C e n t r u m v o o r W i s k u n d e e n I n f o r m a t i c a

Size: px
Start display at page:

Download "C e n t r u m v o o r W i s k u n d e e n I n f o r m a t i c a"

Transcription

1 C e n t r u m v o o r W s k u n d e e n I n f o r m a t c a Probablty, Networks and Algorthms Probablty, Networks and Algorthms Traffc-splttng networks operatng under alpha-far sharng polces and balanced farness P.M.D. Leshout REPORT PNA-R0707 JULY 2007

2 Centrum voor Wskunde en Informatca (CWI) s the natonal research nsttute for Mathematcs and Computer Scence. It s sponsored by the Netherlands Organsaton for Scentfc Research (NWO). CWI s a foundng member of ERCIM, the European Research Consortum for Informatcs and Mathematcs. CWI's research has a theme-orented structure and s grouped nto four clusters. Lsted below are the names of the clusters and n parentheses ther acronyms. Probablty, Networks and Algorthms (PNA) Software Engneerng (SEN) Modellng, Analyss and Smulaton (MAS) Informaton Systems (INS) Copyrght 2007, Stchtng Centrum voor Wskunde en Informatca P.O. Box 94079, 1090 GB Amsterdam (NL) Kruslaan 413, 1098 SJ Amsterdam (NL) Telephone Telefax ISSN

3 Traffc-splttng networks operatng under alpha-far sharng polces and balanced farness ABSTRACT We consder a data network n whch, besdes classes of users that use specfc routes, one class of users can splt ts traffc over several routes. We consder load balancng at the packetlevel, mplyng that traffc of ths class of users can be dvded among several routes at the same tme. Assumng that load balancng s based on an alpha-far sharng polcy, we show that the network has multple possble behavors. In partcular, we show that some classes of users, dependng on the state of the network, share capacty accordng to some Dscrmnatory Processor Sharng (DPS) model, whereas each of the remanng classes of users behaves as n a sngle-class sngle-node model. We compare the performance of ths network wth that of a smlar network, where packet-level load balancng s based on balanced farness. We derve explct expressons for the mean number of users under balanced farness, and show by conductng extensve smulaton experments that these provde accurate approxmatons for the ones under alpha-far sharng Mathematcs Subject Classfcaton: 60K25; 68M20 Keywords and Phrases: Traffc splttng; Alpha-far sharng; Balanced farness Note: Ths research has been funded by the Dutch Bsk/BRICKS (Basc Research n Informatcs for Creatng the Knowledge Socety) project.

4

5 Traffc-Splttng Networks Operatng under Alpha-far Sharng Polces and Balanced Farness P. Leshout CWI P.O. Box 94079, 1090 GB Amsterdam, The Netherlands Emal: July 7, 2007 Abstract We consder a data network n whch, besdes classes of users that use specfc routes, one class of users can splt ts traffc over several routes. We consder load balancng at the packet-level, mplyng that traffc of ths class of users can be dvded among several routes at the same tme. Assumng that load balancng s based on an alpha-far sharng polcy, we show that the network has multple possble behavors. In partcular, we show that some classes of users, dependng on the state of the network, share capacty accordng to some Dscrmnatory Processor Sharng (DPS) model, whereas each of the remanng classes of users behaves as n a sngle-class sngle-node model. We compare the performance of ths network wth that of a smlar network, where packet-level load balancng s based on balanced farness. We derve explct expressons for the mean number of users under balanced farness, and show by conductng extensve smulaton experments that these provde accurate approxmatons for the ones under alphafar sharng. 1 1 Ths research has been funded by the Dutch BSIK/BRICKS (Basc Research n Informatcs for Creatng the Knowledge Socety) project. 1

6 1 Introducton The performance of communcaton networks can be mproved when the servce demands are effcently dvded among the avalable resources, so-called load balancng. One can apply ether statc or dynamc load balancng. In the former case the balancng s not affected by the state of the network, whereas n the latter case t does depend on the system state. It s clear that better performance can be acheved when usng dynamc load balancng, but t s often hard to fnd the optmal load balancng polcy. Even for smple systems such a dynamc load balancng problem has non-trval solutons [16]. In ths paper we analyze load balancng n data networks carryng elastc traffc, as consdered by [12]. Transfers n such networks can be represented by flows. We may dstngush between load balancng at the flow-level or the packet-level, dependng on whether an arrvng flow s entrely drected to a specfc route (that t uses untl the flow s fnshed) or a flow can be splt between several routes, respectvely. Ths paper deals wth packet-level load balancng,.e., we assume that packets of a flow can be dvded among several routes. Due to the dynamc nature of traffc, t s n general complcated to analyze the performance of such networks. Flows arrve accordng to some stochastc process and brng along a random amount of work. For each gven number of flows present n the system, the allocated servce rates are determned by some sharng polcy. As soon as the number of flows changes, t s assumed that these rates are adapted nstantly. We analyze a network n whch, besdes classes of users that use specfc routes, one class of users can splt ts traffc over several routes. We note that ths network s the smplest system to analyze the performance and potental gans of load balancng at the packet-level, and t s therefore of partcular nterest to gan nsght. In addton, ths system also accounts for rather explct results. We assume that packet-level load balancng s based on an alpha-far bandwdth sharng polcy as ntroduced n [13]. The famly of alpha-far polces covers several common notons of farness as specal cases, such as max-mn farness (α ), proportonal farness (α 1) and maxmum throughput (α 0). In [14] t has also been shown that the case α = 2, wth addtonal class weghts set nversely proportonal to the respectve round trp tmes, provdes a reasonable modelng abstracton for the bandwdth sharng realzed by TCP (Transmsson Control Protocol) n the Internet. We show that the above network has multple possble behavors. In partcular, we show that packet-level load balancng based on alpha-far sharng mples that some classes of users, dependng on the state of the network, share capacty accordng to some Dscrmnatory Processor Sharng (DPS) model, whereas each of the remanng classes of users behaves as n a sngle-class sngle-node model. The flow-level performance of the above network s compared wth that of a smlar network, where packet-level load balancng s based on balanced farness, so-called nsenstve load balancng at the packet-level. The term nsenstve refers to the fact that the correspondng steady-state dstrbuton depends on the traffc characterstcs through the traffc ntensty only. Insenstve load balancng at the flow-level was frst ntroduced n [4], and extended to nsenstve load balancng at the packet-level n [10]. Optmal nsenstve load balancng at the flow-level utlzng local state nformaton was addressed n [1]. In [8] t was shown that one can acheve 2

7 1 2 L Fgure 1: The bandwdth-sharng network. even better performance f capacty allocaton and load balancng are optmzed jontly. A comparson between packet-level and flow-level nsenstve load balancng was conducted n [11]. Assumng Posson arrvals and exponentally dstrbuted servce requrements, the dynamcs of the flow populaton may be descrbed by a Markov process under both packet-level load balancng polces. We derve closed-form expressons for the mean number of users of each class under nsenstve load balancng. Extensve smulaton experments show that these are also qute accurate approxmatons for the ones n a smlar network where load balancng s based on alpha-far sharng, for whch no explct expressons are avalable. The above results are n lne wth the fndngs of [3], n whch t was shown that the performance of networks operatng under unweghted max-mn farness, unweghted proportonal farness and balanced farness s nearly smlar. The results n ths paper suggest that balanced farness s n fact a reasonable approxmaton for all unweghted alpha-far sharng polces. The remander of ths paper s organzed as follows. In Secton 2 we frst provde a detaled model descrpton, and ntroduce balanced farness and alpha-far sharng. In the next secton we consder the model for a fxed flow populaton, and we characterze how bandwdth s allocated under both polces. In Secton 4 we consder the model at large tme-scales, such that the state of the network vares, and we derve explct expressons for the mean number of users under balanced farness, and show by conductng extensve smulaton experments that these provde accurate approxmatons for the ones under alpha-far sharng. In the next secton we examne the gan than one can acheve for both polces by usng packet-level load balancng nstead of statc or flow-level load balancng. Secton 6 concludes wth some fnal observatons. 2 Model We consder the network as depcted n Fgure 1. The network conssts of L nodes, where node has servce rate C, = 1,...,L. There are L + 1 classes of users. Class requres servce at node, = 1,...,L, whereas class 0 can be served at all nodes at the same tme,.e., class-0 users can splt ther traffc. We assume that class- users arrve accordng to a Posson process of rate λ, and have exponentally dstrbuted servce requrements wth mean µ 1, = 0,...,L. The arrval processes are all ndependent. The traffc load of class s then ρ = λ µ 1. Let n = (n 0,...,n L ) denote the state of the network, wth n representng the number of class- users. 3

8 2.1 Balanced farness We frst assume that the bandwdth s shared accordng to balanced farness, as ntroduced n [4]. Let φ (n) denote the servce rate allocated to class, = 0,...,L, wth balanced farness, when the network s n state n (here φ 0 (n) = L =1 φ 0(n)). These servce rates have to satsfy the balance condtons φ (n e j ) φ (n) = φ j(n e ) φ j (n),j = 0,...,L, n,n j > 0, (1) where e denotes the ( + 1)th unt vector n R L+1. All balanced servce rates can be expressed n terms of a unque balance functon Φ( ), so that Φ(0) = 1 and φ (n) = Φ(n e ) Φ(n) n : n > 0, = 0,...,L. (2) Hence, characterzaton of Φ(n) mples that φ (n) s characterzed as well. Defne Φ(n) = 0 f n / N L+1 0. In order to obtan Φ(n), we need to solve the followng maxmzaton problem for each n N L+1 0 \{0}: (BF) max Φ(n) 1 s.t. L j=1 φ 0j (n) = Φ(n e 0) Φ(n) φ (n) = Φ(n e ), Φ(n) = 1,...,L φ 0 (n) + φ (n) C, = 1,...,L φ 0 (n),φ (n) 0, = 1,...,L. It s clear that Φ(n) can be obtaned recursvely: the Φ(n e )s are requred to determne Φ(n). Also note that (BF) s a smple LP-problem, whch can be solved usng standard LP algorthms. In Secton 3.1, however, we solve (BF) by rewrtng the LP-problem n terms of a related network. 2.2 Alpha-far sharng We next assume that the network operates under a so-called alpha-far sharng polcy, as ntroduced n [13]. When the network s n state n 0, the servce rate x allocated to each of the class- users s obtaned by solvng the followng optmzaton problem: (AF) max G(x) s.t. n 0 x 0 + n x C, = 1,...,L x 0,x 0, = 1,...,L, where the objectve functon G(x) s defned by { ( L =1 n G(x) := 0 κ x 0) 1 α 0 1 α + L =1 n x κ 1 α 1 α f α (0, )\{1}; n 0 κ 0 log( L =1 x 0) + L =1 n κ log(x ) f α = 1. 4

9 The κ s are non-negatve class weghts, and α (0, ) may be nterpreted as a farness coeffcent. The cases α 0, α 1 and α correspond to allocatons whch acheve maxmum throughput, proportonal farness, and max-mn farness, respectvely. The value of x 0 denotes how much capacty s assgned to path (that requres servce at node ) of class 0. Here x 0 = L =1 x 0 denotes how much capacty s assgned to a sngle class-0 user n the network. Let s (n) := x n denote the total servce rate allocated to class, = 0,...,L. 3 Statc settng In ths secton we consder the model for a fxed flow populaton,.e., the state n N L+1 0 \{0} s fxed, and we derve how bandwdth s shared between the varous classes n case of balanced farness and alpha-far sharng, respectvely. The dffculty n solvng problem (BF) and (AF), as presented n the prevous secton, s that no explct expressons are avalable for ther optmal solutons. We frst show that the network depcted n Fgure 1 s equvalent to another network. In order to do so, let us frst ntroduce the noton of the capacty set. The allocatons φ(n) = (φ 0 (n),...,φ L (n)) and s(n) = (s 0 (n),...,s L (n)) are clearly constraned by the capacty set C R L+1 C := x 0 : a 1,...,a L 0, + : L a j = 1, a x 0 + x C, = 1,...,L, j=1.e., φ(n) C and s(n) C for all n N L+1 0. It s straghtforward to show that the capacty set C can also be expressed as L C := x 0 : L x j C j, x C, = 1,...,L, j=0 j=1.e., C = C. Snce C s the capacty set correspondng to the tree network depcted n Fgure 2, t follows that the networks depcted n Fgures 1 and 2 are n fact equvalent. The tree has a common lnk wth capacty C C L, and L + 1 branches wth capactes, C 1,...,C L, respectvely. In ths network class- users requre servce at the node wth servce rate C and at the common lnk, = 1,...,L, whereas class-0 users only requre servce at the common lnk. Note that each class of users corresponds to a specfc route n the tree network. As a sde remark we menton that n general t s not true that a network (where some classes of users can splt ther traffc over several routes at the same tme) can be converted n a tree network. In fact, f we extend the model depcted n Fgure 1 by addng a class of users that requres servce at all L nodes smultaneously, then t s already not possble to represent the network as a tree network. However, we note that n general one may stll be able to convert a traffc-splttng network n some other network (wth dummy nodes) wthout traffc splttng. 3.1 Balanced farness In ths subsecton we derve the balanced farness allocaton by solvng problem (BF). Snce the models depcted n Fgures 1 and 2 are equvalent, t follows that the balance functon Φ( ) 5

10 class 0 C 1 class 1 L =1 C C L 1 classl 1 C L class L Fgure 2: Tree network correspondng to tree network concdes wth Φ( ),.e., Φ( ) = Φ( ), see [3]. In the followng lemma we present the soluton of the optmzaton problem (BF). Lemma 3.1 The balanced farness functon Φ(n) satsfes, wth Φ(0) = 1, { } Φ(n e 1 ) Φ(n) = max,..., Φ(n e L), C 1 C L L =0 Φ(n e ) L =1 C, n N L+1 0 \{0}. (3) Proof: From the above t follows that we can obtan Φ( ) by determnng Φ( ), as they are the same. Subsequently, Φ( ) s obtaned by usng Equaton (2) n [5]. We note that Lemma 3.1 s n agreement wth Equaton (19) n [10]. From Lemma 3.1 t follows that Φ(n) can be obtaned recursvely. The total servce rate allocated to class, = 0,...,L, n each state n N L+1 0 can be obtaned usng Lemma 3.1 and (2). 3.2 Alpha-far sharng In ths subsecton we focus on the alpha-far allocaton, that s obtaned by solvng problem (AF). Smlar to the prevous subsecton, we can obtan the alpha-far allocaton s(n) by determnng the alpha-far allocaton s(n) n the tree network, as both networks are the same, mplyng that s(n) = s(n). In order to obtan s(n) we need to solve the followng maxmzaton problem: (AF 2) max H(x) L s.t. n x =0 n x C, L C, =1 = 1,...,L, x 0, = 0,...,L, (4) 6

11 where the objectve functon H(x) s defned by H(x) := { L =0 n x κ 1 α 1 α f α (0, )\{1}; L =0 n κ log(x ) f α = 1. Below we show that (AF 2) s solvable, but the optmal soluton strongly depends on the state n 0. We present a smple algorthm for obtanng the alpha-far allocaton. Lemma 3.2 The alpha-far allocaton s(n) can be obtaned wth the followng algorthm: Set Stop:=False Set S := {0,...,L} WHILE Stop=False DO Determne the S -class DPS allocaton: s (n) := n IF s (n) C for all S\{0} THEN set Stop:=True ELSE Take any S\{0} such that s (n) > C Set S := S\{ } Set s (n) := C END END j S\{0} C j j S n j j, S Proof: Frst consder the Karush-Kuhn-Tucker (KKT) necessary condtons for problem (AF 2). If x s an optmal soluton to problem (AF2), then there exst constants p 0, = 0,...,L, such that, n 0 κ 0 x α 0 n κ x α n 0 p 0 ; (5) n (p 0 + p ), = 1,...,L; (6) ( L p 0 C =1 ) L n x = 0; (7) =0 p (C n x ) = 0, = 1,...,L. (8) Note that (5) and (6) hold for any α (0, ). Solvng (5)-(8) for (x 0,...,x L ) and (p 0,...,p L ) yelds L L! q=1 q!(l q)! = 2L 1 possble solutons, however, dependng on the state of the network n, only one of the 2 L 1 solutons, x, s such that p 0, = 0,...,L,.e., ths s the optmal soluton for (AF2). For each of the other solutons there exsts at least one Lagrange parameter that s negatve, mplyng that these solutons cannot be optmal. Note that the exstence of a unque optmal soluton x for (AF2) also follows as H(x) s strctly concave and the constrants are lnear. Straghtforward calculus shows that the correspondng alphafar allocaton s (n) = s (n) = n x, = 0,...,L, can be obtaned by the above algorthm. The algorthm reflects that 2 L 1 solutons exst for (5)-(8), but t also shows that only one of these solutons, x, s found after termnaton of the algorthm. The Lagrange parameters 7

12 correspondng to x are such that p = 0 f S\{0}, and p > 0 f / S\{0}, where S s the set obtaned after termnaton of the algorthm. Furthermore, p 0 = 0 f n 0 = 0 and f there exsts an such that n = 0, = 1,...,L, otherwse p 0 > 0. 4 Flow-level dynamcs In the prevous secton we consdered the model for a fxed flow populaton, and we derved expressons for the balanced farness and alpha-far allocatons n each state of the network. In ths secton we analyze the model at suffcently large tme scales. In ths case we also have to take the random nature of the traffc nto account,.e., the state of the network n vares at large tme scales. 4.1 Balanced farness Let N(t) = (N 0 (t),...,n L (t)) denote the state of the network at tme t. Snce we assumed Posson arrvals and exponentally dstrbuted servce requrements, N(t) s a Markov process wth transton rates: q(n,n + e ) = λ ; q(n,n e ) = µ φ (n), = 0,...,L, n case of balanced farness. In [3] t was shown that the process N(t) s stable f there exsts ( ρ 01,..., ρ 0L ) such that L =1 or equvalently, f ρ 0 = ρ 0 and ρ 0 + ρ < C, = 1,...,L, (9) L ρ < =0 L C j and ρ < C, = 1,...,L. (10) j=1 It may be verfed from (1) that the steady-state queue length dstrbuton s gven by π(n) = 1 L G(ρ) Φ(n) ρ n, n N L+1 0, (11) =0 where the normalzaton constant G(ρ) equals G(ρ) = G(ρ 0,...,ρ L ) =... n 0 =0 n L =0 Φ(n) As a sde remark we menton that (11) n fact holds for much more general traffc characterstcs, see [4] for a more detaled treatment. When applyng Lttle s formula we fnd that L =0 ρ n. EN BF = ρ G(ρ) ρ G(ρ) = ρ log G(ρ), = 0,...,L, (12) ρ 8

13 .e., characterzaton of G(ρ) mples that EN BF, = 0,...,L, s known as well. By explotng the results of [6] on tree networks we can determne G(ρ), and t can be verfed that ths results n G(ρ) = 1 1 L =0 ρ L =1 C L =1 ρ 1 L =1 C ( ). (13) L =1 1 ρ C Then by usng (12) we can obtan a closed-form expresson for EN BF, = 0,...,L. The expresson for EN BF, = 1,...,L, s n general qute complcated, n contrast to the expresson for the mean number of class-0 users, whch s gven by EN BF 0 = ρ 0 L =1 C L =0 ρ. From (13) t follows that EN BF, = 0,...,L, s fnte f the stablty condton (10) holds. 4.2 Alpha-far sharng As before, let N(t) = (N 0 (t),...,n L (t)) denote the state of the network at tme t. In case of alpha-far sharng N(t) s a Markov process wth transton rates: q(n,n + e ) = λ ; q(n,n e ) = µ s (n), = 0,...,L. Snce our network s equvalent to the tree network depcted n Fgure 2, t follows from Theorem 1 n [2] that the process N(t) s stable f (9) holds. Lemma 3.2 shows that, dependng on the state of the network n N L+1 0, the network has 2 L 1 possble behavors. Ths llustrates the complcaton of fndng closed-form expressons for the mean number of users of each class. In fact, so far no expressons for the mean number of users are avalable n case of alpha-far sharng. To gan some nsght, we derve n ths secton approxmatons for the mean number of users of each class,.e., EN AF, = 0,...,L. The approxmatons are valdated by means of smulaton experments. We consder the case where the network conssts of L = 2 nodes, but we note that the approxmatons can be extended to the case L > 2 n a smlar fashon. Usng Lemma 3.2 n Secton 3.2, t follows that the network, dependng on the state n, has three possble behavors: () f ( n 1 > C (κ2 ) 1/α ( ) ) 1/α 1 κ0 n 2 + n 0, C 2 κ 1 κ 1 then classes 0 and 2 behave as n a two-class DPS model wth capacty C 2 and relatve weghts, = 0,2, whereas class 1 behaves as an M/M/1 queue wth arrval rate λ 1 and servce rate µ 1 C 1 ; () If n 1 < C 1 C 2 ( κ2 κ 1 ) 1/α n 2 ( κ0 κ 1 ) 1/α n 0, then classes 0 and 1 behave as n a two-class DPS model wth capacty C 1 and relatve weghts, = 0,1, whereas class 2 behaves as an M/M/1 queue wth arrval rate λ 2 and servce 9

14 rate µ 2 C 2 ; () otherwse the network wll behave as n a three-class DPS model wth capacty C 1 + C 2 and relatve weghts, = 0,1,2. If the network were to behave as () all the tme and f ρ 1 < C 1 and ρ 0 + ρ 2 < C 2 (stablty condtons), then by explotng the results of [7] we would obtan ( ) µ 0 ρ EN () 0 = EN () 1 = EN () 2 = EN () 0 = ρ C 2 ρ 0 ρ 2 ρ 1 C 1 ρ 1 ; ρ C 2 ρ 0 ρ 2 ρ C 1 ρ 0 ρ 1 0 µ 0 (C 2 ρ 0 ) + 2 µ 2 (C 2 ρ 2 ) ( ) µ 2 ρ µ 0 (C 2 ρ 0 ) + 2 µ 2 (C 2 ρ 2 ) Lkewse, when the network behaves as () and f ρ 2 < C 2 and ρ 0 +ρ 1 < C 1 (stablty condtons), we fnd ( ) µ 0 ρ EN () 1 = EN () 2 = ρ C 1 ρ 0 ρ 1 ρ 2 C 2 ρ 2. 0 µ 0 (C 1 ρ 0 ) + 1 µ 1 (C 1 ρ 1 ) ( ) µ 1 ρ µ 0 (C 1 ρ 0 ) + 1 µ 1 (C 1 ρ 1 ) If the network behaves as a three-class DPS model,.e., as (), and f ρ 0 + ρ 1 + ρ 2 < C 1 + C 2 (stablty condton), then one can obtan the mean number of users of each class by solvng the followng set of lnear equatons for EN (), = 0,1,2: (C 1 + C 2 )EN () λ 2 j=0 j λ j λ EN() + λ λ EN() j ;. ; ; j µ j + µ = ρ, = 0,1,2, where λ := λ 0 +λ 1 +λ 2, see [7]. In ths case there also exsts a closed-form expresson for EN (), = 0,1,2, but t s complcated. We propose the followng approxmaton: EN AF EN AP, = 0,1,2, where EN0 AP := EN () 0 ; EN1 AP := max{en () 1, EN() 1 }; EN2 AP := max{en () 2, EN () 2 }. It can be verfed that EN0 AP s bounded f ρ 0 +ρ 1 +ρ 2 < C 1 +C 2, EN1 AP s bounded f ρ 1 < C 1 and ρ 0 + ρ 1 + ρ 2 < C 1 + C 2, and EN2 AP s bounded f ρ 2 < C 2 and ρ 0 + ρ 1 + ρ 2 < C 1 + C 2. Hence, the EN AP s are only all bounded f (9) holds,.e., f the process N(t) s also stable. In [3] t was argued that the performance of a network under proportonal farness (α = 1) and max-mn farness (α ) s closely approxmated by that under balanced farness. Therefore, we also propose the followng approxmaton: EN AF EN BF, = 0,1,2. The value of EN BF, = 0,1,2, can be obtaned usng (12), and s ndependent of the value of α. 10

15 γ EN0 AF EN1 AF EN2 AF EN0 AP EN1 AP EN2 AP EN0 BF EN1 BF EN2 BF Table 1: Smulaton results for scenaro I. γ α EN0 AF EN1 AF EN2 AF EN0 AP EN1 AP EN2 AP EN0 BF EN1 BF EN2 BF Table 2: Smulaton results for scenaro II. To examne the accuracy of the above approxmatons we have performed smulaton experments. We consder the settng wth C 1 = C 2 = 1, and we take λ = γ, µ = 1, = 0,1,2, such that ρ 0 = ρ 1 = ρ 2 = γ. We frst consder scenaro I, where κ = 1, = 0,1,2. Subsequently, we consder scenaro II, where κ 0 = 5, κ 1 = 1 and κ 2 = 2. In scenaro II we let the traffc load γ and the alpha-far coeffcent α vary, whereas n scenaro I we only let γ vary, as t can be verfed that the EN AF s and EN AP s are ndependent of the value of α n scenaro I. To ensure stablty we assume that γ < 2 3. The results are reported n Tables 1 and 2. Each reported smulaton value n these (and other) tables s measured over events,.e., arrvals or departures. Remark: We have also determned a 95% confdence nterval (CI) for each lsted smulaton value n ths paper, but these are not presented. We note, however, that the relatve effcency,.e., the rato of the half-length of the CI to the reported smulaton value, s less than 3% for all lsted cases n Tables 1, 2, 5 and 6, and less than 10% for all lsted cases n Tables 3 and 4. Table 1 compares the value of EN AF and EN BF, = 0,1,2, for scenaro I. The results show that EN AF obtaned by smulaton wth the approxmatons EN AP EN AP, = 0,1,2. Also, 11

16 the table shows that EN0 AF EN0 BF and EN AF EN BF, = 1,2. Overall we see that both approxmatons are accurate n case of equal class weghts, especally for low traffc load. Table 2 reports the results correspondng to scenaro II,.e., n case of unequal class weghts. In ths case EN AF and EN AP do depend on the value of α, as s shown n the table. Agan, we see that EN AF EN AP, = 0,1,2. For low traffc loads both approxmatons perform qute well, but for hgh traffc loads we see that the balanced farness approxmaton s less accurate than the other one. Tables 1 and 2 show that EN AF EN AP, = 0,1,2, whch may be explaned as follows. Frst note that the rate allocated to class 1 s smaller than or equal to C 1 at all moments n tme under alpha-far sharng, whereas rate C 1 s contnuously avalable to class 1 n (). Clearly, ths mples that EN1 AF EN () 1. Wth smlar reasonng, we fnd that ENAF 2 EN () 2. Snce class- users cannot be allocated more than C, = 1,2, under alpha-far sharng, whereas n the three-class DPS model the upper bound s C 1 + C 2 for both classes, one may expect that EN AF EN (), = 1,2. For any state n N 3 0 \{0} t can be verfed that the alpha-far allocaton to class 0 s larger or equal than the one obtaned n the three-class DPS model, so one would expect EN0 AF EN () 0 at frst sght. However, recall that we argued that the number of users of classes 1 and 2 n the model operatng under alpha-far sharng wll (on average) be larger than n the three-class DPS model, whch causes that the total servce allocated to class 0 n the model operatng under alpha-far sharng s less than or equal to that n the three-class DPS model,.e., we may also expect EN0 AF EN () 0. The above reasonng ndeed suggests that EN AF EN AP, = 0,1, Flud and quas-statonary regmes To test the performance of the two approxmatons n case of extreme parameter values, we now assume that the flow dynamcs of the varous classes occur on wdely separate tme scales,.e., n flud and quas-statonary regmes. Formally, let λ (r) := λ f (r) and µ (r) := µ f (r), where f (r) represents the tme scale assocated wth class as functon of r, = 0,...,L. Note that the traffc ntensty of class equals ρ (r) := λ (r) /µ (r) = ρ, = 0,...,L, so t s ndependent of r. Let N (r) be the number of class- flows n the r-th system. Before analyzng the qualty of the approxmatons, we frst present the followng useful proposton. Proposton 4.1 Assume that L+1 classes of users share C unts of capacty accordng to DPS, where class has relatve weght κ, = 0,...,L. If f (r)/f 1 (r) 0 as r, = 1,...,L,.e., hgher ndexed classes operate on faster tme scales, then EN ( ) = ρ C L j= ρ 1 + j=0 κ j κ ( C L r=j ρ r ρ ρ ) ( j C L r=j+1 ρ r ), = 0,...,L. Proof: In [9] the above result was already proved for L = 1. For L > 1 the authors showed that EN ( ) j, j = 1,...,L, could be obtaned by determnng EN ( ), = 0,...,j 1,.e., as a recurson. Straghtforward calculus, however, shows that ths recurson reduces to the above result. 12

17 γ EN0 AF EN1 AF EN2 AF EN AP( ) 0 EN AP( ) 1 EN AP( ) 2 EN0 BF EN1 BF EN2 BF Table 3: Results correspondng to the flud and quas-statonary regmes (scenaro I). Let us return to the settng wth L = 2 nodes and L+1 = 3 classes of users. Proposton 4.1 allows us to obtan smple closed-form expressons for E AP, = 0,1,2, when r. Assumng that hgher ndexed classes operate on faster tme scales and that the stablty condtons (10) hold, we fnd EN AP( ) 0 := EN AP( ) 1 := max ρ 0 ; C 1 + C 2 ρ 0 ρ 1 ρ 2 { ρ 1 ρ 1 C 1 + C 2 ρ 1 ρ ρ 0ρ 1, ; C 1 ρ 1 1 (C 1 + C 2 ρ 0 ρ 1 ρ 2)(C 1 + C 2 ρ 1 ρ 2) { } EN AP( ) ρ 2 ρ 2 1 j ρ jρ 2 2 := max, + C 2 ρ 2 C 1 + C 2 ρ 2 j=0 2 (C 1 + C 2 2 r=j ρr)(c1 + C2 2 r=j+1 ρr). In case of equal class weghts, κ = κ, = 0,1,2, t s not hard to see that EN AP( ) 0 = ρ 0 C 1 + C 2 ρ 0 ρ 1 ρ 2 ; { EN AP( ) ρ1 1 = max, C 1 ρ 1 { EN AP( ) ρ2 2 = max, C 2 ρ 2 ρ 1 C 1 + C 2 ρ 0 ρ 1 ρ 2 ρ 2 C 1 + C 2 ρ 0 ρ 1 ρ 2 Clearly, the EN AP( ) s strongly depend on the orderng of the classes wth respect to the tme scales. In case of other orderngs than the one mentoned above one can obtan expressons n a smlar fashon. The accuracy of the approxmatons n the flud and quas-statonary regmes s examned by performng smulaton experments. We take C 1 = C 2 = 1, λ 0 = γ, λ 1 = 10γ, λ 2 = 100γ, µ 0 = 1, µ 1 = 10, µ 2 = 100, so that ρ = γ, = 0,1,2, and assume that hgher ndexed classes operate on faster tme scales. Tables 3 and 4 report the results for scenaro I and II, respectvely. Recall that the EN AF s and EN AP( ) s are ndependent of the value of α n scenaro I, whereas they are senstve to the value of α n scenaro II. The tables show that also n the flud and quas-statonary regmes the approxmatons are promsng. } ; }. } 5 Comparson wth statc and flow-level load balancng In the prevous sectons we consdered load balancng at the packet-level. In ths secton we quantfy how much better packet-level load balancng s than statc and flow-level load balancng. 13

18 γ α EN0 AF EN1 AF EN2 AF EN AP( ) 0 EN AP( ) 1 EN AP( ) 2 EN0 BF EN1 BF EN2 BF Table 4: Results correspondng to the flud and quas-statonary regmes (scenaro II). We consder the same parameter values as n the prevous secton (wthout consderng flud and quas-statonary regmes), and calculate the mean number of users of each class under statc and flow-level load balancng, so that we can make a comparson wth packet-level load balancng. As before, we frst assume that load balancng s based on balanced farness, and subsequently on alpha-far sharng. 5.1 Balanced farness When statc or flow-level load balancng s used, that s based on balanced farness, we now need to keep track of the number of class-0 users at node, = 1,2. Let n 0 denote the number of class- users at node, = 1,2. Then the balance functon s gven by (see [1]) Φ(n) = and we obtan φ 0 (n) = ( n 01 + n 1 n 1 )( n 02 + n 2 n 2 C n 1+n 01 1 C n 2+n 02 2 n 0 n 0 + n C ; φ (n) = ), n n 0 + n C, = 1,2. Hence, at both nodes capacty s shared accordng to egaltaran Processor Sharng (PS). Let us frst consder statc load balancng. Clearly, consderng the symmetrc parameter settng of the prevous secton, the optmal statc polcy s to route class-0 arrvals to node, = 1,2, wth probablty 1 2. Usng the parameter values of the prevous secton, we thus fnd 14

19 γ ENBF st 0 ENBF st 1 ENBF st 2 BF fl EN0 BF fl EN1 BF fl EN2 EN0 BF EN1 BF EN2 BF Table 5: Results for statc, flow-level and packet-level load balancng n case of balanced farness. that class- (class-0) users arrve accordng to a Posson process of rate γ ( 1 2γ) at node, and both class-0 and class- users have exponentally dstrbuted servce requrements wth mean 1, = 1,2. Recallng that C = 1, = 1,2, and snce capacty s shared accordng to PS at both nodes, t s a straghtforward exercse to show that EN BFst := γ 1 3 = 0,1,2, 2γ, where EN0 BFst denotes the mean number of class-0 users n the network (at node 1 or node 2). In Table 5 we report the EN BFst s for dfferent values of the load γ. Usng the closed-form expressons for EN BFst and EN BF, = 0,1,2, t s straghtforward to derve that EN BFst 0 EN BF 0 = 2; EN BFst EN BF = 4 4γ 4 5γ 1, = 1,2, gven that the load γ of each class s smaller than 2 3. In case of flow-level load balancng t s optmal (under the current settng) to route class-0 users to node 1 f n 01 +n 1 < n 02 +n 2, and to node 2 f n 01 +n 1 > n 02 +n 2. If n 01 +n 1 = n 02 +n 2 then an arrvng class-0 user s sent to node wth probablty 1 2, = 1,2. In other words, an arrvng class-0 user should jon the shortest queue, see [15]. Snce no explct expressons are known for the mean number of users EN BFfl of class, = 0,1,2, under flow-level load balancng, we have performed smulaton experments to obtan these values. The results are also reported n Table 5. Table 5 shows that packet-level load balancng outperforms both statc and flow-level load balancng, and flow-level load balancng s better than statc load balancng, as was expected, EN BFst, = 0,1,2. For low values of γ (low loads), the results are qute smlar, but for hgher loads the dfferences become more sgnfcant. We note that these results are n lne wth the fndngs of [11]..e., EN BF EN BFfl 5.2 Alpha-far sharng In case statc or flow-level load balancng s executed through alpha-far sharng, we also need to be aware of the number of class-0 users at nodes 1 and 2. In case n class- users and n 0 class-0 users are present at node, the allocated servce rates are s (n) = n C 0 n 0 + n, s 0(n) = 0 n 0 C 0 n 0 + n, = 1,2. 15

20 γ α ENAF st 0 ENAF st 1 ENAF st 2 AF fl EN0 AF fl EN1 AF fl EN2 EN0 AF EN1 AF EN2 AF Table 6: Results for statc, flow-level and packet-level load balancng n case of alpha-far sharng (scenaro II). Hence, capacty s shared accordng to DPS wth relatve weghts 0 and at node, = 1,2. Agan, due to symmetrc parameter values, n case of statc load balancng t s optmal to route class-0 arrvals to node, = 1,2, wth probablty 1 2. Usng the parameter values of the prevous secton, we thus fnd that class- (class-0) users arrve accordng to a Posson process of rate γ ( 1 2γ) at node, and both class-0 and class- users have exponentally dstrbuted servce requrements wth mean 1, = 1,2. Usng that C = 1, = 1,2, and snce capacty s shared accordng to DPS at both nodes, the results of [7] mply that ( ) EN0 AFst 1 2 := γ γ γ (1 1 2γ) + κ1/α1 (1 γ) + EN AFst := γ γ 1 + ( 1 2 γ 0 0 (1 1 2γ) + κ1/α (1 γ) ) ( ) γ (1 1 2γ) + κ1/α 2 (1 γ), = 1,2. Note that EN AFst = EN BFst, = 0,1,2, n case of equal class weghts. Therefore, we only focus on scenaro II, and these results are shown n Table 6. The optmal flow-level load balancng polcy s as before to jon the shortest queue, see [15]. As no explct expressons for the mean number of users EN AFfl of class, = 0,1,2, are avalable under flow-level load balancng, we resort to smulaton experments to obtan these values. Note that EN AFfl = EN BFfl, = 0,1,2, n case of equal class weghts, so we only report the results correspondng to scenaro II, see Table ;

21 Tables 6 shows that packet-level load balancng performs better than both statc and flowlevel load balancng: EN AF EN AFfl EN AFst, = 0,1,2. Agan, the results seem to vary more n case of hgh values of γ. 6 Concluson We analyzed a network consstng of L nodes, wth L + 1 classes of users. Class- users requre servce at node only, = 1,...,L, whereas class-0 users can splt ther traffc over the L nodes. We consdered load balancng at the packet-level, mplyng that class-0 users can splt ther traffc over the L nodes at the same tme. We assumed that load balancng was based on balanced farness and an alpha-far bandwdth sharng polcy, respectvely. We characterzed how bandwdth s allocated n each state of the network under these two polces. Assumng Posson arrvals and exponentally dstrbuted servce requrements, we derved expressons (approxmatons) for the mean number of users of each class under these two polces. For both polces we also showed that one can acheve sgnfcant performance gans f one performs packet-level load balancng nstead of statc or flow-level load balancng, especally for hghly loaded systems. A topc for further research s extendng the results to a more general network, e.g., so-called lnear networks where some classes can splt ther traffc over multple nodes at the same tme. In ths case t s consderably harder, f possble at all, to derve expressons for the mean number of users of each class under the above-mentoned polces, as the network does not reduce to a tree network. Acknowledgments The author wshes to thank Sem Borst and Mchel Mandjes for valuable dscussons. References [1] T. Bonald, M. Jonckheere, A. Proutère (2004). Insenstve load balancng. In: Proceedngs of the ACM SIGMETRICS/Performance 2004 Conference, New York, USA, [2] T. Bonald, L. Massoulé (2001). Impact of farness on Internet performance. In: Proceedngs of the ACM SIGMETRICS/Performance 2001 Conference, Boston, USA, [3] T. Bonald, L. Massoulé, A. Proutère, J. Vrtamo (2006). A queueng analyss of max-mn farness, proportonal farness and balanced farness. Queueng Systems and Applcatons, 53: [4] T. Bonald, A. Proutère (2003). Insenstve bandwdth sharng n data networks. Queueng Systems, 44: [5] T. Bonald, A. Proutère, J.W. Roberts, J. Vrtamo (2003). Computatonal aspects of balanced farness. In: Proceedngs of the 18th Internatonal Teletraffc Congress, Berln, Germany,

22 [6] T. Bonald, J. Vrtamo (2004). Calculatng the flow level performance of balanced farness n tree networks Performance Evaluaton, 58: [7] G. Fayolle, I. Mtran, R. Iasnogorodsk (1980). Sharng a processor among many job classes. Journal of the ACM, 27: [8] M. Jonckheere, J. Vrtamo (2005). Optmal nsenstve routng and bandwdth sharng n smple data networks. In: Proceedngs of the ACM SIGMETRICS 2005 Conference, Banff, Canada, [9] G. van Kessel, R. Nunez-Queja, S. Borst (2005). Dfferentated bandwdth sharng wth dsparate flow szes. In: Proceedngs of the IEEE INFOCOM 2005 Conference, Mam, USA, [10] J. Leno, J. Vrtamo (2005). Insenstve traffc splttng n data networks. In: Proceedngs of the 19th Internatonal Teletraffc Congress, [11] J. Leno, J. Vrtamo (2006). Insenstve load balancng n data networks. Computer Networks, 50: [12] L. Massoulé, J.W. Roberts (2000). Bandwdth sharng and admsson control for elastc traffc. Telecommuncaton Systems, 15: [13] J. Mo, J. Walrand (2000). Far end-to-end wndow-based congeston control. IEEE/ACM Transactons on Networkng, 8: [14] J. Padhye, V. Frou, D. Towsley, J. Kurose (2000). Modelng TCP Reno performance: A smple model and ts emprcal valdaton. IEEE/ACM Transactons on Networkng, 8: [15] D. Towsley, P.D. Sparaggs, C.G. Cassandras (1992). Optmal routng and buffer allocaton for a class of fnte capacty queueng systems. IEEE Transactons on Automatc Control, 37: [16] W. Whtt (1986). Decdng whch queue to jon: some counterexamples. Operatons Research, 34:

Queueing Networks II Network Performance

Queueing Networks II Network Performance Queueng Networks II Network Performance Davd Tpper Assocate Professor Graduate Telecommuncatons and Networkng Program Unversty of Pttsburgh Sldes 6 Networks of Queues Many communcaton systems must be modeled

More information

ECE559VV Project Report

ECE559VV Project Report ECE559VV Project Report (Supplementary Notes Loc Xuan Bu I. MAX SUM-RATE SCHEDULING: THE UPLINK CASE We have seen (n the presentaton that, for downlnk (broadcast channels, the strategy maxmzng the sum-rate

More information

Resource Allocation with a Budget Constraint for Computing Independent Tasks in the Cloud

Resource Allocation with a Budget Constraint for Computing Independent Tasks in the Cloud Resource Allocaton wth a Budget Constrant for Computng Independent Tasks n the Cloud Wemng Sh and Bo Hong School of Electrcal and Computer Engneerng Georga Insttute of Technology, USA 2nd IEEE Internatonal

More information

Lecture 4: November 17, Part 1 Single Buffer Management

Lecture 4: November 17, Part 1 Single Buffer Management Lecturer: Ad Rosén Algorthms for the anagement of Networs Fall 2003-2004 Lecture 4: November 7, 2003 Scrbe: Guy Grebla Part Sngle Buffer anagement In the prevous lecture we taled about the Combned Input

More information

TCOM 501: Networking Theory & Fundamentals. Lecture 7 February 25, 2003 Prof. Yannis A. Korilis

TCOM 501: Networking Theory & Fundamentals. Lecture 7 February 25, 2003 Prof. Yannis A. Korilis TCOM 501: Networkng Theory & Fundamentals Lecture 7 February 25, 2003 Prof. Yanns A. Korls 1 7-2 Topcs Open Jackson Networks Network Flows State-Dependent Servce Rates Networks of Transmsson Lnes Klenrock

More information

Minimisation of the Average Response Time in a Cluster of Servers

Minimisation of the Average Response Time in a Cluster of Servers Mnmsaton of the Average Response Tme n a Cluster of Servers Valery Naumov Abstract: In ths paper, we consder task assgnment problem n a cluster of servers. We show that optmal statc task assgnment s tantamount

More information

Module 9. Lecture 6. Duality in Assignment Problems

Module 9. Lecture 6. Duality in Assignment Problems Module 9 1 Lecture 6 Dualty n Assgnment Problems In ths lecture we attempt to answer few other mportant questons posed n earler lecture for (AP) and see how some of them can be explaned through the concept

More information

Structure and Drive Paul A. Jensen Copyright July 20, 2003

Structure and Drive Paul A. Jensen Copyright July 20, 2003 Structure and Drve Paul A. Jensen Copyrght July 20, 2003 A system s made up of several operatons wth flow passng between them. The structure of the system descrbes the flow paths from nputs to outputs.

More information

Suggested solutions for the exam in SF2863 Systems Engineering. June 12,

Suggested solutions for the exam in SF2863 Systems Engineering. June 12, Suggested solutons for the exam n SF2863 Systems Engneerng. June 12, 2012 14.00 19.00 Examner: Per Enqvst, phone: 790 62 98 1. We can thnk of the farm as a Jackson network. The strawberry feld s modelled

More information

Numerical Heat and Mass Transfer

Numerical Heat and Mass Transfer Master degree n Mechancal Engneerng Numercal Heat and Mass Transfer 06-Fnte-Dfference Method (One-dmensonal, steady state heat conducton) Fausto Arpno f.arpno@uncas.t Introducton Why we use models and

More information

3.1 Expectation of Functions of Several Random Variables. )' be a k-dimensional discrete or continuous random vector, with joint PMF p (, E X E X1 E X

3.1 Expectation of Functions of Several Random Variables. )' be a k-dimensional discrete or continuous random vector, with joint PMF p (, E X E X1 E X Statstcs 1: Probablty Theory II 37 3 EPECTATION OF SEVERAL RANDOM VARIABLES As n Probablty Theory I, the nterest n most stuatons les not on the actual dstrbuton of a random vector, but rather on a number

More information

6. Stochastic processes (2)

6. Stochastic processes (2) 6. Stochastc processes () Lect6.ppt S-38.45 - Introducton to Teletraffc Theory Sprng 5 6. Stochastc processes () Contents Markov processes Brth-death processes 6. Stochastc processes () Markov process

More information

6. Stochastic processes (2)

6. Stochastic processes (2) Contents Markov processes Brth-death processes Lect6.ppt S-38.45 - Introducton to Teletraffc Theory Sprng 5 Markov process Consder a contnuous-tme and dscrete-state stochastc process X(t) wth state space

More information

Problem Set 9 Solutions

Problem Set 9 Solutions Desgn and Analyss of Algorthms May 4, 2015 Massachusetts Insttute of Technology 6.046J/18.410J Profs. Erk Demane, Srn Devadas, and Nancy Lynch Problem Set 9 Solutons Problem Set 9 Solutons Ths problem

More information

Winter 2008 CS567 Stochastic Linear/Integer Programming Guest Lecturer: Xu, Huan

Winter 2008 CS567 Stochastic Linear/Integer Programming Guest Lecturer: Xu, Huan Wnter 2008 CS567 Stochastc Lnear/Integer Programmng Guest Lecturer: Xu, Huan Class 2: More Modelng Examples 1 Capacty Expanson Capacty expanson models optmal choces of the tmng and levels of nvestments

More information

Module 3 LOSSY IMAGE COMPRESSION SYSTEMS. Version 2 ECE IIT, Kharagpur

Module 3 LOSSY IMAGE COMPRESSION SYSTEMS. Version 2 ECE IIT, Kharagpur Module 3 LOSSY IMAGE COMPRESSION SYSTEMS Verson ECE IIT, Kharagpur Lesson 6 Theory of Quantzaton Verson ECE IIT, Kharagpur Instructonal Objectves At the end of ths lesson, the students should be able to:

More information

Simultaneous Optimization of Berth Allocation, Quay Crane Assignment and Quay Crane Scheduling Problems in Container Terminals

Simultaneous Optimization of Berth Allocation, Quay Crane Assignment and Quay Crane Scheduling Problems in Container Terminals Smultaneous Optmzaton of Berth Allocaton, Quay Crane Assgnment and Quay Crane Schedulng Problems n Contaner Termnals Necat Aras, Yavuz Türkoğulları, Z. Caner Taşkın, Kuban Altınel Abstract In ths work,

More information

The Minimum Universal Cost Flow in an Infeasible Flow Network

The Minimum Universal Cost Flow in an Infeasible Flow Network Journal of Scences, Islamc Republc of Iran 17(2): 175-180 (2006) Unversty of Tehran, ISSN 1016-1104 http://jscencesutacr The Mnmum Unversal Cost Flow n an Infeasble Flow Network H Saleh Fathabad * M Bagheran

More information

MMA and GCMMA two methods for nonlinear optimization

MMA and GCMMA two methods for nonlinear optimization MMA and GCMMA two methods for nonlnear optmzaton Krster Svanberg Optmzaton and Systems Theory, KTH, Stockholm, Sweden. krlle@math.kth.se Ths note descrbes the algorthms used n the author s 2007 mplementatons

More information

College of Computer & Information Science Fall 2009 Northeastern University 20 October 2009

College of Computer & Information Science Fall 2009 Northeastern University 20 October 2009 College of Computer & Informaton Scence Fall 2009 Northeastern Unversty 20 October 2009 CS7880: Algorthmc Power Tools Scrbe: Jan Wen and Laura Poplawsk Lecture Outlne: Prmal-dual schema Network Desgn:

More information

More metrics on cartesian products

More metrics on cartesian products More metrcs on cartesan products If (X, d ) are metrc spaces for 1 n, then n Secton II4 of the lecture notes we defned three metrcs on X whose underlyng topologes are the product topology The purpose of

More information

CHAPTER 5 NUMERICAL EVALUATION OF DYNAMIC RESPONSE

CHAPTER 5 NUMERICAL EVALUATION OF DYNAMIC RESPONSE CHAPTER 5 NUMERICAL EVALUATION OF DYNAMIC RESPONSE Analytcal soluton s usually not possble when exctaton vares arbtrarly wth tme or f the system s nonlnear. Such problems can be solved by numercal tmesteppng

More information

COMPARISON OF SOME RELIABILITY CHARACTERISTICS BETWEEN REDUNDANT SYSTEMS REQUIRING SUPPORTING UNITS FOR THEIR OPERATIONS

COMPARISON OF SOME RELIABILITY CHARACTERISTICS BETWEEN REDUNDANT SYSTEMS REQUIRING SUPPORTING UNITS FOR THEIR OPERATIONS Avalable onlne at http://sck.org J. Math. Comput. Sc. 3 (3), No., 6-3 ISSN: 97-537 COMPARISON OF SOME RELIABILITY CHARACTERISTICS BETWEEN REDUNDANT SYSTEMS REQUIRING SUPPORTING UNITS FOR THEIR OPERATIONS

More information

Perfect Competition and the Nash Bargaining Solution

Perfect Competition and the Nash Bargaining Solution Perfect Competton and the Nash Barganng Soluton Renhard John Department of Economcs Unversty of Bonn Adenauerallee 24-42 53113 Bonn, Germany emal: rohn@un-bonn.de May 2005 Abstract For a lnear exchange

More information

Games of Threats. Elon Kohlberg Abraham Neyman. Working Paper

Games of Threats. Elon Kohlberg Abraham Neyman. Working Paper Games of Threats Elon Kohlberg Abraham Neyman Workng Paper 18-023 Games of Threats Elon Kohlberg Harvard Busness School Abraham Neyman The Hebrew Unversty of Jerusalem Workng Paper 18-023 Copyrght 2017

More information

Generalized Linear Methods

Generalized Linear Methods Generalzed Lnear Methods 1 Introducton In the Ensemble Methods the general dea s that usng a combnaton of several weak learner one could make a better learner. More formally, assume that we have a set

More information

Difference Equations

Difference Equations Dfference Equatons c Jan Vrbk 1 Bascs Suppose a sequence of numbers, say a 0,a 1,a,a 3,... s defned by a certan general relatonshp between, say, three consecutve values of the sequence, e.g. a + +3a +1

More information

Additional Codes using Finite Difference Method. 1 HJB Equation for Consumption-Saving Problem Without Uncertainty

Additional Codes using Finite Difference Method. 1 HJB Equation for Consumption-Saving Problem Without Uncertainty Addtonal Codes usng Fnte Dfference Method Benamn Moll 1 HJB Equaton for Consumpton-Savng Problem Wthout Uncertanty Before consderng the case wth stochastc ncome n http://www.prnceton.edu/~moll/ HACTproect/HACT_Numercal_Appendx.pdf,

More information

Chapter - 2. Distribution System Power Flow Analysis

Chapter - 2. Distribution System Power Flow Analysis Chapter - 2 Dstrbuton System Power Flow Analyss CHAPTER - 2 Radal Dstrbuton System Load Flow 2.1 Introducton Load flow s an mportant tool [66] for analyzng electrcal power system network performance. Load

More information

Global Sensitivity. Tuesday 20 th February, 2018

Global Sensitivity. Tuesday 20 th February, 2018 Global Senstvty Tuesday 2 th February, 28 ) Local Senstvty Most senstvty analyses [] are based on local estmates of senstvty, typcally by expandng the response n a Taylor seres about some specfc values

More information

A PROBABILITY-DRIVEN SEARCH ALGORITHM FOR SOLVING MULTI-OBJECTIVE OPTIMIZATION PROBLEMS

A PROBABILITY-DRIVEN SEARCH ALGORITHM FOR SOLVING MULTI-OBJECTIVE OPTIMIZATION PROBLEMS HCMC Unversty of Pedagogy Thong Nguyen Huu et al. A PROBABILITY-DRIVEN SEARCH ALGORITHM FOR SOLVING MULTI-OBJECTIVE OPTIMIZATION PROBLEMS Thong Nguyen Huu and Hao Tran Van Department of mathematcs-nformaton,

More information

How Bad Is Suboptimal Rate Allocation?

How Bad Is Suboptimal Rate Allocation? How Bad Is Suboptmal Rate Allocaton? Tan Lan, Xaojun Ln 2, Mung Chang, Ruby Lee Department of Electrcal Engneerng, Prnceton Unversty, NJ 08544, USA 2 School of Electrcal and Computer Engneerng, Purdue

More information

ANSWERS. Problem 1. and the moment generating function (mgf) by. defined for any real t. Use this to show that E( U) var( U)

ANSWERS. Problem 1. and the moment generating function (mgf) by. defined for any real t. Use this to show that E( U) var( U) Econ 413 Exam 13 H ANSWERS Settet er nndelt 9 deloppgaver, A,B,C, som alle anbefales å telle lkt for å gøre det ltt lettere å stå. Svar er gtt . Unfortunately, there s a prntng error n the hnt of

More information

Application of Queuing Theory to Waiting Time of Out-Patients in Hospitals.

Application of Queuing Theory to Waiting Time of Out-Patients in Hospitals. Applcaton of Queung Theory to Watng Tme of Out-Patents n Hosptals. R.A. Adeleke *, O.D. Ogunwale, and O.Y. Hald. Department of Mathematcal Scences, Unversty of Ado-Ekt, Ado-Ekt, Ekt State, Ngera. E-mal:

More information

LOW BIAS INTEGRATED PATH ESTIMATORS. James M. Calvin

LOW BIAS INTEGRATED PATH ESTIMATORS. James M. Calvin Proceedngs of the 007 Wnter Smulaton Conference S G Henderson, B Bller, M-H Hseh, J Shortle, J D Tew, and R R Barton, eds LOW BIAS INTEGRATED PATH ESTIMATORS James M Calvn Department of Computer Scence

More information

On the Multicriteria Integer Network Flow Problem

On the Multicriteria Integer Network Flow Problem BULGARIAN ACADEMY OF SCIENCES CYBERNETICS AND INFORMATION TECHNOLOGIES Volume 5, No 2 Sofa 2005 On the Multcrtera Integer Network Flow Problem Vassl Vasslev, Marana Nkolova, Maryana Vassleva Insttute of

More information

2.3 Nilpotent endomorphisms

2.3 Nilpotent endomorphisms s a block dagonal matrx, wth A Mat dm U (C) In fact, we can assume that B = B 1 B k, wth B an ordered bass of U, and that A = [f U ] B, where f U : U U s the restrcton of f to U 40 23 Nlpotent endomorphsms

More information

Chapter 13: Multiple Regression

Chapter 13: Multiple Regression Chapter 13: Multple Regresson 13.1 Developng the multple-regresson Model The general model can be descrbed as: It smplfes for two ndependent varables: The sample ft parameter b 0, b 1, and b are used to

More information

Chapter 5. Solution of System of Linear Equations. Module No. 6. Solution of Inconsistent and Ill Conditioned Systems

Chapter 5. Solution of System of Linear Equations. Module No. 6. Solution of Inconsistent and Ill Conditioned Systems Numercal Analyss by Dr. Anta Pal Assstant Professor Department of Mathematcs Natonal Insttute of Technology Durgapur Durgapur-713209 emal: anta.bue@gmal.com 1 . Chapter 5 Soluton of System of Lnear Equatons

More information

Lecture 10 Support Vector Machines II

Lecture 10 Support Vector Machines II Lecture 10 Support Vector Machnes II 22 February 2016 Taylor B. Arnold Yale Statstcs STAT 365/665 1/28 Notes: Problem 3 s posted and due ths upcomng Frday There was an early bug n the fake-test data; fxed

More information

Linear Approximation with Regularization and Moving Least Squares

Linear Approximation with Regularization and Moving Least Squares Lnear Approxmaton wth Regularzaton and Movng Least Squares Igor Grešovn May 007 Revson 4.6 (Revson : March 004). 5 4 3 0.5 3 3.5 4 Contents: Lnear Fttng...4. Weghted Least Squares n Functon Approxmaton...

More information

Economics 101. Lecture 4 - Equilibrium and Efficiency

Economics 101. Lecture 4 - Equilibrium and Efficiency Economcs 0 Lecture 4 - Equlbrum and Effcency Intro As dscussed n the prevous lecture, we wll now move from an envronment where we looed at consumers mang decsons n solaton to analyzng economes full of

More information

Computing Correlated Equilibria in Multi-Player Games

Computing Correlated Equilibria in Multi-Player Games Computng Correlated Equlbra n Mult-Player Games Chrstos H. Papadmtrou Presented by Zhanxang Huang December 7th, 2005 1 The Author Dr. Chrstos H. Papadmtrou CS professor at UC Berkley (taught at Harvard,

More information

Assortment Optimization under MNL

Assortment Optimization under MNL Assortment Optmzaton under MNL Haotan Song Aprl 30, 2017 1 Introducton The assortment optmzaton problem ams to fnd the revenue-maxmzng assortment of products to offer when the prces of products are fxed.

More information

Formulas for the Determinant

Formulas for the Determinant page 224 224 CHAPTER 3 Determnants e t te t e 2t 38 A = e t 2te t e 2t e t te t 2e 2t 39 If 123 A = 345, 456 compute the matrx product A adj(a) What can you conclude about det(a)? For Problems 40 43, use

More information

Psychology 282 Lecture #24 Outline Regression Diagnostics: Outliers

Psychology 282 Lecture #24 Outline Regression Diagnostics: Outliers Psychology 282 Lecture #24 Outlne Regresson Dagnostcs: Outlers In an earler lecture we studed the statstcal assumptons underlyng the regresson model, ncludng the followng ponts: Formal statement of assumptons.

More information

Appendix B: Resampling Algorithms

Appendix B: Resampling Algorithms 407 Appendx B: Resamplng Algorthms A common problem of all partcle flters s the degeneracy of weghts, whch conssts of the unbounded ncrease of the varance of the mportance weghts ω [ ] of the partcles

More information

Real-Time Systems. Multiprocessor scheduling. Multiprocessor scheduling. Multiprocessor scheduling

Real-Time Systems. Multiprocessor scheduling. Multiprocessor scheduling. Multiprocessor scheduling Real-Tme Systems Multprocessor schedulng Specfcaton Implementaton Verfcaton Multprocessor schedulng -- -- Global schedulng How are tasks assgned to processors? Statc assgnment The processor(s) used for

More information

Errors for Linear Systems

Errors for Linear Systems Errors for Lnear Systems When we solve a lnear system Ax b we often do not know A and b exactly, but have only approxmatons  and ˆb avalable. Then the best thng we can do s to solve ˆx ˆb exactly whch

More information

Analysis of Queuing Delay in Multimedia Gateway Call Routing

Analysis of Queuing Delay in Multimedia Gateway Call Routing Analyss of Queung Delay n Multmeda ateway Call Routng Qwe Huang UTtarcom Inc, 33 Wood Ave. outh Iseln, NJ 08830, U..A Errol Lloyd Computer Informaton cences Department, Unv. of Delaware, Newark, DE 976,

More information

Research Article Green s Theorem for Sign Data

Research Article Green s Theorem for Sign Data Internatonal Scholarly Research Network ISRN Appled Mathematcs Volume 2012, Artcle ID 539359, 10 pages do:10.5402/2012/539359 Research Artcle Green s Theorem for Sgn Data Lous M. Houston The Unversty of

More information

CIE4801 Transportation and spatial modelling Trip distribution

CIE4801 Transportation and spatial modelling Trip distribution CIE4801 ransportaton and spatal modellng rp dstrbuton Rob van Nes, ransport & Plannng 17/4/13 Delft Unversty of echnology Challenge the future Content What s t about hree methods Wth specal attenton for

More information

Negative Binomial Regression

Negative Binomial Regression STATGRAPHICS Rev. 9/16/2013 Negatve Bnomal Regresson Summary... 1 Data Input... 3 Statstcal Model... 3 Analyss Summary... 4 Analyss Optons... 7 Plot of Ftted Model... 8 Observed Versus Predcted... 10 Predctons...

More information

Supplement: Proofs and Technical Details for The Solution Path of the Generalized Lasso

Supplement: Proofs and Technical Details for The Solution Path of the Generalized Lasso Supplement: Proofs and Techncal Detals for The Soluton Path of the Generalzed Lasso Ryan J. Tbshran Jonathan Taylor In ths document we gve supplementary detals to the paper The Soluton Path of the Generalzed

More information

COS 521: Advanced Algorithms Game Theory and Linear Programming

COS 521: Advanced Algorithms Game Theory and Linear Programming COS 521: Advanced Algorthms Game Theory and Lnear Programmng Moses Charkar February 27, 2013 In these notes, we ntroduce some basc concepts n game theory and lnear programmng (LP). We show a connecton

More information

Pricing and Resource Allocation Game Theoretic Models

Pricing and Resource Allocation Game Theoretic Models Prcng and Resource Allocaton Game Theoretc Models Zhy Huang Changbn Lu Q Zhang Computer and Informaton Scence December 8, 2009 Z. Huang, C. Lu, and Q. Zhang (CIS) Game Theoretc Models December 8, 2009

More information

Lecture Notes on Linear Regression

Lecture Notes on Linear Regression Lecture Notes on Lnear Regresson Feng L fl@sdueducn Shandong Unversty, Chna Lnear Regresson Problem In regresson problem, we am at predct a contnuous target value gven an nput feature vector We assume

More information

NUMERICAL DIFFERENTIATION

NUMERICAL DIFFERENTIATION NUMERICAL DIFFERENTIATION 1 Introducton Dfferentaton s a method to compute the rate at whch a dependent output y changes wth respect to the change n the ndependent nput x. Ths rate of change s called the

More information

The Exact Formulation of the Inverse of the Tridiagonal Matrix for Solving the 1D Poisson Equation with the Finite Difference Method

The Exact Formulation of the Inverse of the Tridiagonal Matrix for Solving the 1D Poisson Equation with the Finite Difference Method Journal of Electromagnetc Analyss and Applcatons, 04, 6, 0-08 Publshed Onlne September 04 n ScRes. http://www.scrp.org/journal/jemaa http://dx.do.org/0.46/jemaa.04.6000 The Exact Formulaton of the Inverse

More information

The Study of Teaching-learning-based Optimization Algorithm

The Study of Teaching-learning-based Optimization Algorithm Advanced Scence and Technology Letters Vol. (AST 06), pp.05- http://dx.do.org/0.57/astl.06. The Study of Teachng-learnng-based Optmzaton Algorthm u Sun, Yan fu, Lele Kong, Haolang Q,, Helongang Insttute

More information

Lectures - Week 4 Matrix norms, Conditioning, Vector Spaces, Linear Independence, Spanning sets and Basis, Null space and Range of a Matrix

Lectures - Week 4 Matrix norms, Conditioning, Vector Spaces, Linear Independence, Spanning sets and Basis, Null space and Range of a Matrix Lectures - Week 4 Matrx norms, Condtonng, Vector Spaces, Lnear Independence, Spannng sets and Bass, Null space and Range of a Matrx Matrx Norms Now we turn to assocatng a number to each matrx. We could

More information

princeton univ. F 17 cos 521: Advanced Algorithm Design Lecture 7: LP Duality Lecturer: Matt Weinberg

princeton univ. F 17 cos 521: Advanced Algorithm Design Lecture 7: LP Duality Lecturer: Matt Weinberg prnceton unv. F 17 cos 521: Advanced Algorthm Desgn Lecture 7: LP Dualty Lecturer: Matt Wenberg Scrbe: LP Dualty s an extremely useful tool for analyzng structural propertes of lnear programs. Whle there

More information

2E Pattern Recognition Solutions to Introduction to Pattern Recognition, Chapter 2: Bayesian pattern classification

2E Pattern Recognition Solutions to Introduction to Pattern Recognition, Chapter 2: Bayesian pattern classification E395 - Pattern Recognton Solutons to Introducton to Pattern Recognton, Chapter : Bayesan pattern classfcaton Preface Ths document s a soluton manual for selected exercses from Introducton to Pattern Recognton

More information

On the correction of the h-index for career length

On the correction of the h-index for career length 1 On the correcton of the h-ndex for career length by L. Egghe Unverstet Hasselt (UHasselt), Campus Depenbeek, Agoralaan, B-3590 Depenbeek, Belgum 1 and Unverstet Antwerpen (UA), IBW, Stadscampus, Venusstraat

More information

The optimal delay of the second test is therefore approximately 210 hours earlier than =2.

The optimal delay of the second test is therefore approximately 210 hours earlier than =2. THE IEC 61508 FORMULAS 223 The optmal delay of the second test s therefore approxmately 210 hours earler than =2. 8.4 The IEC 61508 Formulas IEC 61508-6 provdes approxmaton formulas for the PF for smple

More information

The Order Relation and Trace Inequalities for. Hermitian Operators

The Order Relation and Trace Inequalities for. Hermitian Operators Internatonal Mathematcal Forum, Vol 3, 08, no, 507-57 HIKARI Ltd, wwwm-hkarcom https://doorg/0988/mf088055 The Order Relaton and Trace Inequaltes for Hermtan Operators Y Huang School of Informaton Scence

More information

EEL 6266 Power System Operation and Control. Chapter 3 Economic Dispatch Using Dynamic Programming

EEL 6266 Power System Operation and Control. Chapter 3 Economic Dispatch Using Dynamic Programming EEL 6266 Power System Operaton and Control Chapter 3 Economc Dspatch Usng Dynamc Programmng Pecewse Lnear Cost Functons Common practce many utltes prefer to represent ther generator cost functons as sngle-

More information

Foundations of Arithmetic

Foundations of Arithmetic Foundatons of Arthmetc Notaton We shall denote the sum and product of numbers n the usual notaton as a 2 + a 2 + a 3 + + a = a, a 1 a 2 a 3 a = a The notaton a b means a dvdes b,.e. ac = b where c s an

More information

A Hybrid Variational Iteration Method for Blasius Equation

A Hybrid Variational Iteration Method for Blasius Equation Avalable at http://pvamu.edu/aam Appl. Appl. Math. ISSN: 1932-9466 Vol. 10, Issue 1 (June 2015), pp. 223-229 Applcatons and Appled Mathematcs: An Internatonal Journal (AAM) A Hybrd Varatonal Iteraton Method

More information

Applied Stochastic Processes

Applied Stochastic Processes STAT455/855 Fall 23 Appled Stochastc Processes Fnal Exam, Bref Solutons 1. (15 marks) (a) (7 marks) The dstrbuton of Y s gven by ( ) ( ) y 2 1 5 P (Y y) for y 2, 3,... The above follows because each of

More information

Min Cut, Fast Cut, Polynomial Identities

Min Cut, Fast Cut, Polynomial Identities Randomzed Algorthms, Summer 016 Mn Cut, Fast Cut, Polynomal Identtes Instructor: Thomas Kesselhem and Kurt Mehlhorn 1 Mn Cuts n Graphs Lecture (5 pages) Throughout ths secton, G = (V, E) s a mult-graph.

More information

Remarks on the Properties of a Quasi-Fibonacci-like Polynomial Sequence

Remarks on the Properties of a Quasi-Fibonacci-like Polynomial Sequence Remarks on the Propertes of a Quas-Fbonacc-lke Polynomal Sequence Brce Merwne LIU Brooklyn Ilan Wenschelbaum Wesleyan Unversty Abstract Consder the Quas-Fbonacc-lke Polynomal Sequence gven by F 0 = 1,

More information

Analysis of Discrete Time Queues (Section 4.6)

Analysis of Discrete Time Queues (Section 4.6) Analyss of Dscrete Tme Queues (Secton 4.6) Copyrght 2002, Sanjay K. Bose Tme axs dvded nto slots slot slot boundares Arrvals can only occur at slot boundares Servce to a job can only start at a slot boundary

More information

Econ107 Applied Econometrics Topic 3: Classical Model (Studenmund, Chapter 4)

Econ107 Applied Econometrics Topic 3: Classical Model (Studenmund, Chapter 4) I. Classcal Assumptons Econ7 Appled Econometrcs Topc 3: Classcal Model (Studenmund, Chapter 4) We have defned OLS and studed some algebrac propertes of OLS. In ths topc we wll study statstcal propertes

More information

A SEPARABLE APPROXIMATION DYNAMIC PROGRAMMING ALGORITHM FOR ECONOMIC DISPATCH WITH TRANSMISSION LOSSES. Pierre HANSEN, Nenad MLADENOVI]

A SEPARABLE APPROXIMATION DYNAMIC PROGRAMMING ALGORITHM FOR ECONOMIC DISPATCH WITH TRANSMISSION LOSSES. Pierre HANSEN, Nenad MLADENOVI] Yugoslav Journal of Operatons Research (00) umber 57-66 A SEPARABLE APPROXIMATIO DYAMIC PROGRAMMIG ALGORITHM FOR ECOOMIC DISPATCH WITH TRASMISSIO LOSSES Perre HASE enad MLADEOVI] GERAD and Ecole des Hautes

More information

Chapter 11: Simple Linear Regression and Correlation

Chapter 11: Simple Linear Regression and Correlation Chapter 11: Smple Lnear Regresson and Correlaton 11-1 Emprcal Models 11-2 Smple Lnear Regresson 11-3 Propertes of the Least Squares Estmators 11-4 Hypothess Test n Smple Lnear Regresson 11-4.1 Use of t-tests

More information

Markov Chain Monte Carlo Lecture 6

Markov Chain Monte Carlo Lecture 6 where (x 1,..., x N ) X N, N s called the populaton sze, f(x) f (x) for at least one {1, 2,..., N}, and those dfferent from f(x) are called the tral dstrbutons n terms of mportance samplng. Dfferent ways

More information

Amiri s Supply Chain Model. System Engineering b Department of Mathematics and Statistics c Odette School of Business

Amiri s Supply Chain Model. System Engineering b Department of Mathematics and Statistics c Odette School of Business Amr s Supply Chan Model by S. Ashtab a,, R.J. Caron b E. Selvarajah c a Department of Industral Manufacturng System Engneerng b Department of Mathematcs Statstcs c Odette School of Busness Unversty of

More information

x = , so that calculated

x = , so that calculated Stat 4, secton Sngle Factor ANOVA notes by Tm Plachowsk n chapter 8 we conducted hypothess tests n whch we compared a sngle sample s mean or proporton to some hypotheszed value Chapter 9 expanded ths to

More information

Markov chains. Definition of a CTMC: [2, page 381] is a continuous time, discrete value random process such that for an infinitesimal

Markov chains. Definition of a CTMC: [2, page 381] is a continuous time, discrete value random process such that for an infinitesimal Markov chans M. Veeraraghavan; March 17, 2004 [Tp: Study the MC, QT, and Lttle s law lectures together: CTMC (MC lecture), M/M/1 queue (QT lecture), Lttle s law lecture (when dervng the mean response tme

More information

n α j x j = 0 j=1 has a nontrivial solution. Here A is the n k matrix whose jth column is the vector for all t j=0

n α j x j = 0 j=1 has a nontrivial solution. Here A is the n k matrix whose jth column is the vector for all t j=0 MODULE 2 Topcs: Lnear ndependence, bass and dmenson We have seen that f n a set of vectors one vector s a lnear combnaton of the remanng vectors n the set then the span of the set s unchanged f that vector

More information

DUE: WEDS FEB 21ST 2018

DUE: WEDS FEB 21ST 2018 HOMEWORK # 1: FINITE DIFFERENCES IN ONE DIMENSION DUE: WEDS FEB 21ST 2018 1. Theory Beam bendng s a classcal engneerng analyss. The tradtonal soluton technque makes smplfyng assumptons such as a constant

More information

FUZZY GOAL PROGRAMMING VS ORDINARY FUZZY PROGRAMMING APPROACH FOR MULTI OBJECTIVE PROGRAMMING PROBLEM

FUZZY GOAL PROGRAMMING VS ORDINARY FUZZY PROGRAMMING APPROACH FOR MULTI OBJECTIVE PROGRAMMING PROBLEM Internatonal Conference on Ceramcs, Bkaner, Inda Internatonal Journal of Modern Physcs: Conference Seres Vol. 22 (2013) 757 761 World Scentfc Publshng Company DOI: 10.1142/S2010194513010982 FUZZY GOAL

More information

ON A DETERMINATION OF THE INITIAL FUNCTIONS FROM THE OBSERVED VALUES OF THE BOUNDARY FUNCTIONS FOR THE SECOND-ORDER HYPERBOLIC EQUATION

ON A DETERMINATION OF THE INITIAL FUNCTIONS FROM THE OBSERVED VALUES OF THE BOUNDARY FUNCTIONS FOR THE SECOND-ORDER HYPERBOLIC EQUATION Advanced Mathematcal Models & Applcatons Vol.3, No.3, 2018, pp.215-222 ON A DETERMINATION OF THE INITIAL FUNCTIONS FROM THE OBSERVED VALUES OF THE BOUNDARY FUNCTIONS FOR THE SECOND-ORDER HYPERBOLIC EUATION

More information

An (almost) unbiased estimator for the S-Gini index

An (almost) unbiased estimator for the S-Gini index An (almost unbased estmator for the S-Gn ndex Thomas Demuynck February 25, 2009 Abstract Ths note provdes an unbased estmator for the absolute S-Gn and an almost unbased estmator for the relatve S-Gn for

More information

An Admission Control Algorithm in Cloud Computing Systems

An Admission Control Algorithm in Cloud Computing Systems An Admsson Control Algorthm n Cloud Computng Systems Authors: Frank Yeong-Sung Ln Department of Informaton Management Natonal Tawan Unversty Tape, Tawan, R.O.C. ysln@m.ntu.edu.tw Yngje Lan Management Scence

More information

Lecture 12: Discrete Laplacian

Lecture 12: Discrete Laplacian Lecture 12: Dscrete Laplacan Scrbe: Tanye Lu Our goal s to come up wth a dscrete verson of Laplacan operator for trangulated surfaces, so that we can use t n practce to solve related problems We are mostly

More information

DETERMINATION OF TEMPERATURE DISTRIBUTION FOR ANNULAR FINS WITH TEMPERATURE DEPENDENT THERMAL CONDUCTIVITY BY HPM

DETERMINATION OF TEMPERATURE DISTRIBUTION FOR ANNULAR FINS WITH TEMPERATURE DEPENDENT THERMAL CONDUCTIVITY BY HPM Ganj, Z. Z., et al.: Determnaton of Temperature Dstrbuton for S111 DETERMINATION OF TEMPERATURE DISTRIBUTION FOR ANNULAR FINS WITH TEMPERATURE DEPENDENT THERMAL CONDUCTIVITY BY HPM by Davood Domr GANJI

More information

Uncertainty and auto-correlation in. Measurement

Uncertainty and auto-correlation in. Measurement Uncertanty and auto-correlaton n arxv:1707.03276v2 [physcs.data-an] 30 Dec 2017 Measurement Markus Schebl Federal Offce of Metrology and Surveyng (BEV), 1160 Venna, Austra E-mal: markus.schebl@bev.gv.at

More information

U.C. Berkeley CS294: Beyond Worst-Case Analysis Luca Trevisan September 5, 2017

U.C. Berkeley CS294: Beyond Worst-Case Analysis Luca Trevisan September 5, 2017 U.C. Berkeley CS94: Beyond Worst-Case Analyss Handout 4s Luca Trevsan September 5, 07 Summary of Lecture 4 In whch we ntroduce semdefnte programmng and apply t to Max Cut. Semdefnte Programmng Recall that

More information

Error Probability for M Signals

Error Probability for M Signals Chapter 3 rror Probablty for M Sgnals In ths chapter we dscuss the error probablty n decdng whch of M sgnals was transmtted over an arbtrary channel. We assume the sgnals are represented by a set of orthonormal

More information

Limit Theorems for Markovian Bandwidth Sharing Networks with Rate Constraints

Limit Theorems for Markovian Bandwidth Sharing Networks with Rate Constraints Lmt Theorems for Markovan Bandwdth Sharng Networks wth Rate Constrants Josh Reed Stern School of Busness New York Unversty Bert Zwart CWI September 9, 214 Abstract Bandwdth sharng networks as consdered

More information

Workshop: Approximating energies and wave functions Quantum aspects of physical chemistry

Workshop: Approximating energies and wave functions Quantum aspects of physical chemistry Workshop: Approxmatng energes and wave functons Quantum aspects of physcal chemstry http://quantum.bu.edu/pltl/6/6.pdf Last updated Thursday, November 7, 25 7:9:5-5: Copyrght 25 Dan Dll (dan@bu.edu) Department

More information

SOJOURN TIME IN A QUEUE WITH CLUSTERED PERIODIC ARRIVALS

SOJOURN TIME IN A QUEUE WITH CLUSTERED PERIODIC ARRIVALS Journal of the Operatons Research Socety of Japan 2003, Vol. 46, No. 2, 220-241 2003 he Operatons Research Socety of Japan SOJOURN IME IN A QUEUE WIH CLUSERED PERIODIC ARRIVALS Da Inoue he oko Marne and

More information

Online Appendix. t=1 (p t w)q t. Then the first order condition shows that

Online Appendix. t=1 (p t w)q t. Then the first order condition shows that Artcle forthcomng to ; manuscrpt no (Please, provde the manuscrpt number!) 1 Onlne Appendx Appendx E: Proofs Proof of Proposton 1 Frst we derve the equlbrum when the manufacturer does not vertcally ntegrate

More information

An Interactive Optimisation Tool for Allocation Problems

An Interactive Optimisation Tool for Allocation Problems An Interactve Optmsaton ool for Allocaton Problems Fredr Bonäs, Joam Westerlund and apo Westerlund Process Desgn Laboratory, Faculty of echnology, Åbo Aadem Unversty, uru 20500, Fnland hs paper presents

More information

THE CHINESE REMAINDER THEOREM. We should thank the Chinese for their wonderful remainder theorem. Glenn Stevens

THE CHINESE REMAINDER THEOREM. We should thank the Chinese for their wonderful remainder theorem. Glenn Stevens THE CHINESE REMAINDER THEOREM KEITH CONRAD We should thank the Chnese for ther wonderful remander theorem. Glenn Stevens 1. Introducton The Chnese remander theorem says we can unquely solve any par of

More information

One-sided finite-difference approximations suitable for use with Richardson extrapolation

One-sided finite-difference approximations suitable for use with Richardson extrapolation Journal of Computatonal Physcs 219 (2006) 13 20 Short note One-sded fnte-dfference approxmatons sutable for use wth Rchardson extrapolaton Kumar Rahul, S.N. Bhattacharyya * Department of Mechancal Engneerng,

More information

Exponential Type Product Estimator for Finite Population Mean with Information on Auxiliary Attribute

Exponential Type Product Estimator for Finite Population Mean with Information on Auxiliary Attribute Avalable at http://pvamu.edu/aam Appl. Appl. Math. ISSN: 193-9466 Vol. 10, Issue 1 (June 015), pp. 106-113 Applcatons and Appled Mathematcs: An Internatonal Journal (AAM) Exponental Tpe Product Estmator

More information

Maximizing the number of nonnegative subsets

Maximizing the number of nonnegative subsets Maxmzng the number of nonnegatve subsets Noga Alon Hao Huang December 1, 213 Abstract Gven a set of n real numbers, f the sum of elements of every subset of sze larger than k s negatve, what s the maxmum

More information