF THE DISTRIBUTION COMPANIES I. INTRODUCTION

II. THE QUALITY AND REGULATION OF THE DISTRIBUTION COMPANIES I. INTRODUCTION Fronter Methodology to fx Qualty goals n Electrcal Energy Dstrbuton Copanes R. Rarez 1, A. Sudrà 2, A. Super 3, J.Bergas 4, R.Vllafáfla 5 1-2 -3-4-5 - CITCEA - UPC UPC., Unversdad Poltécnca de Cataluña,

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

A Novel Dynamic Role-Based Access Control Scheme in User Hierarchy

A Novel Dynamic Role-Based Access Control Scheme in User Hierarchy Journal of Coputatonal Inforaton Systes 6:7(200) 2423-2430 Avalable at http://www.jofcs.co A Novel Dynac Role-Based Access Control Schee n User Herarchy Xuxa TIAN, Zhongqn BI, Janpng XU, Dang LIU School

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

Least Squares Fitting of Data

Least Squares Fitting of Data Least Squares Fttng of Data Davd Eberly Geoetrc Tools, LLC http://www.geoetrctools.co/ Copyrght c 1998-2016. All Rghts Reserved. Created: July 15, 1999 Last Modfed: January 5, 2015 Contents 1 Lnear Fttng

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

Web Service-based Business Process Automation Using Matching Algorithms

Web Service-based Business Process Automation Using Matching Algorithms Web Servce-based Busness Process Autoaton Usng Matchng Algorths Yanggon K and Juhnyoung Lee 2 Coputer and Inforaton Scences, Towson Uversty, Towson, MD 2252, USA, yk@towson.edu 2 IBM T. J. Watson Research

More information

PRIOR ROBUST OPTIMIZATION. Balasubramanian Sivan. A dissertation submitted in partial fulfillment of the requirements for the degree of

PRIOR ROBUST OPTIMIZATION. Balasubramanian Sivan. A dissertation submitted in partial fulfillment of the requirements for the degree of PRIOR ROBUST OPTIMIZATION By Balasubraanan Svan A dssertaton subtted n partal fulfllent of the requreents for the degree of Doctor of Phlosophy (Coputer Scences) at the UNIVERSITY OF WISCONSIN MADISON

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

Near Optimal Online Algorithms and Fast Approximation Algorithms for Resource Allocation Problems

Near Optimal Online Algorithms and Fast Approximation Algorithms for Resource Allocation Problems Near Optal Onlne Algorths and Fast Approxaton Algorths for Resource Allocaton Probles ABSTRACT Nhl R Devanur Mcrosoft Research Redond WA USA ndev@crosoftco Balasubraanan Svan Coputer Scences Dept Unv of

More information

A Statistical Model for Detecting Abnormality in Static-Priority Scheduling Networks with Differentiated Services

A Statistical Model for Detecting Abnormality in Static-Priority Scheduling Networks with Differentiated Services A Statstcal odel for Detectng Abnoralty n Statc-Prorty Schedulng Networks wth Dfferentated Servces ng L 1 and We Zhao 1 School of Inforaton Scence & Technology, East Chna Noral Unversty, Shangha 0006,

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

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

This paper concerns the evaluation and analysis of order

This paper concerns the evaluation and analysis of order ORDER-FULFILLMENT PERFORMANCE MEASURES IN AN ASSEMBLE- TO-ORDER SYSTEM WITH STOCHASTIC LEADTIMES JING-SHENG SONG Unversty of Calforna, Irvne, Calforna SUSAN H. XU Penn State Unversty, Unversty Park, Pennsylvana

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

Robust Design of Public Storage Warehouses. Yeming (Yale) Gong EMLYON Business School

Robust Design of Public Storage Warehouses. Yeming (Yale) Gong EMLYON Business School Robust Desgn of Publc Storage Warehouses Yemng (Yale) Gong EMLYON Busness School Rene de Koster Rotterdam school of management, Erasmus Unversty Abstract We apply robust optmzaton and revenue management

More information

Description of the Force Method Procedure. Indeterminate Analysis Force Method 1. Force Method con t. Force Method con t

Description of the Force Method Procedure. Indeterminate Analysis Force Method 1. Force Method con t. Force Method con t Indeternate Analyss Force Method The force (flexblty) ethod expresses the relatonshps between dsplaceents and forces that exst n a structure. Prary objectve of the force ethod s to deterne the chosen set

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

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

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

THE DISTRIBUTION OF LOAN PORTFOLIO VALUE * Oldrich Alfons Vasicek

THE DISTRIBUTION OF LOAN PORTFOLIO VALUE * Oldrich Alfons Vasicek HE DISRIBUION OF LOAN PORFOLIO VALUE * Oldrch Alfons Vascek he amount of captal necessary to support a portfolo of debt securtes depends on the probablty dstrbuton of the portfolo loss. Consder a portfolo

More information

denote the location of a node, and suppose node X . This transmission causes a successful reception by node X for any other node

denote the location of a node, and suppose node X . This transmission causes a successful reception by node X for any other node Fnal Report of EE359 Class Proect Throughput and Delay n Wreless Ad Hoc Networs Changhua He changhua@stanford.edu Abstract: Networ throughput and pacet delay are the two most mportant parameters to evaluate

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

Relay Secrecy in Wireless Networks with Eavesdropper

Relay Secrecy in Wireless Networks with Eavesdropper Relay Secrecy n Wreless Networks wth Eavesdropper Parvathnathan Venktasubramanam, Tng He and Lang Tong School of Electrcal and Computer Engneerng Cornell Unversty, Ithaca, NY 14853 Emal : {pv45, th255,

More information

The Power of Slightly More than One Sample in Randomized Load Balancing

The Power of Slightly More than One Sample in Randomized Load Balancing The Power of Slghtly More than One Sample n Randomzed oad Balancng e Yng, R. Srkant and Xaohan Kang Abstract In many computng and networkng applcatons, arrvng tasks have to be routed to one of many servers,

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

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

Yixin Jiang and Chuang Lin. Minghui Shi and Xuemin Sherman Shen*

Yixin Jiang and Chuang Lin. Minghui Shi and Xuemin Sherman Shen* 198 Int J Securty Networks Vol 1 Nos 3/4 2006 A self-encrypton authentcaton protocol for teleconference servces Yxn Jang huang Ln Departent of oputer Scence Technology Tsnghua Unversty Beng hna E-al: yxang@csnet1cstsnghuaeducn

More information

An MILP model for planning of batch plants operating in a campaign-mode

An MILP model for planning of batch plants operating in a campaign-mode An MILP model for plannng of batch plants operatng n a campagn-mode Yanna Fumero Insttuto de Desarrollo y Dseño CONICET UTN yfumero@santafe-concet.gov.ar Gabrela Corsano Insttuto de Desarrollo y Dseño

More information

Technical Report, SFB 475: Komplexitätsreduktion in Multivariaten Datenstrukturen, Universität Dortmund, No. 1998,04

Technical Report, SFB 475: Komplexitätsreduktion in Multivariaten Datenstrukturen, Universität Dortmund, No. 1998,04 econstor www.econstor.eu Der Open-Access-Publkatonsserver der ZBW Lebnz-Inforatonszentru Wrtschaft The Open Access Publcaton Server of the ZBW Lebnz Inforaton Centre for Econocs Becka, Mchael Workng Paper

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

Packet Reorderng Analysis

Packet Reorderng Analysis On Montorng of End-to-End Packet Reorderng over the Internet Bn Ye 1 Anura P. Jayasuana 1 Nschal M. Pratla 2 1Coputer Networkng Research laboratory, Colorado State Unversty, Fort Collns, CO 8523, USA 2

More information

CloudMedia: When Cloud on Demand Meets Video on Demand

CloudMedia: When Cloud on Demand Meets Video on Demand CloudMeda: When Cloud on Demand Meets Vdeo on Demand Yu Wu, Chuan Wu, Bo L, Xuanja Qu, Francs C.M. Lau Department of Computer Scence, The Unversty of Hong Kong, Emal: {ywu,cwu,xjqu,fcmlau}@cs.hku.hk Department

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

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

Modeling and Assessment Performance of OpenFlow-Based Network Control Plane

Modeling and Assessment Performance of OpenFlow-Based Network Control Plane ISSN (Onlne): 2319-7064 Index Coperncus Value (2013): 6.14 Ipact Factor (2013): 4.438 Modelng and Assessent Perforance of OpenFlo-Based Netork Control Plane Saer Salah Al_Yassn Assstant Teacher, Al_Maon

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

On File Delay Minimization for Content Uploading to Media Cloud via Collaborative Wireless Network

On File Delay Minimization for Content Uploading to Media Cloud via Collaborative Wireless Network On Fle Delay Mnmzaton for Content Uploadng to Meda Cloud va Collaboratve Wreless Network Ge Zhang and Yonggang Wen School of Computer Engneerng Nanyang Technologcal Unversty Sngapore Emal: {zh0001ge, ygwen}@ntu.edu.sg

More information

Online Advertisement, Optimization and Stochastic Networks

Online Advertisement, Optimization and Stochastic Networks Onlne Advertsement, Optmzaton and Stochastc Networks Bo (Rambo) Tan and R. Srkant Department of Electrcal and Computer Engneerng Unversty of Illnos at Urbana-Champagn Urbana, IL, USA 1 arxv:1009.0870v6

More information

Stochastic Games on a Multiple Access Channel

Stochastic Games on a Multiple Access Channel Stochastc Games on a Multple Access Channel Prashant N and Vnod Sharma Department of Electrcal Communcaton Engneerng Indan Insttute of Scence, Bangalore 560012, Inda Emal: prashant2406@gmal.com, vnod@ece.sc.ernet.n

More information

Naglaa Raga Said Assistant Professor of Operations. Egypt.

Naglaa Raga Said Assistant Professor of Operations. Egypt. Volue, Issue, Deceer ISSN: 77 8X Internatonal Journal of Adanced Research n Coputer Scence and Software Engneerng Research Paper Aalale onlne at: www.jarcsse.co Optal Control Theory Approach to Sole Constraned

More information

Retailers must constantly strive for excellence in operations; extremely narrow profit margins

Retailers must constantly strive for excellence in operations; extremely narrow profit margins Managng a Retaler s Shelf Space, Inventory, and Transportaton Gerard Cachon 300 SH/DH, The Wharton School, Unversty of Pennsylvana, Phladelpha, Pennsylvana 90 cachon@wharton.upenn.edu http://opm.wharton.upenn.edu/cachon/

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

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

An Adaptive Cross-layer Bandwidth Scheduling Strategy for the Speed-Sensitive Strategy in Hierarchical Cellular Networks

An Adaptive Cross-layer Bandwidth Scheduling Strategy for the Speed-Sensitive Strategy in Hierarchical Cellular Networks An Adaptve Cross-layer Bandwdth Schedulng Strategy for the Speed-Senstve Strategy n erarchcal Cellular Networks Jong-Shn Chen #1, Me-Wen #2 Department of Informaton and Communcaton Engneerng ChaoYang Unversty

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

CLoud computing technologies have enabled rapid

CLoud computing technologies have enabled rapid 1 Cost-Mnmzng Dynamc Mgraton of Content Dstrbuton Servces nto Hybrd Clouds Xuana Qu, Hongxng L, Chuan Wu, Zongpeng L and Francs C.M. Lau Department of Computer Scence, The Unversty of Hong Kong, Hong Kong,

More information

An Error Detecting and Tagging Framework for Reducing Data Entry Errors in Electronic Medical Records (EMR) System

An Error Detecting and Tagging Framework for Reducing Data Entry Errors in Electronic Medical Records (EMR) System 201 IEEE Internatonal Conference on Bonforatcs and Boedcne An Error Detectng and Taggng Fraework for Reducng Data Entry Errors n Electronc Medcal Records (EMR Syste Yuan Lng, Yuan An College of Coputng

More information

PSYCHOLOGICAL RESEARCH (PYC 304-C) Lecture 12

PSYCHOLOGICAL RESEARCH (PYC 304-C) Lecture 12 14 The Ch-squared dstrbuton PSYCHOLOGICAL RESEARCH (PYC 304-C) Lecture 1 If a normal varable X, havng mean µ and varance σ, s standardsed, the new varable Z has a mean 0 and varance 1. When ths standardsed

More information

Bandwdth Packng E. G. Coman, Jr. and A. L. Stolyar Bell Labs, Lucent Technologes Murray Hll, NJ 07974 fegc,stolyarg@research.bell-labs.com Abstract We model a server that allocates varyng amounts of bandwdth

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

How To Write A Powerpoint Powerpoint Commandbook For A Data Center

How To Write A Powerpoint Powerpoint Commandbook For A Data Center Internatonal Journal of Innovatve Research n Scence, Engneerng and Technology (An ISO 3297: 2007 ertfed Organzaton) VM Assgnent Algorth Based ost Effectve achng n loud outng Raya.R 1 Assstant Professor,Det

More information

Performance Analysis and Comparison of QoS Provisioning Mechanisms for CBR Traffic in Noisy IEEE 802.11e WLANs Environments

Performance Analysis and Comparison of QoS Provisioning Mechanisms for CBR Traffic in Noisy IEEE 802.11e WLANs Environments Tamkang Journal of Scence and Engneerng, Vol. 12, No. 2, pp. 143149 (2008) 143 Performance Analyss and Comparson of QoS Provsonng Mechansms for CBR Traffc n Nosy IEEE 802.11e WLANs Envronments Der-Junn

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

How To Make A Dstrbuted Ddos Attack More Successful

How To Make A Dstrbuted Ddos Attack More Successful PacketScore: Statstcs-based Overload Control aganst Dstrbuted Denal-of-Servce Attacks Yoohwan K * Wng Cheong Lau * Moo Choo Chuah ** H. Jonathan Chao EECS Departent Bell Labs CSE Departent ECE Departent

More information

Two-Phase Traceback of DDoS Attacks with Overlay Network

Two-Phase Traceback of DDoS Attacks with Overlay Network 4th Internatonal Conference on Sensors, Measureent and Intellgent Materals (ICSMIM 205) Two-Phase Traceback of DDoS Attacks wth Overlay Network Zahong Zhou, a, Jang Wang2, b and X Chen3, c -2 School of

More information

MAC Layer Service Time Distribution of a Fixed Priority Real Time Scheduler over 802.11

MAC Layer Service Time Distribution of a Fixed Priority Real Time Scheduler over 802.11 Internatonal Journal of Software Engneerng and Its Applcatons Vol., No., Aprl, 008 MAC Layer Servce Tme Dstrbuton of a Fxed Prorty Real Tme Scheduler over 80. Inès El Korb Ecole Natonale des Scences de

More information

Evolution of Internet Infrastructure in the 21 st century: The Role of Private Interconnection Agreements

Evolution of Internet Infrastructure in the 21 st century: The Role of Private Interconnection Agreements Evoluton of Internet Infrastructure n the 21 st century: The Role of Prvate Interconnecton Agreements Rajv Dewan*, Marshall Fremer, and Pavan Gundepud {dewan, fremer, gundepudpa}@ssb.rochester.edu Smon

More information

INTRODUCTION TO MERGERS AND ACQUISITIONS: FIRM DIVERSIFICATION

INTRODUCTION TO MERGERS AND ACQUISITIONS: FIRM DIVERSIFICATION XV. INTODUCTION TO MEGES AND ACQUISITIONS: FIM DIVESIFICATION In the ntroducton to Secton VII, t was noted that frs can acqure assets by ether undertakng nternally-generated new projects or by acqurng

More information

Value Driven Load Balancing

Value Driven Load Balancing Value Drven Load Balancng Sherwn Doroud a, Esa Hyytä b,1, Mor Harchol-Balter c,2 a Tepper School of Busness, Carnege Mellon Unversty, 5000 Forbes Ave., Pttsburgh, PA 15213 b Department of Communcatons

More information

A Design Method of High-availability and Low-optical-loss Optical Aggregation Network Architecture

A Design Method of High-availability and Low-optical-loss Optical Aggregation Network Architecture A Desgn Method of Hgh-avalablty and Low-optcal-loss Optcal Aggregaton Network Archtecture Takehro Sato, Kuntaka Ashzawa, Kazumasa Tokuhash, Dasuke Ish, Satoru Okamoto and Naoak Yamanaka Dept. of Informaton

More information

Elastic Systems for Static Balancing of Robot Arms

Elastic Systems for Static Balancing of Robot Arms . th World ongress n Mechans and Machne Scence, Guanajuato, Méco, 9- June, 0 _ lastc Sstes for Statc alancng of Robot rs I.Sonescu L. uptu Lucana Ionta I.Ion M. ne Poltehnca Unverst Poltehnca Unverst Poltehnca

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

A generalized hierarchical fair service curve algorithm for high network utilization and link-sharing

A generalized hierarchical fair service curve algorithm for high network utilization and link-sharing Computer Networks 43 (2003) 669 694 www.elsever.com/locate/comnet A generalzed herarchcal far servce curve algorthm for hgh network utlzaton and lnk-sharng Khyun Pyun *, Junehwa Song, Heung-Kyu Lee Department

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