On Distributed Computation Rate Optimization for Deploying Cloud Computing Programming Frameworks

Size: px
Start display at page:

Download "On Distributed Computation Rate Optimization for Deploying Cloud Computing Programming Frameworks"

Transcription

1 On Distributed omputation ate Optimization for Depoying oud omputing Programming Framewors Jia Liu athy H. Xia Ness B. Shroff Xiaodong Zhang Dept. of Eectrica and omputer Engineering Dept. of Integrated Systems Engineering Dept. of omputer Science and Engineering The Ohio State University, oumbus, OH 43210, U.S.A. iu, ABSTAT With the rapidy growing chaenges of big data anaytics, the need for efficient and distributed agorithms to optimize coud computing performances is unprecedentedy high. In this paper, we consider how to optimay depoy a coud computing programming framewor e.g., Mapeduce and Dryad over a given underying networ hardware infrastructure to maximize the end-to-end computation rate and minimize the overa computation and communication costs. The main contributions in this paper are three-fod: i we deveop a new networ fow mode with a generaized fowconservation aw to enabe a systematic design of distributed agorithms for computation rate utiity maximization probems UM in coud computing; ii based on the networ fow mode, we revea ey separabe properties of the dua functions of Probem UM, which further ead to a distributed agorithm design; and iii we offer important networing insights and meaningfu economic interpretations for the proposed agorithm and point out their connections to and distinctions from distributed agorithms design in traditiona data communications networs. This paper serves as an important first step towards the deveopment of a theoretica foundation for distributed computation anaytics in coud computing. 1. INTODUTION With the rapid advances in information technoogies, recent years have witnessed the growing chaenges in storing, processing, and anayzing arge data sets in many areas, such as socia networs web-services, genomic research, networ traffic management, compex physics simuations, environmenta research, just to name a few [1, 2]. Traditiona centraized reationa database management systems, first deveoped more than four decades earier, cannot hande the unstructured and dynamic nature of the arge data sets nowadays and the performance of reationa database management systems scaes poory as the data sets sizes increase. As a resut, distributed coud computing patforms that have a arge number of networed computing nodes with massivey parae structures have emerged as an attractive soution for handing big data anaytics. Among coud computing programming framewors for big data anaytics, perhaps the most famous exampe is the so- caed Mapeduce [3], designed initiay by Googe for scanning arge amounts of textua data to create web search indexes for the entire Internet. Another notabe exampe is Dryad [4], which is Microsoft s counterpart to Mapeduce. Dryad has been seen by some researchers as an approach to improve the pitfas of the Mapeduce framewor [5] in that: i it aows for genera styes of computation by using directed-acycic graph DAG that are much more than just map and reduce phases; and ii it aows communications between stages to happen over more than just fies stored in distributed fie systems. Furthermore, recent research showed that Mapeduce and Dryad can be mathematicay unified under a we-defined matrix-based mode [6]. However, as promising as they are, the actua depoyments of coud computing framewors such as Mapeduce and Dryad remain in their infancy and there are many technica issues to be resoved. For exampe, in the most popuar Mapeduce impementationown as the Hadoop project [7], there is ony one active job tracer to schedue and monitor a map and reduce tass. This not ony poses the singe-point-of-faiure vunerabiity, but aso has a poor scaabiity that defeats the whoe purpose of distributedness in coud computing. To date, athough there exist some heuristic design rue-of-thumbs e.g., moving computation nodes coser to their data sources to avoid unnecessary communications, itte effort has been made to estabish a mathematica foundation to systematicay deveop distributed agorithms and contro schemes that optimize the depoyments and performances of coud computing programming framewors. Hence, the goa of our paper is to tae the first step to fi this gap. In this paper, we study how to optimay decompose and aocate the subcomputation tass in a coud computing programming framewor over an underying hardware infrastructure, such that the end-to-end computation rate can be maximized and the overa computation and communication costs can be minimized. Further, agorithms for soving this probem need to be impemented in a distributed fashion. Toward this end, motivated by Dryad which aso subsumes Mapeduce as pointed out earier, we mode a coud computing programming framewor as computing a set of generic functions Θ } that share a common set of distributed incoming data sources, where the data dependency can be represented by a mutipe-input mutipe-output directed acycic graph MIMO-DAG, as shown in the ius- opyright is hed by author/owners. Performance Evauation eview, Vo. 40, No. 4, March

2 χ 1 χ 2 χ 3 χ 4 g 1 λ e 1 χ 1 + χ 2 g 3 g 4 g 2 e 4 e 2 λ e 5 λ e λ 6 λ e λ 3 e 9 e e 10 7 λ λ e 8 λ λ λ e 11 λ e 12 λ e 13 g 10 g 11 Θ 1 =χ 1 + χ 2 χ 2 + χ 3 Θ 2 = χ 1 +3χ 2 + χ 3 + χ 4 a An iustrative exampe of a coud computing programming framewor n1 n2 n3 n4 n5 n6 n7 n8 n9 n10 n11 n12 n13 n14 n15 n16 b A 16-node 2-D torus interconnection topoogy. Figure 1: An iustrative exampe of depoying a coud computing programming framewor over an underying networ infrastructure: a omputing functions Θ 1 and Θ 2 at computation rate λ; the data dependencies are represented by a MIMO- DAG structure; b An iustration of a 2-D torus interconnection topoogy commony seen in arge data centers or supercomputers. trated exampe in Fig. 1a. 1 On the other hand, the underying networed system upon which any coud computing framewor needs to be depoyed can aso be represented by a graph. For exampe, Fig. 1b iustrates a 2-D torus interconnection topoogy commony found in arge data centers or supercomputers that support coud computing jobs. As wi be shown ater, to capture the ey features of coud computing programming framewors and incorporate computation and communication constraints of the underying networs, we deveop a generic mathematica modeing framewor, based on which we formuate the computation rate utiity maximization probem UM. Our main resuts and contributions in this paper are three-fod: By accounting for in-networ data aggregation fows, we show that one can construct a new networ fow mode with a generaized fow-conservation aw to address both 1 The addition and mutipication operations in Fig. 1a are ony for iustrative purposes. Each vertex coud be any genera data processing operations. communication imits and computation costs in Probem UM. We point out that this utiity framewor and the associated generaized fow-conservation aw are important in that they shares the same separabe structure as in the cassica networ utiity maximization probems in data communication networs [8, 9]. Thus, they enabe the design of poynomia-time distributed agorithms for coud computing by drawing experiences gained from the cassica networ utiity maximization theory. By appropriatey reformuating Probem UM in the Lagrangian dua domain, we revea ey separabe properties of the dua function of Probem UM. These ey structura properties enabe us to obtain cosed-form expressions for the prima and dua update schemes, which further eads to a fuy distributed agorithmic impementation for big data anaytics in coud computing. We offer important networing insights and economics interpretations for the proposed distributed agorithm. We aso point out the connections to and distinctions from the dua decomposition based distributed agorithms in the cassica networ utiity maximization theory for data communications networs, thus further advancing our understanding of distributed approaches in coud computing networ optimization theory. We aso provide numerica exampes to show the efficacy of our proposed distributed agorithm. To our nowedge, this paper is among the first that treat the design of distributed agorithms in MIMO-DAG based coud computing framewors e.g., Dryad via a rigorous theoretica networ-fow optimization approach. The remainder of this paper is organized as foows. In Section 2, we review some reated wor in the iterature, putting our wor in a comparative perspective. Section 3 introduces the new networ fow mode with a generaized fow-conservation aw for depoying coud computing framewors. Section 4 deveops the principa components of our proposed distributed agorithm for soving Probem UM. Section 5 provides numerica resuts and Section 6 concudes this paper. 2. ELATED WOK Our wor is cosey reated to i distributed cross-ayer utiity maximization theory for data communication networs see, e.g., [10] for an overview and ii in-networ computation techniques. In the in-networ computation iterature, our wor is most reated to [11], where Shah et a. deveoped a networ fow mode for the in-networ computation in sensor networ appications. The mode in [11] extends the conventiona fow-conservation aw in the networ fow iterature [12] to in-networ computation appications. However, the mode in [11] is restricted to simpe tree topoogies that are used for data aggregation in sensor networs. In contrast, our networ fow mode wors with generic MIMO-DAG, which are the most appropriate modes for compex coud computing programming framewors, such as Dryad and Mapeduce. Aso, in our networ mode, we incorporate genera networ utiity and communication/computation cost functions, which were not considered in [11]. Our networ mode aso shares some simiarities with the oad shedding and distributed resource contro probem of stream processing networs e.g., [13 15]. But our wor 64 Performance Evauation eview, Vo. 40, No. 4, March 2013

3 differs from these wors in the foowing two important aspects: First, athough the issue of fow imbaance in stream processing networs was aso pointed out in [14,15], the fow imbaance in [14,15] was caused by different fow production and consumption rates between upstream and downstream nodes. In contrast, the fow imbaance in this wor is due to subcomputation in couds, which is a fundamentay different cause. Second, in [13], the tas-to-server assignment reationship is assumed to be given and the authors ony studied the end-to-end utiity rate maximization. In contrast, the tas-to-sever assignment reationship is aso part of the overa optimization in our wor. Our networ fow mode aso has connections with the graph embedding probems in graph theory e.g., [16 18]. But these probems differ from ours in that their embedding objectives were to minimize some graph-theoretic performance metrics, such as diation i.e., the maximum distance in the networ between adjacent tree nodes. 3. NETWOK MODEL AND POBLEM FO- MULATION In this section, we first present the modeing detais of coud computing programming framewors and the underying networ infrastructure in Section 3.1. Then, the concept of mapping between a programming framewor and networ infrastructure is introduced in Section 3.2. Next, we deveop a new networ fow mode with a generaized fow-conservation aw in Section 3.3. Finay, based on the networ fow mode, we formuate the computation rate optimization probem in Section Modeing oud omputing Framewors and Networ Infrastructure As mentioned earier, we mode a coud computing programming framewor by a MIMO-DAG. Here, we denote a MIMO-DAG by D = V, E}, where V and E represent the sets of vertices and edges, respectivey, with V = V and E = E. We note that this MIMO-DAG structure captures the parae processing and muti-stage aggregation nature of typica coud computing programming framewors, such as Dryad or mutipe concatenations of Map/educe. We suppose that there are S vertices of in-degree-zero in D, corresponding to the S distributed input data sources. Without oss of generaity w..o.g, we abe these source vertices as g 1,...,g S. For modeing convenience, there are K vertices in D having in-degree-one and out-degree-zero, corresponding to the virtua sins that absorb the fina output of Θ, =1,...,K. Again, w..o.g, we abe such sins as g V K+1,...,g V. The remaining nodes g S+1,...,g V 1 perform the computations prescribed by Θ. The input data stream at each source g i is denoted as χ i } =0, where χ i represents the i-th input eementattime instant. The infinite input data streams coud represent, for exampe, the big data phenomenon exempified in the arge data sets processing in by the Mapeduce or Dryad framewors in coud computing. In this paper, a sources are assumed to be synchronized. We note that the synchronization in coud computing systems is aso an activey research topic see, e.g., [19] and references therein and its detais are beyond the scope of this paper. For simpicity, we assume that a input data streams and a directed edges in the MIMO- DAG structure have a homogeneous input rate λ, which can aso be thought of as the end-to-end computation rate of Θ }. We remar that the cases with heterogeneous rates due to compression or expansion processing can be exercised simiary by introducing coefficients in front of the data rates [14] in our utiity maximization formuation described ater. One can further transform such cases into a homogeneous case using modeing techniques in [13] by changing the measure and absorbing the coefficient into the resource usage parameters. Hence, this homogeneous rate assumption does not ose any generaity. We et he and te denote the head and tai vertices of each directed edge e in E, respectivey. We et S e E he g 1,...,g S}} be the set of a edges originating from the sources. We note that S S in genera. Liewise, we et K e E te g V K+1,...,g V }} denote the set of edges terminated at the sins. Note that, since each sin edge represents a unique output see Fig. 1a, we have K = K. W..o.g, we abe the edges in E in such a way that the first S edges e 1,...,e S and the ast K edges e +1,...,e E are the source and sin edges, respectivey. In Fig. 1a, for exampe, S = e 1,...,e 6 } and K = e 12,e 13 }. We point out that each directed edge in D can be viewed as a subcomputation tas. For exampe, in Fig. 1a, edge e 7 corresponds to the computation resut χ 1 + χ 2. Therefore, the terms edge and subcomputation are sometimes used interchangeaby in this paper. It shoud be noted, however, that a subcomputation does not necessariy correspond to a unique edge. For exampe, in Fig. 1a, edges e 7 and e 8 a carry the same subcomputation χ 1+χ 2. The set of successor edges of e in D is defined as Ψe e E te =he}. For exampe, in Fig. 1a, the successor edges of e 2 are e 7 and e 8. Note that from the structure of D, wehaveψe i= for a e i K. Finay, we remar that the MIMO-DAG structure degenerates into a tree topoogy if it has a singe sin and each edge ony has one successor edge. Therefore, the tree-structure mode considered in [11] can be viewed as a specia case of our mode. On the other hand, in coud computing environments e.g., data centers or supercomputers, we have a networed system as the physica computing and communication infrastructure i.e., a coud computing hardware patform. Usuay, the hardware patform aso exhibits certain carefuy constructed parae structures. For exampe, Fig. 1b iustrates a 2-D torus interconnection topoogy that is commony seen in coud hardware patforms. Advanced supercomputers nowadays coud empoy torus interconnections with even higher dimensions e.g., 5-D torus in IBM Bue Gene/Q architecture. But we point out that these specia topoogy properties are not essentia to our networ modes and our distributed agorithm design, both of which appy to genera interconnection topoogies. In this paper, we aso mode the underying coud computing hardware patform by a graph G = N, L}, where N and L are the sets of nodes and ins with N = N and L = L, respectivey. Depending on the size and scae of the hardware system, each node in N coud represent, e.g., a PU core, a node card, or even a computer rac. Each in in L represents the communication connection between the nodes. It coud represent a oca high-speed bus if the nodes are co-ocated on the same card or a fiber in if the nodes are ocated at different racs. As mentioned earier, to avoid unnecessary data commu- Performance Evauation eview, Vo. 40, No. 4, March

4 nications in coud computing modes such as Map/educe and Dryad, it is highy desirabe to aocate subcomputation tass invoving input data sources and outputs at their physica ocations. Therefore, w..o.g, we abe the nodes of N as n 1,...,n N in such a way that the first S nodes n 1,...,n S correspond to the physica ocations of the S data sources of D. 3.2 Mapping a oud omputing Framewor onto a Networ Infrastructure eary, depoying a given coud computing framewor on a coud computing hardware patform amounts to mapping a edges in D onto G in an appropriate fashion. As in standard graph theory terms, we define a path P as a sequence of nodes in N such that every two adjacent nodes and n in P satisfy,n L. The first and the ast nodes in P are caed the start and end nodes and are denoted as αp and βp, respectivey. In the extreme case, P coud contain ony one node, say. In this case, αp =βp = and the in, degenerates into a sef-oop. Here, if an edge is mapped to a path P,itmeans that the data are transmitted from αp via the specified ins to βp and then the corresponding subcomputation is performed at βp. In the specia case, if an edge is mapped to a sef-oop, then the corresponding subcomputation tas is performed ocay and no communication is needed. We denote the set of a paths in G as P. Avaid mapping of D onto G is defined as foows: Definition 1. A mapping M : E P is vaid if: 1 αme i = if e i S and he i = g, = 1,...,S; 2 βme i = if e i Kand is the physica output ocation of e i ; and 3 αme j = βme i if e j Ψe i. In Definition 1, conditions 1 and 2 impy that the source and the sin nodes in D have to match to their physica ocations in the underying networ, and condition 3 represents that the ogica successor reationships of the edges in D need to be respected after the mapping. In genera, there are many different ways to map D onto G. For exampe, it is not difficut to see that Figs. 2a and 2b iustrate two vaid mappings of the MIMO-DAG structure in Fig. 1a onto the networ in Fig. 1b. Note that in a vaid mapping, the node sequence corresponding to an edge in D coud consist of mutipe connected ins in G. For exampe, in Figure 2b, edge e 8 consists of ins n 5,n 9, n 9,n 10, and n 10,n 11. We note that each mapping may use a different computation rate, as shown in Figs. 2a and 2b. For exampe, in Fig. 2a, the abe e i : 2 represents that the computation rate of each edge e i is 2 in this mapping. Further, a timesharing among vaid mappings aso yieds a vaid mapping. For instance, Fig. 3 shows a time-sharing between the two mappings in Figs. 2a and 2b, each accounting for 50% of time. It is not difficut to see that the computation rate of this time-sharing mapping is = A Networ Fow Mode with a Generaized Fow-onservation Law It can be seen from the above discussions that the computation rate region of a coud computing frameworovera n1 n5 given networ infrastructure can be exhausted by a timesharing strategies among a possibe vaid mappings. As a resut, any performance metrics reated to the compue1:2 e 7 :2 e2:2 e8:2 n2 n6 e 12 :2 e 5 :2 e3:2 e9:2 e10:2 e13:2 e4:2 e 11 :2 e6:2 n9 n10 n11 n12 n13 n14 n15 n16 a Mapping 1 with computation rate λ =2. n 1 n 2 n 3 n 4 e1:3 e2:3 e3:3 e7:3 e 8 :3 e4:3 e9:3 n3 n7 e5:3 e6:3 n5 n6 n7 n8 e 12 :3 e 10 :3 e11:3 n9 n10 n11 n12 n13 n14 n15 e 13 :3 n4 n8 n16 b Mapping 2 with computation rate λ =3. Figure 2: Two vaid mappings of the MIMO-DAG structure in Fig. 1a onto the networ in Fig. 1b. Each mapping has a different computation rate. tation rate can be optimized by determining an optima time-sharing strategy. However, since the tota number of vaid mappings is exponentia in N, finding an optima timesharing strategy directy is intractabe. Therefore, we need to expoit additiona structure of the networed system to address this chaenge. To this end, our basic idea is to deveop a new networ fow mode with a generaized fow-conservation aw for computation anaytics. The fundamenta rationae behind this approach is that, simiar to the cassica mode in the muticommodity networ fow iterature [12], the structura properties of the new networ fow mode woud aso ead to poynomia time soutions if designed appropriatey. Unfortunatey, deveoping a fow-conserved networ fow mode for computation anaytics is not-trivia. Due to performing computations at each node in the networ, the cassica fow-conservation aw [12] fais to hod in computation anaytics networ fows. For exampe, from the zoom-in view for node n 10 in Fig. 4, it can be seen that the tota incoming fow rate is: x e 7 n 9,n 10 + xe 8 n 9,n 10 + xe 9 n 10,n 11 + xe 9 n 10,n 11 =5. However, due to the addition computations performed at 66 Performance Evauation eview, Vo. 40, No. 4, March 2013

5 n1 n2 n3 n4 n5 n6 n7 n8 n9 n13 e2:1 e 1 :1 e1:1.5 e7:1.5 e7:1 e 8 :1.5 e12:1.5 e8:1 e2:1.5 n10 n14 e 5 :1 e4:1 e3:1 e9:1 e10:1.5 e6:1 e3:1.5 e6:1.5 e 11 :1 e4:1.5 e 12 :1 e9:1.5 e9:1 n11 n15 e13:1 e11:1.5 e10:1.5 e13:1.5 Figure 3: A time-sharing between two mappings in Fig. 2a and Fig. 2b, each accounting for 50% of time. e9:1.5 e7:1 e8:1.5 in n10,n9 in n 9,n 10 n 10 e12:1 in n 10,n 14 n12 n16 in n11,n10 in n10,n11 e9:1.5 e9:1 e8:1.5 Figure 4: Zoom-in view of node n 10 in Fig. 3. Due to the addition computations, the tota incoming and outgoing fow rates at n 10 are not equa. node n 10, the tota outgoing fow rate is: x e 9 n 10,n 9 + xe 8 n 10,n 11 + xe 12 n 10,n 14 =4. This shows that, for computation anaytics in coud computing, the tota input and output networ fow are unbaanced. Athough the traditiona fow-conservation aw no onger hods here, upon a coser oo at each incoming edge and their corresponding outgoing or successor edges at n 7,itis not difficut to observe another form of fow-conservation aw. As shown in Fig. 4, for the incoming edge e 4 that does not invove any computation at n 7,wehave: x e 9 n 11,n 10 = xe 9 n 10,n 9, i.e., the incoming rate equas the outgoing rate. The same is aso true for the incoming edge e 8. On the other hand, for the incoming edge e 7 that invoves in the addition with incoming edge e 9,wehave: x e 7 n 9,n 10 = xe 12 n 10,n 14 and x e 9 n 11,n 10 = xe 12 n 10,n 14, i.e., the outgoing rate of the successor edge of e 7 is equa to the incoming rate of e 7. To mode this new form of fow-conservation aw, it is important to recognize that: for a subcomputation, say e, that is injecting to node n, a portion of its fow is terminated at n to compute the successor edges of e i.e., Ψe. For convenience, we abe the ins in the physica networ as 1,...,L instead of using node pairs n j,. We et x e 0 represent the fow amount of subcomputation tas e on in. In a vaid mappings M, since the start node of a source edge e i Sand the end node of a sin edge e i Kare aways mapped to their respective physica nodes in the networ, we et Srce i αme i and Dste i βme i be the physica source and end nodes of e i and e i for simpicity. Then, based on the previous observations, we have the foowing resut: Lemma 1. When computation anaytics are performed over an underying networed system, the foowing generaized fow-conservation aw hods: O O O O I I = λ, e i S, e j Ψe i, = Srce i, 1 =0, e i E\K, e j Ψe i,and Srce i if e i S, 2 =0, e i K, Dste i, I I = λ, 3 e i K, = Dste i, 4 where O and I represent the sets of outgoing and incoming ins at node, respectivey; and y e i represents the subcomputation e i generated at node. Here, the equaities in 1 and 2 represent that the tota incoming fow rate of any non-sin edge e i at node shoud be equa to the sum of the outgoing fow rates of e i and its successor edge e j, and this reationship hods for every successor edge of e i. Moreover, Eq.1 states that if e i is a source edge and happens to be its source node, then the net injection rate of e i at is λ. On the other hand, Eqs.3 and 4 say that, for a sin edge e i that does not have any successor edge, the net output fow rate shoud be equa to λ at its physica ocation and zero esewhere. 3.4 Probem Formuation We et denote the capacity of in. Since the tota networ fow traversing a in cannot exceed the in s capacity, we have E xe i, =1,...,L. We et the networ utiity be a function of λ, denoted by Uλ : +. We assume that Uλ is concave, monotonicay increasing, and twice continuousy differentiabe. The concavity of U represents the diminishing returns effect. When U is inear as a specia case of concavity, we are simpy maximizing the computation rate itsef. In the physica networ, each outgoing edge at a node represents a subcomputation see, e.g., edges e 9, e 10, and e 13 in Fig. 4, which incurs certain costs e.g., consumed energy per unit amountofcomputation. We et ρ represent the unit cost for performing computation at node. Then, the tota cost due to computation at node can be computed Performance Evauation eview, Vo. 40, No. 4, March

6 as: S ρ λ + y e i + i= S +1 ρ I y e i + I O + O. 5 In 5, the term λ + y e i + I xe i O xe i is the tota fow rate of source edge e i that is terminated at node and used to compute its successor edges. Liewise, the term y e i + I xe i O xe i represents the tota fow rate of non-source and non-sin edge e i terminated at. Our objective is to maximize the net networ utiity, defined as networ utiity minus the tota costs. Putting together the discussions earier, we can formuate the probem as foows: UM: Max Uλ O S N =1 ρ + i= S +1 λ + y e i ρ y e i + + I I O s.t. Eqs. 1, 2, 3, 4, E, =1,...,L, 0, i, ; λ 0. Now, it is not difficut to recognize that with the proposed networ fow mode in Lemma 1, Probem UM is a convex program. Moreover, the nice separabe structure of the objective function enabes the design of distributed agorithm to sove Probem UM, which wi be the focus of the next section. 4. DISTIBUTED SOLUTION POEDUE In this section, we wi present the ey steps in designing a distributed agorithm based on dua decomposition to sove Probem UM. In Section 4.1, we wi first reformuate Probem UM into its Lagrangian dua probem and show how to appropriatey decompose the Lagrangian dua probem. Based on these resuts, we introduce the basic idea in designing a distributed agorithm in Section 4.2. In Section 4.3, we offer some interesting networing and economics interpretations of the proposed distributed agorithm. Then, we wi present some numerica resuts in Section Lagrangian Dua eformuation and Decomposition As mentioned earier, since Probem UM is a convex program, it can be equivaenty soved in its Lagrangian dua domain because of a zero duaity gap [20]. To sove the Probem UM in its Lagrangian dua domain, we first sighty modify the constraints in 1 4 into inequaity constraints as foows: O O O O I I λ, e i S, e j Ψe i, = Srce i, 6 0, e i E\K, e j Ψe i, and Srce iife i S, 7 0, e i K, Dste i, I I λ, 8 e i K, = Dste i, 9 It is not difficut to show that these modifications do not affect the soution at optimaity. Interestingy, these modifications can aso be interpreted from a networ stabiity perspective: the tota service rate at each node is no ess than the tota arriva rate. Next, we associate dua variabes 0 and 0 for each constraint in 6 7 and 8 9, respectivey. For notationa simpicity, we use Ψ i to represent the index set j : e j Ψe i}. Aso, we use vectors x, μ, and w to group a x-, μ- and w-variabes. By accommodating the constraints into the objective function and combining reated terms, we have that the Lagrangian can be written as foows: Lλ, x, μ, w =Uλ O + + E + N j Ψ i =1 E N i= S +1 i= 1 =1 ρ S N =1 y e i O O S + λ Srci + λ j Ψ i ρ + λ + y e i I E i=+1 I + I O I Dsti. 10 Then, the Lagrangian dua function can be written as: } E Φμ, w = sup Lλ, x, μ, w xe i,, λ,x x e.11 i 0, i, ; λ 0. Finay, the dua probem can be written as: D: Minimize Φu 12 subject to u 0. Next, it is important to recognize that the Lagrangian function in 11 possesses a decomposabe structure, which eads to a distributed computation scheme. More specificay, by appropriatey switching summation orders and rearranging terms in 11, we have the foowing resut see Appendix A for proof detais: 68 Performance Evauation eview, Vo. 40, No. 4, March 2013

7 Proposition 2. The Lagrangian in 11 can be decomposed in a source-wise and in-wise fashion as foows: Φμ, w =Φ Fμ, w+ L =1 Φ μ, w+ N =1 Φ μ, w, where Φ Fμ, w, Φ μ, w, andφn μ, w represent the fow contro, per-in routing, and per-node computation subprobems at each source, each in, and each node, respectivey. Here, Φ Fμ, w and Φ μ, w are defined as foows, respectivey: Φ F μ, w max Uλ λ 0 [ S λ j Ψ i Φ μ, w max s.t. N Srci + ρ E =1 E i=+1 Tx wi x } E, 0, i., w Dsti]} ; where, =1,...,N, i =1,...,E, are defined as: ρ + j Ψ i, i =1,...,E K, Φ μ, w max y e i + j Ψ i, i = E K +1,...,E; 0, i 13 E [ y e j ŵ i ye i ]}, j Ψ i where ŵ i, =1,...,N, i =1,...,E, are defined as: ρ, i =1,...,E K,, i = E K +1,...,E. 14 Based on Proposition 2, the Lagrangian dua probem D in 12 can be transformed into the foowing master dua probem: MD: Minimize Φ Fμ, w+ L =1 Φ μ, w+ N =1 Φ μ, w subject to μ 0, w 0. Then, the tas of soving the Lagrangian dua probem D bois down to distributedy soving the subprobemsφ Fμ, w, Φ μ, w, and Φ μ, w, and then the master dua probem MD. 4.2 Design of Distributed Agorithm Note that at each source node, the fow contro and computation subprobems Φ F has a concave objective function with a singe non-negative decision variabe. Thus, Φ F can be triviay and efficienty soved e.g., by simpe bisection search method. Aso, it can be observed that the routing subprobem Φ at each in is a ower dimensiona inear programming probem with a singe constraint, OEV 2 variabes, and a probem coefficients are ocay avaiabe, which means that it can be efficienty soved as Agorithm 1 A subgradient agorithm for soving Probem UM. Initiaization: 1. hoose initia starting points μ 0 and w 0. Let m =0. Main Iteration: 2. ompute the source computation rate λm by soving the fow contro subprobem Φ Fμ m, w m by using, e.g., bisection method. 3. ompute the routing decisions m at each in by soving the routing subprobem Φ μ m, w m, a inear programming probem. 4. hoose an appropriate step size π m.ompute the subgradients d ij μ, m and di w, m using 17 and 18 with λm and m. 5. Update dua variabes μ m+1 and w m+1 with d ij μ, m and d i w, m. 6. If μ m+1 μ m <ɛand w m+1 w m <ɛ, then return λm and m. Otherwise, et + 1 and go to Step 2. we. For the master dua probem MD, since its objective function is piece-wise differentiabe, one can appy subgradient method [20]. More specificay, starting with an initia μ 0 and w 0 and after evauating Φ F μ m, w m and Φ μm, w m in the m-th iteration, we update the dua variabes as foows: m +1=max m +1=maxwi m π m d ij μ, m, 0}, 15 m πm d i w, m, 0}, 16 where π m > 0 denotes a positive scaar step size, and d ij μ, m and d ij w, m represent the subgradients of the dua variabes and in the m-th iteration, respectivey. It is nown that for the subgradient iterative schemes in 15 and 16 to converge, a sufficient condition is that the step size π m satisfies π m 0asm, m=0 =, and m=0 πm 2 [20]. A popuar step size seection strategy is the divergent harmonic series: π m = β, m =1, 2,..., m where β is some given system parameter. For the master dua probem MD, the subgradient for the Lagrangian dua function in 10 can be computed as: d ij μ, m = O O I λ1 ei S, =Srce i }, 17 d i w, m = I + λ1 ei S, =Dste i }, 18 where 1 } represents the indicator function, which equas 1 if the condition in } is satisfied and 0 otherwise. To summarize, the design of distributed subgradient agorithm for soving Probem UM is iustrated in Agorithm Networing and Economics Interpretations Severa interesting networing and economics insights for the subgradient-based distributed agorithm are in order. Queue ength interpretations of the dua variabes: Performance Evauation eview, Vo. 40, No. 4, March

8 It can be seen that by dividing the step size π m on both sides of 15 and 16 and etting Q ij m /π m and Q i m wi /πm,wehave: Q ij m +1=max Q ij m y e j + y e i + I O } + λ1 ei S, =Srce i }, 0, 19 Q i m +1=max Q i m λ1 ei S, =Dste i } + I O } + y e i, A coser oo reveas that 19 and 20 behave exacty the same as queue ength evoution as seen in traditiona data communications networs. Indeed, if we et Q ij m represent the queuing bacog for computing the successive edge j of edge i at node in time instant m, then this queuing bacog and the dua variabe m are intimatey reated differ ony by a scaing factor π m. By the same toen, Qi m can be simiary interpreted as the queuing bacog of the terminating edges e i Kat node m. onnections to the bac-pressure agorithm: eary, in cases where in-networ computation is disabed, i.e., by forcing x e j = 0 for a e j Ψe i and for a, the networ degenerates into a traditiona communication networ. As expected, it is easy to chec that the queue ength evoutions in 19 and 20 reduce to the conventiona queue ength evoution: Q i m +1=max Q i m O + I } + λ 1 ei S, =Srce i } 1 ei S, =Dste i }, Moreover, since a dua μ-variabes become 0, it is easy to verify that the computation subprobem Φ admits a trivia optima soution: y e i =0,, i. Aso, with dua μ-variabes being 0, it is not difficut to see that the routing subprobem Φ w admits a trivia soution as foows: for a given in, pic a subcomputation tas e i that has the argest Tx wi x -vaue, say e i, and et e i use up the in capacity. This is exacty the same strategy used in the ceebrated bac-pressure agorithm that was first discovered in [21]. Pricing interpretation of the dua subgradient updating scheme: The dua variabes m and m can aso be economicay interpreted as congestion prices at node during the m-iteration. The dua updating scheme of the subgradient agorithm can then be viewed as a pricing scheme. When node becomes increasingy congested, then d ij μ, < 0. From 15, we can see the price of node wi be increased. On the other hand, when node becomes ess congested, then d ij μ, > 0. Again, from 16, it can be sen that the price wi be decreased. Fow ate North Out South Out East Out West Out Sef omp Node ID Figure 5: Outgoing ins and sef-oop fow rates at each node for edge e 12. Fow ate North Out South Out East Out West Out Sef omp Node ID Figure 6: Outgoing ins and sef-oop fow rates at each node for edge e NUMEIAL ESULTS In this section, we use Fig. 1 as an exampe to iustrate our proposed networ fow mode and distributed soution procedure. That is, we want to optimize the depoyment of the MIMO-DAG computation framewor in Fig. 1a onto the 16-node 2-D torus interconneced networ in Fig. 1b. The physica ocations of both Θ 1 and Θ 2 are at node n 16. Here, we et the capacity of each in in Fig. 1b be 10 and et the per-node unit computation cost be Our objective is to maximize the computation rate, i.e., etting Uλ =λ. After optimization, the maximum computation rate is Due to space imitation and the arge number of optima in fow rate variabes for this exampe = 1040 variabes, we ony pot in Figs. 5 and 6 the outgoing ins and sef-oops fow rates for edges e 12 and e 13 to iustrate part of the optima soution. The North Out, South Out, East Out, and West Out in Figs. 5 and 6 represent the outgoing ins at each node in Fig. 1b aong the specified directions, respectivey. eca from Fig. 1a that edges e 12 and e 13 are sin edges, which correspond to the fina resuts of functions Θ 1 and Θ 2, respectivey. Surprisingy, it can be seen that the computations of these sin edges are not depoyed cose to the physica output node n 16. The majority of the computations of e 12 and e 13 are done at node 2 see the rates of sef-oops at node 2 and node 1. This shows that the heuristic proximity depoyment rue may not be optima. From Figs. 5 and 6, it is aso not difficut to see the optima routing paths for edges e 12 and 70 Performance Evauation eview, Vo. 40, No. 4, March 2013

9 Normaized ompuation ate Normaized Noda omputation ost Figure 7: The change of maximum end-to-end computation rate with respect to the change of per-node unit computation cost. Normaized ompuation ate Normaized Lin apacity Figure 8: The change of maximum end-to-end computation rate with respect to the change of in capacity. e 13. For exampe, the optima routing paths for edge e 13 are: n W 1 n N 4 n 16, n E 2 n E 3 n N 4 n 16, n 2 S n 6 E n 7 E n 8 S n 12 S n 16, n 2 W n 1 W n 4 N n 16, n N 2 n E 14 n E 15 n 16, n S 12 n 16, where the etter above each arrow denotes the routing direction at that hop. Next, we iustrate in Figs. 7 and 8 the changes of maximum end-to-end computation rate with respect to the changes of per-node unit computation cost and in capacity, respectivey. In Fig. 7, we can see as expected that the end-to-end computation rate decreases as the unit computation cost at each node increases from 1 to 10 units step size is Liewise, we can see from Fig. 8 the end-to-end computation rate increases as the in capacity increases from 10 to ONLUSION In this paper, we investigated the design of distributed agorithms for coud computing programming framewors depoyments. We formuated the computation rate utiity maximization probem UM by deveoping a new networ fow mode with a generaized fow-conservation aw. Based on this enabing framewor, we deveoped a dua decomposition based distributed agorithm to sove Probem UM. We provided important networing interpretations and ey impementation insights for our proposed agorithm and pointed out the connections and distinctions to distributed agorithms design in traditiona data communications networs. oectivey, these resuts serve as the first buiding boc of a new theoretica framewor for the depoyment of coud computing programming framewors. 7. EFEENES [1] A. S. Szaay, Extreme data-intensive scientific computing, IEEE omputing in Science and Engineering, vo. 13, no. 6, pp , Nov [2] Data, data everywhere, Economist, [Onine]. Avaiabe: [3] J. Dean and S. Ghemawat, Mapeduce: Simpified data processing on arge custers, in Proc. USENIX OSDI, San Francisco, A, Dec. 6-8, 2004, pp [4] M. Isard, M. Budiu, Y. Yu, A. Birre, and D. Fettery, Dryad: Distributed data-parae programs from sequentia buiding bocs, in Proc. AM SIGOPS/Eurosys, Lisboa, Portuga, Mar , 2007, pp [5] K.-H. Lee, Y.-J. Lee, H. hoi, Y. D. hung, and B. Moon, Parae data processing with Mapeduce: Asurvey, AM SIGMOD ecord, vo. 40, no. 4, pp , Dec [6] Y. Huai,. Lee, S. Zhang,. H. Xia, and X. Zhang, DOT: A matrix mode for anayzing, optimizing and depoying software for big data anaytics in distributed systems, in Proc. AM SO, ascais, Portuga, Oct , [7] Hadoop. [8] M. hiang, S. H. Low, A.. aderban, and J.. Doye, Layering as optimization decomposition: A mathematica theory of networ architecture, Proc. IEEE, vo. 95, no. 1, pp , Jan [9] F. P. Key, A. K. Mauo, and D. K. H. Tan, ate contro in communications networs: Shadow prices, proportiona fairness and stabiity, Journa of the Operationa esearch Society, vo. 49, pp , [10] X. Lin, N. B. Shroff, and. Sriant, A tutoria on cross-ayer optimization in wireess networs, IEEE J. Se. Areas ommun., vo. 24, no. 8, pp , Aug [11] V. Shah, B. K. Dey, and D. Manjunath, Networ fows for functions, in Proc. IEEE Internationa Symposium on Information Theory ISIT, St. Petersburg, ussia, Ju.31 Aug.5, 2011, pp [12] M. S. Bazaraa, J. J. Jarvis, and H. D. Sherai, Linear Programming and Networ Fows, 4th ed. New Yor: John Wiey & Sons Inc., [13] H. Feng, Z. Liu,. Xia, and L. Zhang, Load shedding and distributed resource contro of stream processing networs, Performance Evauation, vo. 64, no. 9-12, pp , Oct [14] H. Zhao,. H. Xia, Z. Liu, and D. Towsey, A unified modeing framewor for distributed resource aocation of genera for and join processing networs, in Proc. AM Sigmetrics, New Yor, NY, Jun , [15] Z. Liu, A. Tang,. H. Xia, and L. Zhang, A decentraized contro mechanism for stream processing Performance Evauation eview, Vo. 40, No. 4, March

10 networs, Annas of Operations esearch, vo. 170, no. 1, pp , Sep [16] F. T. Leighton, M. J. Newman, A. G. anade, and E. J. Schwabe, Dynamic tree embeddings in butterfies and hypercubes, SIAM Journa of omputing, vo. 21, no. 4, pp , Aug [17] O. Wohmuth and F. Mayer-Lindenberg, A method for the embedding of arbitrary trees into hypercubes, in Proc. AM Symposium on Appied omputing, 1998, pp [18] V. Heun and E. W. Mayr, Efficient dynamic embeddings of arbitrary binary trees into hypercubes, Journa of Agorithms, vo. 43, pp , [19] A. W. Mai, A. Par, and. M. Fujimoto, Optimistic synchronization of parae simuations in coud computing environements, in Proc. IEEE Internationa onference on oud omputing, Bangaore, India, Sep , 2009, pp [20] M. S. Bazaraa, H. D. Sherai, and. M. Shetty, Noninear Programming: Theory and Agorithms, 3rd ed. New Yor, NY: John Wiey & Sons Inc., [21] L. Tassiuas and A. Ephremides, Stabiity properties of constrained queuing systems and scheduing poicies for maximum throughput in mutihop radio networs, IEEE Trans. Autom. ontro, vo. 37, no. 12, pp , Dec APPENDIX A. POOF OF POPOSITION 2 To derive the subprobem Φ Fμ, w, we combine a terms in 10 reated to λ, which eads to: [ S Uλ λ j Ψ i N Srci + ρ =1 E i=+1 ] Dsti. Note that the above expression is exacty the objective function of Φ F in Proposition 2. Next, we derive the objective function of Φ,whichis reativey more invoved. First, it is not difficut to verify that by switching the summation order from node-based to in-based, we can rewrite the fouth term in 10 as: E N i= 1 =1 = L E =1 i=+1 O Tx wi x I N E ye i. =1 i=+1 22 By the same toen, we can immediatey rewrite the partia second term excuding the summation invoving λ in 10 as: = L =1 N =1 ρ O I ρ Tx ρ x N N =1 i= S +1 =1 i= S +1 ρ y e i ρ y e i. 23 Now, for the more compex third term in 10, we have E N j Ψ i =1 a = = b = E =1 E =1 + N N E =1 L =1 =1 O O I + y e j j Ψ i O N E j Ψ i I y e j Tx x j Ψ i j Ψ i N E + =1 j Ψ i I j Ψ i j Ψ i x e j, 24 where a hods because the summations of the -variabes do not invove index j and can be taen outside of the summation with respect to index j; and b foows from the same toen as in 22 and 23 and the fact that switching the summation orders of the x e j -variabes does not change their sum vaue. Next, by adding 22, 23, 24 together and defining new w- and ŵ-variabes as in 13 and 14, we arrive at a summation of the foowing two terms: } L E Tx wi x, 25 N =1 [ E j Ψ i y e j ŵ i ye i ], 26 which is exacty the objective function of Φ and Φ. Note that the outer summation in 25 is with respect to in indices, each summand can be decomposed and computed at each in ocay. Liewise, the outer summation in 26 is with respect to node indices, each summand can be decomposed and computed at each node ocay. Finay, noting that the constraints E xe i,, can aso be separated in a in-wise fashion, we then have that the maximization of Lλ, x, μ, w can be equivaenty computed as the sum of a series of maximization subprobems at the source and each in, which are defined as in Proposition 2. This competes the proof. 72 Performance Evauation eview, Vo. 40, No. 4, March 2013

Secure Network Coding with a Cost Criterion

Secure Network Coding with a Cost Criterion Secure Network Coding with a Cost Criterion Jianong Tan, Murie Médard Laboratory for Information and Decision Systems Massachusetts Institute of Technoogy Cambridge, MA 0239, USA E-mai: {jianong, medard}@mit.edu

More information

Multi-Robot Task Scheduling

Multi-Robot Task Scheduling Proc of IEEE Internationa Conference on Robotics and Automation, Karsruhe, Germany, 013 Muti-Robot Tas Scheduing Yu Zhang and Lynne E Parer Abstract The scheduing probem has been studied extensivey in

More information

Simultaneous Routing and Power Allocation in CDMA Wireless Data Networks

Simultaneous Routing and Power Allocation in CDMA Wireless Data Networks Simutaneous Routing and Power Aocation in CDMA Wireess Data Networks Mikae Johansson *,LinXiao and Stephen Boyd * Department of Signas, Sensors and Systems Roya Institute of Technoogy, SE 00 Stockhom,

More information

Pricing and Revenue Sharing Strategies for Internet Service Providers

Pricing and Revenue Sharing Strategies for Internet Service Providers Pricing and Revenue Sharing Strategies for Internet Service Providers Linhai He and Jean Warand Department of Eectrica Engineering and Computer Sciences University of Caifornia at Berkeey {inhai,wr}@eecs.berkeey.edu

More information

Fast Robust Hashing. ) [7] will be re-mapped (and therefore discarded), due to the load-balancing property of hashing.

Fast Robust Hashing. ) [7] will be re-mapped (and therefore discarded), due to the load-balancing property of hashing. Fast Robust Hashing Manue Urueña, David Larrabeiti and Pabo Serrano Universidad Caros III de Madrid E-89 Leganés (Madrid), Spain Emai: {muruenya,darra,pabo}@it.uc3m.es Abstract As statefu fow-aware services

More information

Load Balance vs Energy Efficiency in Traffic Engineering: A Game Theoretical Perspective

Load Balance vs Energy Efficiency in Traffic Engineering: A Game Theoretical Perspective Load Baance vs Energy Efficiency in Traffic Engineering: A Game Theoretica Perspective Yangming Zhao, Sheng Wang, Shizhong Xu and Xiong Wang Schoo of Communication and Information Engineering University

More information

Chapter 3: e-business Integration Patterns

Chapter 3: e-business Integration Patterns Chapter 3: e-business Integration Patterns Page 1 of 9 Chapter 3: e-business Integration Patterns "Consistency is the ast refuge of the unimaginative." Oscar Wide In This Chapter What Are Integration Patterns?

More information

500 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 31, NO. 3, MARCH 2013

500 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 31, NO. 3, MARCH 2013 500 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 3, NO. 3, MARCH 203 Cognitive Radio Network Duaity Agorithms for Utiity Maximization Liang Zheng Chee Wei Tan, Senior Member, IEEE Abstract We

More information

Pricing Internet Services With Multiple Providers

Pricing Internet Services With Multiple Providers Pricing Internet Services With Mutipe Providers Linhai He and Jean Warand Dept. of Eectrica Engineering and Computer Science University of Caifornia at Berkeey Berkeey, CA 94709 inhai, wr@eecs.berkeey.edu

More information

IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 31, NO. 12, DECEMBER 2013 1

IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 31, NO. 12, DECEMBER 2013 1 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 31, NO. 12, DECEMBER 2013 1 Scaabe Muti-Cass Traffic Management in Data Center Backbone Networks Amitabha Ghosh, Sangtae Ha, Edward Crabbe, and Jennifer

More information

Lecture 7 Datalink Ethernet, Home. Datalink Layer Architectures

Lecture 7 Datalink Ethernet, Home. Datalink Layer Architectures Lecture 7 Dataink Ethernet, Home Peter Steenkiste Schoo of Computer Science Department of Eectrica and Computer Engineering Carnegie Meon University 15-441 Networking, Spring 2004 http://www.cs.cmu.edu/~prs/15-441

More information

Betting Strategies, Market Selection, and the Wisdom of Crowds

Betting Strategies, Market Selection, and the Wisdom of Crowds Betting Strategies, Market Seection, and the Wisdom of Crowds Wiemien Kets Northwestern University w-kets@keogg.northwestern.edu David M. Pennock Microsoft Research New York City dpennock@microsoft.com

More information

Teamwork. Abstract. 2.1 Overview

Teamwork. Abstract. 2.1 Overview 2 Teamwork Abstract This chapter presents one of the basic eements of software projects teamwork. It addresses how to buid teams in a way that promotes team members accountabiity and responsibiity, and

More information

Comparison of Traditional and Open-Access Appointment Scheduling for Exponentially Distributed Service Time

Comparison of Traditional and Open-Access Appointment Scheduling for Exponentially Distributed Service Time Journa of Heathcare Engineering Vo. 6 No. 3 Page 34 376 34 Comparison of Traditiona and Open-Access Appointment Scheduing for Exponentiay Distributed Service Chongjun Yan, PhD; Jiafu Tang *, PhD; Bowen

More information

Finance 360 Problem Set #6 Solutions

Finance 360 Problem Set #6 Solutions Finance 360 Probem Set #6 Soutions 1) Suppose that you are the manager of an opera house. You have a constant margina cost of production equa to $50 (i.e. each additiona person in the theatre raises your

More information

CERTIFICATE COURSE ON CLIMATE CHANGE AND SUSTAINABILITY. Course Offered By: Indian Environmental Society

CERTIFICATE COURSE ON CLIMATE CHANGE AND SUSTAINABILITY. Course Offered By: Indian Environmental Society CERTIFICATE COURSE ON CLIMATE CHANGE AND SUSTAINABILITY Course Offered By: Indian Environmenta Society INTRODUCTION The Indian Environmenta Society (IES) a dynamic and fexibe organization with a goba vision

More information

With the arrival of Java 2 Micro Edition (J2ME) and its industry

With the arrival of Java 2 Micro Edition (J2ME) and its industry Knowedge-based Autonomous Agents for Pervasive Computing Using AgentLight Fernando L. Koch and John-Jues C. Meyer Utrecht University Project AgentLight is a mutiagent system-buiding framework targeting

More information

Scheduling in Multi-Channel Wireless Networks

Scheduling in Multi-Channel Wireless Networks Scheduing in Muti-Channe Wireess Networks Vartika Bhandari and Nitin H. Vaidya University of Iinois at Urbana-Champaign, USA vartikab@acm.org, nhv@iinois.edu Abstract. The avaiabiity of mutipe orthogona

More information

GREEN: An Active Queue Management Algorithm for a Self Managed Internet

GREEN: An Active Queue Management Algorithm for a Self Managed Internet : An Active Queue Management Agorithm for a Sef Managed Internet Bartek Wydrowski and Moshe Zukerman ARC Specia Research Centre for Utra-Broadband Information Networks, EEE Department, The University of

More information

Vendor Performance Measurement Using Fuzzy Logic Controller

Vendor Performance Measurement Using Fuzzy Logic Controller The Journa of Mathematics and Computer Science Avaiabe onine at http://www.tjmcs.com The Journa of Mathematics and Computer Science Vo.2 No.2 (2011) 311-318 Performance Measurement Using Fuzzy Logic Controer

More information

A Similarity Search Scheme over Encrypted Cloud Images based on Secure Transformation

A Similarity Search Scheme over Encrypted Cloud Images based on Secure Transformation A Simiarity Search Scheme over Encrypted Coud Images based on Secure Transormation Zhihua Xia, Yi Zhu, Xingming Sun, and Jin Wang Jiangsu Engineering Center o Network Monitoring, Nanjing University o Inormation

More information

Normalization of Database Tables. Functional Dependency. Examples of Functional Dependencies: So Now what is Normalization? Transitive Dependencies

Normalization of Database Tables. Functional Dependency. Examples of Functional Dependencies: So Now what is Normalization? Transitive Dependencies ISM 602 Dr. Hamid Nemati Objectives The idea Dependencies Attributes and Design Understand concepts normaization (Higher-Leve Norma Forms) Learn how to normaize tabes Understand normaization and database

More information

Fixed income managers: evolution or revolution

Fixed income managers: evolution or revolution Fixed income managers: evoution or revoution Traditiona approaches to managing fixed interest funds rey on benchmarks that may not represent optima risk and return outcomes. New techniques based on separate

More information

Art of Java Web Development By Neal Ford 624 pages US$44.95 Manning Publications, 2004 ISBN: 1-932394-06-0

Art of Java Web Development By Neal Ford 624 pages US$44.95 Manning Publications, 2004 ISBN: 1-932394-06-0 IEEE DISTRIBUTED SYSTEMS ONLINE 1541-4922 2005 Pubished by the IEEE Computer Society Vo. 6, No. 5; May 2005 Editor: Marcin Paprzycki, http://www.cs.okstate.edu/%7emarcin/ Book Reviews: Java Toos and Frameworks

More information

Load Balancing in Distributed Web Server Systems with Partial Document Replication *

Load Balancing in Distributed Web Server Systems with Partial Document Replication * Load Baancing in Distributed Web Server Systems with Partia Document Repication * Ling Zhuo Cho-Li Wang Francis C. M. Lau Department of Computer Science and Information Systems The University of Hong Kong

More information

Dynamic Pricing Trade Market for Shared Resources in IIU Federated Cloud

Dynamic Pricing Trade Market for Shared Resources in IIU Federated Cloud Dynamic Pricing Trade Market or Shared Resources in IIU Federated Coud Tongrang Fan 1, Jian Liu 1, Feng Gao 1 1Schoo o Inormation Science and Technoogy, Shiiazhuang Tiedao University, Shiiazhuang, 543,

More information

Maintenance activities planning and grouping for complex structure systems

Maintenance activities planning and grouping for complex structure systems Maintenance activities panning and grouping for compex structure systems Hai Canh u, Phuc Do an, Anne Barros, Christophe Berenguer To cite this version: Hai Canh u, Phuc Do an, Anne Barros, Christophe

More information

Assessing Network Vulnerability Under Probabilistic Region Failure Model

Assessing Network Vulnerability Under Probabilistic Region Failure Model 2011 IEEE 12th Internationa Conference on High Performance Switching and Routing Assessing Networ Vunerabiity Under Probabiistic Region Faiure Mode Xiaoiang Wang, Xiaohong Jiang and Achie Pattavina State

More information

Wide-Area Traffic Management for. Cloud Services

Wide-Area Traffic Management for. Cloud Services Wide-Area Traffic Management for Coud Services Joe Wenjie Jiang A Dissertation Presented to the Facuty of Princeton University in Candidacy for the Degree of Doctor of Phiosophy Recommended for Acceptance

More information

Face Hallucination and Recognition

Face Hallucination and Recognition Face Haucination and Recognition Xiaogang Wang and Xiaoou Tang Department of Information Engineering, The Chinese University of Hong Kong {xgwang1, xtang}@ie.cuhk.edu.hk http://mmab.ie.cuhk.edu.hk Abstract.

More information

Application-Aware Data Collection in Wireless Sensor Networks

Application-Aware Data Collection in Wireless Sensor Networks Appication-Aware Data Coection in Wireess Sensor Networks Xiaoin Fang *, Hong Gao *, Jianzhong Li *, and Yingshu Li +* * Schoo of Computer Science and Technoogy, Harbin Institute of Technoogy, Harbin,

More information

We focus on systems composed of entities operating with autonomous control, such

We focus on systems composed of entities operating with autonomous control, such Middeware Architecture for Federated Contro Systems Girish Baiga and P.R. Kumar University of Iinois at Urbana-Champaign A federated contro system (FCS) is composed of autonomous entities, such as cars,

More information

Cooperative Content Distribution and Traffic Engineering in an ISP Network

Cooperative Content Distribution and Traffic Engineering in an ISP Network Cooperative Content Distribution and Traffic Engineering in an ISP Network Wenjie Jiang, Rui Zhang-Shen, Jennifer Rexford, Mung Chiang Department of Computer Science, and Department of Eectrica Engineering

More information

Overview of Health and Safety in China

Overview of Health and Safety in China Overview of Heath and Safety in China Hongyuan Wei 1, Leping Dang 1, and Mark Hoye 2 1 Schoo of Chemica Engineering, Tianjin University, Tianjin 300072, P R China, E-mai: david.wei@tju.edu.cn 2 AstraZeneca

More information

A Supplier Evaluation System for Automotive Industry According To Iso/Ts 16949 Requirements

A Supplier Evaluation System for Automotive Industry According To Iso/Ts 16949 Requirements A Suppier Evauation System for Automotive Industry According To Iso/Ts 16949 Requirements DILEK PINAR ÖZTOP 1, ASLI AKSOY 2,*, NURSEL ÖZTÜRK 2 1 HONDA TR Purchasing Department, 41480, Çayırova - Gebze,

More information

Take me to your leader! Online Optimization of Distributed Storage Configurations

Take me to your leader! Online Optimization of Distributed Storage Configurations Take me to your eader! Onine Optimization of Distributed Storage Configurations Artyom Sharov Aexander Shraer Arif Merchant Murray Stokey sharov@cs.technion.ac.i, {shraex, aamerchant, mstokey}@googe.com

More information

Order-to-Cash Processes

Order-to-Cash Processes TMI170 ING info pat 2:Info pat.qxt 01/12/2008 09:25 Page 1 Section Two: Order-to-Cash Processes Gregory Cronie, Head Saes, Payments and Cash Management, ING O rder-to-cash and purchase-topay processes

More information

Application and Desktop Virtualization

Application and Desktop Virtualization Appication and Desktop Virtuaization Content 1) Why Appication and Desktop Virtuaization 2) Some terms reated to vapp and vdesktop 3) Appication and Desktop Deivery 4) Appication Virtuaization 5)- Type

More information

Early access to FAS payments for members in poor health

Early access to FAS payments for members in poor health Financia Assistance Scheme Eary access to FAS payments for members in poor heath Pension Protection Fund Protecting Peope s Futures The Financia Assistance Scheme is administered by the Pension Protection

More information

A New Statistical Approach to Network Anomaly Detection

A New Statistical Approach to Network Anomaly Detection A New Statistica Approach to Network Anomay Detection Christian Caegari, Sandrine Vaton 2, and Michee Pagano Dept of Information Engineering, University of Pisa, ITALY E-mai: {christiancaegari,mpagano}@ietunipiit

More information

Network/Communicational Vulnerability

Network/Communicational Vulnerability Automated teer machines (ATMs) are a part of most of our ives. The major appea of these machines is convenience The ATM environment is changing and that change has serious ramifications for the security

More information

A Latent Variable Pairwise Classification Model of a Clustering Ensemble

A Latent Variable Pairwise Classification Model of a Clustering Ensemble A atent Variabe Pairwise Cassification Mode of a Custering Ensembe Vadimir Berikov Soboev Institute of mathematics, Novosibirsk State University, Russia berikov@math.nsc.ru http://www.math.nsc.ru Abstract.

More information

Leakage detection in water pipe networks using a Bayesian probabilistic framework

Leakage detection in water pipe networks using a Bayesian probabilistic framework Probabiistic Engineering Mechanics 18 (2003) 315 327 www.esevier.com/ocate/probengmech Leakage detection in water pipe networks using a Bayesian probabiistic framework Z. Pouakis, D. Vaougeorgis, C. Papadimitriou*

More information

Minimizing the Total Weighted Completion Time of Coflows in Datacenter Networks

Minimizing the Total Weighted Completion Time of Coflows in Datacenter Networks Minimizing the Tota Weighted Competion Time of Cofows in Datacenter Networks Zhen Qiu Ciff Stein and Yuan Zhong ABSTRACT Communications in datacenter jobs (such as the shuffe operations in MapReduce appications

More information

Oligopoly in Insurance Markets

Oligopoly in Insurance Markets Oigopoy in Insurance Markets June 3, 2008 Abstract We consider an oigopoistic insurance market with individuas who differ in their degrees of accident probabiities. Insurers compete in coverage and premium.

More information

Migrating and Managing Dynamic, Non-Textua Content

Migrating and Managing Dynamic, Non-Textua Content Considering Dynamic, Non-Textua Content when Migrating Digita Asset Management Systems Aya Stein; University of Iinois at Urbana-Champaign; Urbana, Iinois USA Santi Thompson; University of Houston; Houston,

More information

A train dispatching model based on fuzzy passenger demand forecasting during holidays

A train dispatching model based on fuzzy passenger demand forecasting during holidays Journa of Industria Engineering and Management JIEM, 2013 6(1):320-335 Onine ISSN: 2013-0953 Print ISSN: 2013-8423 http://dx.doi.org/10.3926/jiem.699 A train dispatching mode based on fuzzy passenger demand

More information

Human Capital & Human Resources Certificate Programs

Human Capital & Human Resources Certificate Programs MANAGEMENT CONCEPTS Human Capita & Human Resources Certificate Programs Programs to deveop functiona and strategic skis in: Human Capita // Human Resources ENROLL TODAY! Contract Hoder Contract GS-02F-0010J

More information

On Capacity Scaling in Arbitrary Wireless Networks

On Capacity Scaling in Arbitrary Wireless Networks On Capacity Scaing in Arbitrary Wireess Networks Urs Niesen, Piyush Gupta, and Devavrat Shah 1 Abstract arxiv:07112745v3 [csit] 3 Aug 2009 In recent work, Özgür, Lévêque, and Tse 2007) obtained a compete

More information

A quantum model for the stock market

A quantum model for the stock market A quantum mode for the stock market Authors: Chao Zhang a,, Lu Huang b Affiiations: a Schoo of Physics and Engineering, Sun Yat-sen University, Guangzhou 5175, China b Schoo of Economics and Business Administration,

More information

Risk Margin for a Non-Life Insurance Run-Off

Risk Margin for a Non-Life Insurance Run-Off Risk Margin for a Non-Life Insurance Run-Off Mario V. Wüthrich, Pau Embrechts, Andreas Tsanakas August 15, 2011 Abstract For sovency purposes insurance companies need to cacuate so-caed best-estimate reserves

More information

Risk Margin for a Non-Life Insurance Run-Off

Risk Margin for a Non-Life Insurance Run-Off Risk Margin for a Non-Life Insurance Run-Off Mario V. Wüthrich, Pau Embrechts, Andreas Tsanakas February 2, 2011 Abstract For sovency purposes insurance companies need to cacuate so-caed best-estimate

More information

Iterative Water-filling for Load-balancing in Wireless LAN or Microcellular Networks

Iterative Water-filling for Load-balancing in Wireless LAN or Microcellular Networks terative Water-fiing for Load-baancing in Wireess LAN or Microceuar Networks Jeremy K. Chen Theodore S. Rappaport Gustavo de Veciana Wireess Networking and Communications Group (WNCG), be University of

More information

Randomized Algorithms for Scheduling VMs in the Cloud

Randomized Algorithms for Scheduling VMs in the Cloud Randomized Agorithms for Scheduing VMs in the Coud Javad Ghaderi Coumbia University Abstract We consider the probem of scheduing VMs (Virtua Machines) in a muti-server system motivated by coud computing

More information

Advanced ColdFusion 4.0 Application Development - 3 - Server Clustering Using Bright Tiger

Advanced ColdFusion 4.0 Application Development - 3 - Server Clustering Using Bright Tiger Advanced CodFusion 4.0 Appication Deveopment - CH 3 - Server Custering Using Bri.. Page 1 of 7 [Figures are not incuded in this sampe chapter] Advanced CodFusion 4.0 Appication Deveopment - 3 - Server

More information

Ricoh Healthcare. Process Optimized. Healthcare Simplified.

Ricoh Healthcare. Process Optimized. Healthcare Simplified. Ricoh Heathcare Process Optimized. Heathcare Simpified. Rather than a destination that concudes with the eimination of a paper, the Paperess Maturity Roadmap is a continuous journey to strategicay remove

More information

Big Data projects and use cases. Claus Samuelsen IBM Analytics, Europe csa@dk.ibm.com

Big Data projects and use cases. Claus Samuelsen IBM Analytics, Europe csa@dk.ibm.com Big projects and use cases Caus Samuesen IBM Anaytics, Europe csa@dk.ibm.com IBM Sofware Overview of BigInsights IBM BigInsights Scientist Free Quick Start (non production): IBM Open Patform BigInsights

More information

CONTRIBUTION OF INTERNAL AUDITING IN THE VALUE OF A NURSING UNIT WITHIN THREE YEARS

CONTRIBUTION OF INTERNAL AUDITING IN THE VALUE OF A NURSING UNIT WITHIN THREE YEARS Dehi Business Review X Vo. 4, No. 2, Juy - December 2003 CONTRIBUTION OF INTERNAL AUDITING IN THE VALUE OF A NURSING UNIT WITHIN THREE YEARS John N.. Var arvatsouakis atsouakis DURING the present time,

More information

WHITE PAPER BEsT PRAcTIcEs: PusHIng ExcEl BEyond ITs limits WITH InfoRmATIon optimization

WHITE PAPER BEsT PRAcTIcEs: PusHIng ExcEl BEyond ITs limits WITH InfoRmATIon optimization Best Practices: Pushing Exce Beyond Its Limits with Information Optimization WHITE Best Practices: Pushing Exce Beyond Its Limits with Information Optimization Executive Overview Microsoft Exce is the

More information

Australian Bureau of Statistics Management of Business Providers

Australian Bureau of Statistics Management of Business Providers Purpose Austraian Bureau of Statistics Management of Business Providers 1 The principa objective of the Austraian Bureau of Statistics (ABS) in respect of business providers is to impose the owest oad

More information

Bite-Size Steps to ITIL Success

Bite-Size Steps to ITIL Success 7 Bite-Size Steps to ITIL Success Pus making a Business Case for ITIL! Do you want to impement ITIL but don t know where to start? 7 Bite-Size Steps to ITIL Success can hep you to decide whether ITIL can

More information

Assigning Tasks in a 24-Hour Software Development Model

Assigning Tasks in a 24-Hour Software Development Model Assigning Tasks in a -Hour Software Deveopent Mode Pankaj Jaote, Gourav Jain Departent of Coputer Science & Engineering Indian Institute of Technoogy, Kanpur, INDIA 006 Eai:{jaote, gauravj}@iitk.ac.in

More information

Virtual trunk simulation

Virtual trunk simulation Virtua trunk simuation Samui Aato * Laboratory of Teecommunications Technoogy Hesinki University of Technoogy Sivia Giordano Laboratoire de Reseaux de Communication Ecoe Poytechnique Federae de Lausanne

More information

An Idiot s guide to Support vector machines (SVMs)

An Idiot s guide to Support vector machines (SVMs) An Idiot s guide to Support vector machines (SVMs) R. Berwick, Viage Idiot SVMs: A New Generation of Learning Agorithms Pre 1980: Amost a earning methods earned inear decision surfaces. Linear earning

More information

Life Contingencies Study Note for CAS Exam S. Tom Struppeck

Life Contingencies Study Note for CAS Exam S. Tom Struppeck Life Contingencies Study Note for CAS Eam S Tom Struppeck (Revised 9/19/2015) Introduction Life contingencies is a term used to describe surviva modes for human ives and resuting cash fows that start or

More information

The growth of online Internet services during the past decade has

The growth of online Internet services during the past decade has IEEE DS Onine, Voume 2, Number 4 Designing an Adaptive CORBA Load Baancing Service Using TAO Ossama Othman, Caros O'Ryan, and Dougas C. Schmidt University of Caifornia, Irvine The growth of onine Internet

More information

The definition of insanity is doing the same thing over and over again and expecting different results

The definition of insanity is doing the same thing over and over again and expecting different results insurance services Sma Business Insurance a market opportunity being missed Einstein may not have known much about insurance, but if you appy his definition to the way existing brands are deveoping their

More information

We are XMA and Viglen.

We are XMA and Viglen. alearn with Microsoft 16pp 21.07_Layout 1 22/12/2014 10:49 Page 1 FRONT COVER alearn with Microsoft We are XMA and Vigen. Ca us now on 0115 846 4900 Visit www.xma.co.uk/aearn Emai alearn@xma.co.uk Foow

More information

Chapter 2 Traditional Software Development

Chapter 2 Traditional Software Development Chapter 2 Traditiona Software Deveopment 2.1 History of Project Management Large projects from the past must aready have had some sort of project management, such the Pyramid of Giza or Pyramid of Cheops,

More information

Market Design & Analysis for a P2P Backup System

Market Design & Analysis for a P2P Backup System Market Design & Anaysis for a P2P Backup System Sven Seuken Schoo of Engineering & Appied Sciences Harvard University, Cambridge, MA seuken@eecs.harvard.edu Denis Chares, Max Chickering, Sidd Puri Microsoft

More information

Design Considerations

Design Considerations Chapter 2: Basic Virtua Private Network Depoyment Page 1 of 12 Chapter 2: Basic Virtua Private Network Depoyment Before discussing the features of Windows 2000 tunneing technoogy, it is important to estabish

More information

SELECTING THE SUITABLE ERP SYSTEM: A FUZZY AHP APPROACH. Ufuk Cebeci

SELECTING THE SUITABLE ERP SYSTEM: A FUZZY AHP APPROACH. Ufuk Cebeci SELECTING THE SUITABLE ERP SYSTEM: A FUZZY AHP APPROACH Ufuk Cebeci Department of Industria Engineering, Istanbu Technica University, Macka, Istanbu, Turkey - ufuk_cebeci@yahoo.com Abstract An Enterprise

More information

This paper considers an inventory system with an assembly structure. In addition to uncertain customer

This paper considers an inventory system with an assembly structure. In addition to uncertain customer MANAGEMENT SCIENCE Vo. 51, No. 8, August 2005, pp. 1250 1265 issn 0025-1909 eissn 1526-5501 05 5108 1250 informs doi 10.1287/mnsc.1050.0394 2005 INFORMS Inventory Management for an Assemby System wh Product

More information

COMPARISON OF DIFFUSION MODELS IN ASTRONOMICAL OBJECT LOCALIZATION

COMPARISON OF DIFFUSION MODELS IN ASTRONOMICAL OBJECT LOCALIZATION COMPARISON OF DIFFUSION MODELS IN ASTRONOMICAL OBJECT LOCALIZATION Františe Mojžíš Department of Computing and Contro Engineering, ICT Prague, Technicá, 8 Prague frantise.mojzis@vscht.cz Abstract This

More information

Data Centers as Demand Response Resources in the Electricity Market: Some Preliminary Results

Data Centers as Demand Response Resources in the Electricity Market: Some Preliminary Results Data Centers as Demand Response Resources in the Eectricity Maret: Some Preiminary Resuts Rui Wang ECE Department Drexe University Phiadephia, PA 19104, USA rui.wang@drexe.edu Nagarajan Kandasamy ECE Department

More information

Federal Financial Management Certificate Program

Federal Financial Management Certificate Program MANAGEMENT CONCEPTS Federa Financia Management Certificate Program Training to hep you achieve the highest eve performance in: Accounting // Auditing // Budgeting // Financia Management ENROLL TODAY! Contract

More information

A Branch-and-Price Algorithm for Parallel Machine Scheduling with Time Windows and Job Priorities

A Branch-and-Price Algorithm for Parallel Machine Scheduling with Time Windows and Job Priorities A Branch-and-Price Agorithm for Parae Machine Scheduing with Time Windows and Job Priorities Jonathan F. Bard, 1 Siwate Rojanasoonthon 2 1 Graduate Program in Operations Research and Industria Engineering,

More information

An Integrated Data Management Framework of Wireless Sensor Network

An Integrated Data Management Framework of Wireless Sensor Network An Integrated Data Management Framework of Wireess Sensor Network for Agricutura Appications 1,2 Zhao Liang, 2 He Liyuan, 1 Zheng Fang, 1 Jin Xing 1 Coege of Science, Huazhong Agricutura University, Wuhan

More information

Integrating Risk into your Plant Lifecycle A next generation software architecture for risk based

Integrating Risk into your Plant Lifecycle A next generation software architecture for risk based Integrating Risk into your Pant Lifecyce A next generation software architecture for risk based operations Dr Nic Cavanagh 1, Dr Jeremy Linn 2 and Coin Hickey 3 1 Head of Safeti Product Management, DNV

More information

Best Practices for Push & Pull Using Oracle Inventory Stock Locators. Introduction to Master Data and Master Data Management (MDM): Part 1

Best Practices for Push & Pull Using Oracle Inventory Stock Locators. Introduction to Master Data and Master Data Management (MDM): Part 1 SPECIAL CONFERENCE ISSUE THE OFFICIAL PUBLICATION OF THE Orace Appications USERS GROUP spring 2012 Introduction to Master Data and Master Data Management (MDM): Part 1 Utiizing Orace Upgrade Advisor for

More information

Message. The Trade and Industry Bureau is committed to providing maximum support for Hong Kong s manufacturing and services industries.

Message. The Trade and Industry Bureau is committed to providing maximum support for Hong Kong s manufacturing and services industries. Message The Trade and Industry Bureau is committed to providing maximum support for Hong Kong s manufacturing and services industries. With the weight of our economy shifting towards knowedge-based and

More information

Design and Analysis of a Hidden Peer-to-peer Backup Market

Design and Analysis of a Hidden Peer-to-peer Backup Market Design and Anaysis of a Hidden Peer-to-peer Backup Market Sven Seuken, Denis Chares, Max Chickering, Mary Czerwinski Kama Jain, David C. Parkes, Sidd Puri, and Desney Tan December, 2015 Abstract We present

More information

INDUSTRIAL AND COMMERCIAL

INDUSTRIAL AND COMMERCIAL Finance TM NEW YORK CITY DEPARTMENT OF FINANCE TAX & PARKING PROGRAM OPERATIONS DIVISION INDUSTRIAL AND COMMERCIAL ABATEMENT PROGRAM PRELIMINARY APPLICATION AND INSTRUCTIONS Mai to: NYC Department of Finance,

More information

Business Banking. A guide for franchises

Business Banking. A guide for franchises Business Banking A guide for franchises Hep with your franchise business, right on your doorstep A true understanding of the needs of your business: that s what makes RBS the right choice for financia

More information

Hedge Fund Capital Accounts and Revaluations: Are They Section 704(b) Compliant?

Hedge Fund Capital Accounts and Revaluations: Are They Section 704(b) Compliant? o EDITED BY ROGER F. PILLOW, LL.M. PARTNERSHIPS, S CORPORATIONS & LLCs Hedge Fund Capita Accounts and Revauations: Are They Section 704(b) Compiant? THOMAS GRAY Hedge funds treated as partnerships for

More information

PROPAGATION OF SURGE WAVES ON NON-HOMOGENEOUS TRANSMISSION LINES INDUCED BY LIGHTNING STROKE

PROPAGATION OF SURGE WAVES ON NON-HOMOGENEOUS TRANSMISSION LINES INDUCED BY LIGHTNING STROKE Advances in Eectrica and Eectronic Engineering 98 PROPAGATION OF SRGE WAVES ON NON-HOMOGENEOS TRANSMISSION LINES INDCED BY LIGHTNING STROKE Z. Benešová, V. Kotan WB Pisen, Facuty of Eectrica Engineering,

More information

Chapter 3: JavaScript in Action Page 1 of 10. How to practice reading and writing JavaScript on a Web page

Chapter 3: JavaScript in Action Page 1 of 10. How to practice reading and writing JavaScript on a Web page Chapter 3: JavaScript in Action Page 1 of 10 Chapter 3: JavaScript in Action In this chapter, you get your first opportunity to write JavaScript! This chapter introduces you to JavaScript propery. In addition,

More information

Hybrid Interface Solutions for next Generation Wireless Access Infrastructure

Hybrid Interface Solutions for next Generation Wireless Access Infrastructure tec. Connectivity & Networks Voker Sorhage Hybrid Interface Soutions for next Generation Wireess Access Infrastructure Broadband wireess communication wi revoutionize every aspect of peope s ives by enabing

More information

A Description of the California Partnership for Long-Term Care Prepared by the California Department of Health Care Services

A Description of the California Partnership for Long-Term Care Prepared by the California Department of Health Care Services 2012 Before You Buy A Description of the Caifornia Partnership for Long-Term Care Prepared by the Caifornia Department of Heath Care Services Page 1 of 13 Ony ong-term care insurance poicies bearing any

More information

Leadership & Management Certificate Programs

Leadership & Management Certificate Programs MANAGEMENT CONCEPTS Leadership & Management Certificate Programs Programs to deveop expertise in: Anaytics // Leadership // Professiona Skis // Supervision ENROLL TODAY! Contract oder Contract GS-02F-0010J

More information

LADDER SAFETY Table of Contents

LADDER SAFETY Table of Contents Tabe of Contents SECTION 1. TRAINING PROGRAM INTRODUCTION..................3 Training Objectives...........................................3 Rationae for Training.........................................3

More information

SPOTLIGHT. A year of transformation

SPOTLIGHT. A year of transformation WINTER ISSUE 2014 2015 SPOTLIGHT Wecome to the winter issue of Oasis Spotight. These newsetters are designed to keep you upto-date with news about the Oasis community. This quartery issue features an artice

More information

ONE of the most challenging problems addressed by the

ONE of the most challenging problems addressed by the IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 44, NO. 9, SEPTEMBER 2006 2587 A Mutieve Context-Based System for Cassification of Very High Spatia Resoution Images Lorenzo Bruzzone, Senior Member,

More information

Modeling a Scenario-based Distribution Network Design Problem in a Physical Internet-enabled open Logistics Web

Modeling a Scenario-based Distribution Network Design Problem in a Physical Internet-enabled open Logistics Web 4 th Internationa conference on Information Systems, Logistics and Suppy Chain Quebec City August 26-29, 2012 Modeing a Scenario-based Distribution Network Design Probem in a Physica Internet-enabed open

More information

Business schools are the academic setting where. The current crisis has highlighted the need to redefine the role of senior managers in organizations.

Business schools are the academic setting where. The current crisis has highlighted the need to redefine the role of senior managers in organizations. c r o s os r oi a d s REDISCOVERING THE ROLE OF BUSINESS SCHOOLS The current crisis has highighted the need to redefine the roe of senior managers in organizations. JORDI CANALS Professor and Dean, IESE

More information

How To Deiver Resuts

How To Deiver Resuts Message We sha make every effort to strengthen the community buiding programme which serves to foster among the peope of Hong Kong a sense of beonging and mutua care. We wi continue to impement the District

More information

Efficient Data Partitioning Model for Heterogeneous Graphs in the Cloud

Efficient Data Partitioning Model for Heterogeneous Graphs in the Cloud Efficient Data Partitioning Mode for Heterogeneous Graphs in the Coud Kisung Lee Georgia Institute of Technoogy ksee@gatech.edu Ling Liu Georgia Institute of Technoogy ingiu@cc.gatech.edu ABSTRACT As the

More information

DEGREES OF ORDERS ON TORSION-FREE ABELIAN GROUPS

DEGREES OF ORDERS ON TORSION-FREE ABELIAN GROUPS DEGREES OF ORDERS ON TORSION-FREE ABELIAN GROUPS ASHER M. KACH, KAREN LANGE, AND REED SOLOMON Abstract. We show that if H is an effectivey competey decomposabe computabe torsion-free abeian group, then

More information

How To Get Acedo With Microsoft.Com

How To Get Acedo With Microsoft.Com alearn with Microsoft We are XMA. Ca us now on 0115 846 4900 Visit www.xma.co.uk/aearn Emai alearn@xma.co.uk Foow us @WeareXMA Introduction Use our 'steps to alearn' framework to ensure you cover a bases...

More information

Insertion and deletion correcting DNA barcodes based on watermarks

Insertion and deletion correcting DNA barcodes based on watermarks Kracht and Schober BMC Bioinformatics (2015) 16:50 DOI 10.1186/s12859-015-0482-7 METHODOLOGY ARTICLE Open Access Insertion and deetion correcting DNA barcodes based on watermarks David Kracht * and Steffen

More information