Size: px
Start display at page:

Download ""

Transcription

1 Bandwdth Packng E. G. Coman, Jr. and A. L. Stolyar Bell Labs, Lucent Technologes Murray Hll, NJ Abstract We model a server that allocates varyng amounts of bandwdth to \customers" durng servce. Customers could be computer jobs wth demands for storage bandwdth or they could be calls wth demands for transmsson bandwdth on a network lnk. Servce tmes are constants, each normalzed to 1 tme unt, and the system operates n dscrete tme, wth packng (schedulng) decsons made only at nteger tmes. Demands for bandwdths are for fractons of the total avalable and are lmted to the dscrete set f1=k; 2=k;:::; 1g where k s a gven parameter. More than one customer can be served at a tme, but the total bandwdth allocated to the customers n servce must be at most the total avalable. Customers arrve nk ows and jon a queue. The jth ow has rate j and contans just those customers wth bandwdth demands j=k. We study the performance of the two packng algorthms Frst Ft and Best Ft, both allocatng bandwdth by a greedy rule, the rst scannng the queue n arrval order and the second scannng the queue n decreasng order of bandwdth demand. We determne necessary and sucent condtons for stablty of the system under the two packng rules. The average total bandwdth demand of the arrvals n a tme slot must be less than 1 for stablty under any packng rule,.e., the condton := (=k) < 1 must hold. We prove that f the arrval rates 1 ;:::; are symmetrc,.e., = k, for all ; 1 k, 1, then < 1 s also sucent for stablty under both rules. Our Best Ft result strengthens an earler result conned to Posson ows and equal rates 1 = = ; and does so usng a far smper proof. Our Frst Ft result s completely new. The work here extends earler results on bandwdth packng n multmeda communcaton systems, on storage allocaton n computer systems, and on message transmsson along slotted communcaton channels. It s not surprsng that < 1 s sucent under Best Ft, snce n a congested system, Best Ft tends to serve two complementary (matched) customers n each tme slot, wth bandwdth demands beng =k and (k, )=k for some ; 1 k, 1. It s not so obvous, however, that <1 s also sucent under Frst Ft. Interestngly, when the system becomes congested, Frst Ft exhbts a "self-organzng" property whereby an orderng of the queue by tme of arrval becomes approxmately the same as an orderng by decreasng bandwdth demand. Part of the work of ths author was completed whle he was wth AT&T Labs{Research, Murray Hll, NJ (now located n Florham Park, NJ 07932). 1

2 1 Introducton We study a queueng model of storage and transmsson bandwdth allocaton n computer and communcaton systems. To dene the model, we use the termnology of queueng systems; later, we wll map ths termnology nto that of the applcatons. In our queueng model, customers are allocated avalable bandwdth accordng to ther demands, each customer holdng ts allocaton whle t s beng served, then releasng ts allocaton when t departs from the system. More than one customer can be served at a tme, but the total bandwdth allocated to customers n servce at any tme must be less than the total avalable. Bandwdth demands are dscretzed and speced n fractons; for some gven nteger k>0, a demand can be any multple of 1=k up to k=k =1. The dscretzaton loses no generalty n practce and expands the potental applcatons, as we shall see below. Customers arrve n k ows to a sngle queue, the th ow havng rate and just those customers wth bandwdth demands =k. Each customer servce tme s a constant, whch we take to be the unt of tme. In addton, the system operates n dscrete tme; packng decsons are made, and customer servces begn, only at nteger tmes. Unt tme ntervals begnnng at nteger tmes wll also be called tme slots. Note that our model s a stochastc verson of one dmensonal bn packng [5], where a bn corresponds to the total bandwdth avalable over a tme slot. We study the performance of both the Frst Ft and Best Ft packng rules. At servce completons, both rules scan the queue and pack customer bandwdth demands by a greedy rule: the demand beng consdered s packed f and only f t s for at most the bandwdth stll remanng from the demands already packed for customers to be served n the next tme slot. The derence between the two polces s that Frst-Ft scans the queue n arrval order, whle Best Ft scans the queue n decreasng order of bandwdth demand. Our analyss addresses stablty problems: determne necessary and sucent condtons on the arrval rates such that the system s stable under Frst Ft and Best Ft,.e., the underlyng Markov queueng processes are ergodc. We prove that f the bandwdth-demand rates 1 ;:::; are symmetrc,.e., = k, ; 1 k, 1, then under a very general class of arrval processes, := k (=k) < 1 s necessary and sucent for stablty for both Frst Ft and Best Ft. Snce s the average total bandwdth demand n each tme unt, < 1 s clearly a necessary condton, so our proofs focus entrely on showng that the condton s sucent. In what follows we wll take k =0. It would be trval for us to generalze our results to the case k > 0, but we have chosen not to do so as t creates a lack of symmetry and clutters the analyss. Models smlar to ours were studed n [3], where applcatons to multmeda communcatons were emphaszed. The term `bandwdth packng' was ntroduced n [3] as a name for the class of problems of nterest here. In the applcatons, bandwdth on a network lnk s beng 2

3 allocated to several competng demands n varyng amounts such as those needed for vdeo, audo, and data transmsson. It was proved n [3] that <1was sucent for the stabltyof Best Ft when the nput ows were specalzed to the Posson process and when the demand dstrbuton was unform wth all 's equal. The stablty queston for Frst Ft was left as an open problem. The analyss n [3] appled the classcal potental (Lyapunov) functon approach. Our approach uses the relatvely recent ud-lmt technques descrbed n Secton 2. As a consequence, our proofs are more compact and more easly adapted to general arrval processes. In another mportant applcaton, the avalable bandwdth refers to the storage (memory) bandwdth of a multprocessor system, a model studed n [9]. Customers are jobs usng varyng amounts of storage whle runnng on a computer. The model n [9] ders from ours n requrng a strct FIFO servce dscplne. Ths s a substantal smplcaton of our model, but the analyss n [9] leads to further results, ncludng formulas nvarant measures. Our work solves the stablty problem n ths applcaton for much more ecent packng rules. Our work also contrbutes new results to the analyss of an equvalent model of slotted communcaton systems [6]. In ths new nterpretaton, customers are messages and bandwdth demands are message duratons (fractons of a tme slot); the avalable bandwdth n a tme unt of our model becomes a unt-duraton slot n whch subsets of messages are packed and transmtted. Wth arrvng messages modeled by a dscretzed Markov process, the analyss n [6] focuses on the Next Ft algorthm: When a message arrves and nds no messages watng, t s packed (wll be sent) n the next tme slot. If a message arrves and nds other watng messages, t s packed n the latest tme slot already allocated at least one message, f t ts n the remanng unallocated tme of that slot; otherwse, the message s packed n the next, as yet unused, tme slot (and hence eventually transmtted one tme unt later). Our analyss extends the earler work to the much more ecent Frst Ft and Best Ft packng algorthms. For example, wth equal arrval rates 1 = =, the message-rate capacty under Next Ft s only 3=2, whereas t s 2 under Frst Ft and Best Ft. As a nal applcaton, one that takes us away from the bandwdth nterpretaton, we menton classcal k-server queues. Our model generalzes these queues by allowng customers to requre more than one server durng ther servce. In the termnology of our model, a bandwdth demand of =k s smply a request for servers. The next secton formalzes our probablty model and ntroduces the ud-lmt approach to our stablty problems. Our man results appear n Sectons 3 and 4 as theorems gvng necessary and sucent condtons for stablty under the Frst Ft and Best Ft rules, respectvely. In Secton 5 we present a moment convergence result, whch complements the stablty results for both Frst Ft and Best Ft. Whle our man results are n the stochastc analyss of algorthms, our methods also yeld useful results n the asymptotc average-case analyss of algorthms. In the average-case (or xed-nput) model, a xed number n of customers wth..d. bandwdth demands s gven, and the objectve s the large-n behavor of the expected total bandwdth wasted whle servng the n demands. Secton 6 apples the ud-lmt approach to the average-case 3

4 model by gvng a smple proof that, under Frst Ft and symmetrc bandwdth-demand dstrbutons, the expected total wasted bandwdth s o(n), so the expected number of tme slots needed to serve the n customers exceeds the expected sum of bandwdth demands (n=2) by a o(n) term. Secton 7 concludes the paper wth a dscusson of open problems and the senstvty of the analyss to varous model assumptons. 2 Prelmnares Under Frst Ft, a state of the queue s denoted by an element of the set of all ntely termnated sequences on f1;:::;k, 1g. The length of the sequence s the queue length, and the th element of the sequence gves the bandwdth demand of the customer that was th to arrve among the customers currently watng. Under Best Ft, the arrval order s not needed; the state just needs, for each =1;:::;k, 1 the number of customers watng wth bandwdth demands =k. Hereafter, a type- demand s one for a fracton =k of the bandwdth; type- customers are those wth type- demands. We assume that the aggregate arrval process of the k, 1 customer types can be descrbed by a nte number of ndependent, dscrete-tme regeneratve processes wth nte-mean regeneraton cycles. Our proofs rely on two consequences of ths assumpton: The underlyng queueng process, whch we denote by = ((t); t = 1; 2;:::), s a countable Markov chan, and the functonal strong law of large numbers holds for the nput process. To avod trval complcatons, we also assume that s rreducble and aperodc. These assumptons allow for vrtually any process havng a regeneratve structure, e.g., dscretetme versons of Markov modulated Posson processes, the processes generated by on-o sources, etc. However, to avod complcated notaton, n the rest of the paper we vew the k, 1 nput ows as ndependent wth the th beng an..d. sequence of nteger-valued random varables whch gve the numbers of type- arrvals n [t, 1;t] and have a nte mean, the same for all t =1; 2;:::. Wth ths smplcaton, the underlyng process becomes the queue-content process. In what follows, the norm k(t)k denotes the number of customers watng at tme t. Let (n) denote a process wth an ntal condton such that k (n) (0)k = n. In the analyss to follow, all varables assocated wth a process (n) wll be suppled wth the upper ndex (n). The followng theorem s a corollary of a more general result of Malyshev and Menshkov [10]. Theorem 1 Suppose there exsts an nteger T > 0 such that for any sequence of processes (n), we have lm n!1 E[ 1 n k (n) (nt )k] =0 (1) 4

5 Then s ergodc. It was shown by Rybko and Stolyar [11] that an ergodcty condton of the form (1) naturally leads to a ud-lmt approach to the stablty problem of queueng systems. Ths approach was further developed by Da [7], Chen [2], Stolyar [12], and Da and Meyn [8]. As the form of (1) suggests, the approach studes a ud process x(t) obtaned as a lmt of the sequence of scaled processes 1 n (n) (nt);t 0; at the heart of the approach n ts standard form s a proof that x(t) startng from any ntal state wth norm kx(0)k = 1 reaches 0 n nte tme T and stays there. (In most cases of nterest, ncludng ours, weaker condtons are sucent, e.g., t s enough to verfy that nf t0 kx(t)k < 1, as shown n [12].) In our settng we need to dene what the scalng 1 n (n) (nt) means. In order for ths scalng to make sense, we wll need an alternatve denton of the queueng process. To ths end, we rst adopt the conventon (t) = (btc); t 0, whch allows us to vew as a contnuous-tme process dened for all t 0, but wth new arrvals and servces stll begnnng only at nteger tmes t = 0; 1; 2;:::. Next, we dene the followng random functons assocated wth the process (n) (t): F (n) (t) s the total numberoftype- customers (n) that arrved by tme t 0, ncludng the customers present at tme 0; and ^F (t) s the (n) number of type- customers that were served by tme t 0. Obvously, ^F (0) = 0 for all. As n [11] and [12], we \encode" the ntal state of the system; n partcular, we extend the denton of F (n) (t) to the negatve nterval t 2 [,n; 0) by assumng that the customers present n the system n ts ntal state (n) (0) arrved n the past at tme nstants,(n, 1);,(n, 2);:::; 0, exactly one customer at each tme nstant. In the case of Frst Ft, we requre the order of these arrvals to be the same as n the state at tme 0. By ths conventon F (n) (,n) = 0 for all and n, and P F (n) (0) = n. It s clear that the process (n) = ( (n) (t);t 0) s a projecton of the process S (n) = (F (n) ; ^F (n) ), where and F (n) =(F (n) (t); t,n; =1; 2;:::;k, 1) ^F (n) =(^F (n) (t); t 0; =1; 2;:::;k, 1);.e., a sample path of S (n) unquely denes a sample path of (n). Now consder the scaled process s (n) =(f (n) ; ^f (n) ), where f (n) =(f (n) (t) = 1 n F (n) (nt); t,1; =1; 2;:::;k, 1) and ^f (n) =(^f (n) (t) = 1 (n) ^F (nt); t 0; =1; 2;:::;k, 1) n The followng lemma establshes convergence to a ud process and s a varant of Theorem 4.1 n [7]. 5

6 Lemma 1 The followng statements hold wth probablty 1. For any sequence of processes (n), there exsts a subsequence (m) ; fmg fng, such that for each ; 1 k, 1, (f (m) (t);t,1)! (f (t);t,1) u:o:c: (2) (m) ( ^f (t);t 0)! ( ^f (t);t 0) u:o:c: (3) where the functons f ; ^f ; are non-negatve non-decreasng Lpschtz-contnuous n the gven tme ntervals, and u.o.c. means that convergence s unform on compact sets as n! 1. The lmtng set of functons also satses and for all ; 1 k, 1, for any 0 t 1 t 2, s =(f; ^f) =f(f (t);t,1); ( ^f (t);t 0);; 2;:::;k, 1g f (0) = 1 (4) f (,1) = 0; ^f (0)=0; (5) f (t), f (0) = t; t 0; (6) ^f (t) f (t); t 0 ; (7) k ( ^f (t 2 ), ^f (t 1 )) 1 : (8) Proof. It follows from the strong law of large numbers that, wth probablty 1 for every, (f (n) (t), f (n) (0); t 0)! ( t; t 0) u:o:c: Also, for every n and, the functons (f (n) (n) (t);,1 t 0) and ( ^f (t);t 0) can ncrease by at most k(1=n) n any nterval of length 1=n. Ths mples (2) and (3). We get (6) as a byproduct. Equatons (4) and (5) follow from the constructon representng the ntal state. Equaton (7) follows mmedately from dentons, and the conservaton law (8) from the trval observaton that the total bandwdth of customers served n one tme slot s lmted to 1. 3 Frst Ft To prove that <1 s sucent for stablty under Frst Ft, we need two lemmas. Lemma 2 For any xed T 1 > 1, the followng holds wth probablty 1. A lmtng set of functons s = (f; ^f) dened n Lemma 1 has the followng addtonal property: For all ; 1 k, 1, ^f (T 1 ) >f (0): (9) 6

7 Proof. Consder the sequence of sample paths of the scaled process s (n) convergng to the set of functons s dened n Lemma 1. For any xed > 0 and > 0, we have for all sucently large n, f (n) () < 1+( )(1 + ): Snce as long as the queue s non-empty at least one customer s served n every tme slot, we conclude that ^f (n) ( +1+( )(1 + )) f (n) (); whch mples (9) va a smple passage to the lmt n!1, and by the fact that and can be arbtrarly small. For a set s of functons as dened n Lemma 1, let us dene and let (t) = nff,1j f () > ^f (t)g (10) (t) = mn (t): (11) 1 The proof of the Frst Ft stablty result centers on the analyss of the tmes (t); because of ther useful propertes, one beng a smple relaton to the ud lmt of the queue-length processes k (n) (t)k. For ths reason, we gve below an nformal nterpretaton of these tmes dened n terms of the unscaled processes S (n) = f(f (n) ; ^F (n) )g. Accordng to (10), (t) s the earlest tme by whch the number of type- arrvals exceeds the number of type- departures by tme t. Under sutable condtons to be covered n the lemma below, (t) can be expressed as the nverse of f evaluated at ^f (t),.e., (t) =f,1 ( ^f (t)); (12) as llustrated n Fgure 1. We remark that (t) need not be a smooth functon. For example, the ntal state can be contrved so that f (t) s at n some subnterval of [,1; 0], thus creatng a dscontnutyn (t). On the other hand, as proved later, (t) has to be Lpschtzcontnuous n the nterval 1 < t < 1. Under Frst Ft, the queue of type- customers at tme t conssts of just those type- customers that arrved durng [ (t);t]. Then (t) =t s a type- empty-queue condton. Recallng our dscusson of Theorem 1, we want to show that (t) tends towards t,.e., 0 (t) > 1; and absorbs n the empty-queue condton (t) =t. These and related propertes of the (t) are formalzed n the followng result. Lemma 3 Let T 1 > 1 be xed. There exst xed constants T 2 and T, T 1 T 2 T < 1 such that wth probablty 1, a lmtng set of functons s = (f; ^f) dened n Lemma 1 has the followng addtonal propertes: () We have (T 1 ) > 0 for all ; 1 k, 1: (13) 7

8 f (t) 1 ^f (t),1 (t) t Fgure 1: The functons (t), ^f (t), and f (t) wth f (t) = t + f (0) for t 0. () In the nterval t T 1, every functon (t); 1 k, 1, s non-decreasng Lpschtzcontnuous, and therefore so s (t). () At any regular pont t T 1,.e., a pont where all the dervatves of each of the functons f ; ^f ; ; and exst for all, 1 k, 1, we have 0 (t) = ^f 0 (t)= (14) (t) <t ) 0 (t) 1= (15) ( (t) < j (t) ^ <j) ) 0 j(t) =0: (16) (v) For all t T 2 ; <j ) (t) j (t): (17) (v) If the nput ows are symmetrc,.e., f = k, for every, then we have at any regular pont t T 2, (t) <t ) 0 (t) 1= > 1 (18) (v) For symmetrc nput ows, for all t T; whch s equvalent to the asserton that, for all t T; (t) =t; (19) ^f (t) =f (t) for all ; 1 k, 1: (20) 8

9 Proof Property () follows from Lemmas 1 and 2. When t T 1 > 1, the eects of the ntal state have dsspated and we know that s postve (by property ()), f (t) = f (0) + t, and ^f (t) s nondecreasng Lpschtz-contnuous (by Lemma 1). It follows easly that (t) s nondecreasng Lpschtz-contnuous for t T 1, whch proves property (). For property (), we rst derentate (12) at regular ponts t and substtute f 0 (t) = ; ths gves (14). To prove (15), dene M(t) :=f j (t) =(t)g; so that, snce t s a regular pont, we can wrte for all 2 M(t), (t) =(t); 0 (t) = 0 (t); ^f 0 (t)= = 0 (t): (21) For the unscaled sample path, t s easy to see that for any sucently small > 0 and all sucently large n, at least one customer of a type 2 M(t) wll be served n each tme slot of the tme nterval [nt; n(t + )]. Ths means that P ^f 0 2M (t) (t) 1; whch together wth (21) mples 0 (t) 1= P, thus provng (15). To prove (16) and complete the proof of property (), consder an unscaled sample path at tme nt. If (t) < j (t), then for small > 0 and all sucently large n, there are at least [ j (t), (t)](1, )n type- customers n the queue ahead of any type-j customer. Ths means that, f <j, there exsts a small >0 such that no type-j customers wll be served n the nterval [nt; n(t + )]. Ths n turn mples that enough, and therefore that (16) holds. ^f (n) j (t + ) = ^f (n) j (t) for all n large Property (v) follows from (15) and (16) n property (). The constant T 2 can be chosen to be T 2 = T 1 + max (T 1 ), (T 1 ) 1= P : To prove property (v), we note rst that, by property (v), the set M(t) for t T 2 has the form M(t) =fk, 1;k, 2;:::;rg wth r k, 1. Then we can rewrte (21) as (t) = (t) =:::= r (t) < r,1 (t) (22) 0 (t) =(t) 0 =:::= r(t) 0 > 0 (23) ^f (t)= 0 = = ^f r(t)= 0 r = 0 (t); (24) for some r k, 1. Here, we need to show that, f (t) <t, then 0 (t) 1= (25) Let us assume that k s odd; the proof of (25) for k even s very smlar and left to the reader. Frst, we make an observaton smlar to the one we made n the proof of (16). Consder the unscaled sample path at tme nt. Equaton (22) mples that, for any small > 0 and 9

10 all sucently large n, there are at least ( r,1 (t), (t))(1, )n customers of each type = k, 1;k, 2;:::;r n the queue ahead of any customer of type j = r, 1;:::;1. Ths means that there exsts a small > 0 such that n the nterval [nt; n(t + )] the customers of types r, 1;:::;1 have lower prorty than customers of types k, 1;k, 2;:::;r. More precsely, no customer of type j < r wll be packed n a tme slot as long as a customer of type r can be packed nto that tme slot nstead. Therefore, as far as the behavor of (n) the functons ^f ; r, n the nterval [t; t + ] s concerned, we can gnore the customers of types j <r. In the remander of the proof of property (v), we x wth the above observaton n mnd; we conne ourselves to the behavor of scaled processes n the nterval [t; t + ], and the behavor of the correspondng unscaled processes n the nterval [nt; n(t + )], wth n sucently large. Let p =(k, 1)=2, and note that the symmetry condton = k, ; 1 p; mples " # p := k = k + k, k, k = p = =p+1 : (26) If r p +1,then 0 (t) = 1 P =r 1 P =p+1 = 1 : (27) To see ths, note that exactly one customer of some type r wll be served n each tme slot. For, snce 2r k + 1, two such customers have demands exceedng the total avalable bandwdth. Ths mmedately mples that =r ^f 0 (t) =1 whch means that 0 (t) P =r = 1, and hence that (27) holds. To nsh the proof of property (v), t remans to dspose of the case r p. We wll show that 0 (t) =1=. Frst, we observe that 0 (t) 1=. Ths s because, by an argument smlar to the one used for the case r p + 1 above, we have =p+1 ^f 0 (t) 1 and so 0 (t) 1 P =p+1 = 1 : Now assume that strct nequalty holds, 0 (t) < 1=: (28) 10

11 We wll prove that ths mples that ^f 0 r(t) 1=, whch s a contradcton, snce ^f 0 r(t) = 0 (t) by denton of r. If (28) were to hold, then for any >0, any sucently small ; 0 <<;(wth dependng on ), and all sucently large n (dependng on and ), the followng three observatons would hold for an unscaled sample path n the nterval [nt; n(t + )]: (a) The number of tme slots not servng any customers of types k, 1;:::;k, r + 1, whch do not t together wth type-r customers, s at least 2 4 1, ( 0 (t)+) =k,r n: (b) Consder a tme slot descrbed n (a). If ths slot does not serve any customers of types p; p, 1;:::;r+1,then t must serve at least one type-r customer. (c) The total number of served customers of types = p;:::;r+ 1 does not exceed [ 0 (t)+] p =r+1 1 A n: Snce the number of slots occuped by customers of types p; p, 1;:::;r+1s at most the number of such customers, observatons (a)-(c) mply that the lmtng type-r servce rate has the lower bound ^f 0 r(t) 2 4 1, ( 0 (t)+) Snce >0 can be arbtrarly small, we get ^f 0 r(t) 1, 0 (t) =k,r =k,r+1 3 5, [ 0 (t)+] + By the symmetry condton, P =k,r+1 = P r,1,so p ^f 0 r(t) 1, 0 (t) p =r (t) r = 1, 0 (t) + 0 (t) r > 0 (t) r ; 3 5 : p =r+1 the desred contradcton. Thus, (28) can not hold, we can conclude that 0 (t) = 1=, and property (v) s proved. It follows from property (v) that nfftj (t) =tg T 2 + T 2, (T 2 ) 1=, 1 T 2 + T 2 1=, 1 11

12 Let us choose the constant T to be T = T 2 + T 2 1=, 1 : Snce we know that d (t, (t)) 1, 1= < 0 at any regular pont t T dt 2 such that t, (t) > 0, we conclude that (t) = t for all t T. Ths proves property (v) and hence the lemma. Theorem 2 Suppose the nput ow ntenstes are symmetrc, and <1. Then under Frst Ft s ergodc. Proof The proof s a slght modcaton of the proof of Theorem 4.2 n [7]. In partcular, Lemmas 1 and 3 mply that there exsts a T > 0, whch can be chosen to be an nteger, such that for any sequence of processes f (n) g we have 1 lm n!1 n k (n) (nt )k = n!1 lm (f (n) (T ), ^f (n) (T )) = 0; (29) wth probablty 1. The unform ntegrablty of the sequence f (n) g can be proved n ways smlar to those n [11] and [7]. Unform ntegrablty and the convergence n (29) mply that lm E[ 1 n!1 n k (n) (nt )k] =0: Then the condton n (1) of Theorem 1 holds, and we are done. We can also make strong statements about convergence propertes and the exstence of moments. These apply to Best Ft as well, so we defer these results to Secton 5. 4 Best Ft Dscplne In ths secton, we prove that the analog of Theorem 2 for Best Ft also holds. We use the same general ud-lmt approach, but the arguments wll be smpler. We agan need Lemma 1, but we wll create a new verson of Lemma 3, a smpler verson n that there wll be no need to deal wth the tmes (t) or the encodng of the ntal state; nstead of analyzng t, (t), we wll analyze the derence q (t) :=f (t), ^f (t), provng that t reaches zero n nte tme and stays there. Theorem 3 Suppose the nput ow ntenstes are symmetrc, and <1. Then under Best Ft s ergodc. 12

13 Proof: We need only prove Lemma 4 below; wth Lemma 4 replacng Lemmas 2 and 3, the proof of Theorem 2 wll apply to Best Ft. Lemma 4 Suppose that the nput ows are symmetrc. Then there exst constants 0=T k < T <:::<T 1 = T < 1 such that, wth probablty 1, a lmtng set of functons s dened n Lemma 1 has the followng addtonal property for every =1; 2;:::;: At any regular pont t T +1, ^f (t) <f (t) ) ^f 0 (t) +(1, ); (30) and for any t T, Thus, for all t T, and for all ; 1 k, 1, ^f (t) =f (t) and therefore ^f 0 (t) = : (31) ^f (t) =f (t) (32) Proof All the conventons ntroduced n the proof of Lemma 3 are stll n force. Thus, s (n) ;n =1; 2;::: s the sequence of sample paths of the scaled process s (n) whch converges to s. And when we refer to the unscaled sample path, we mean the correspondng sample path of the process S (n). We consder only the case when k s odd; the proof for even k s analogous.) Dene p := (k, 1)=2, as before, and recall that q (t) := f (t), ^f (t), so that (30) and (31) become q (t) > 0 ) q(t) 0,(1, ) (33) and q (t) = 0 for every t T : (34) We need a couple of key observatons, the rst followng from the fact that, the hgher the bandwdth demand, the hgher the packng prorty under Best Ft. (a) For any xed r, the servce of customers of types ;k,2;:::;rs completely unaected by the servce of customers of types j <r. (b) The condton q (t) > 0 for the lmtng set of functons mples that, for a sucently small, xed > 0, and all n sucently large, the correspondng unscaled sample path s such that n the nterval [tn; (t + )n]: (b 1 ) there are always type- customers avalable for servce; (b 2 ) f p, then every tme slot servng a type-(k, ) customer must serve a type- customer; every tme slot not servng a type- customer, or a type-(k, ) or larger customer must serve one or more customers of types p; p, 1;:::;+1. The proof s by nducton on decreasng from k, 1to1.If = k, 1, t follows easly from observatons (a) and (b 1 ) that at any regular pont t 0 = T k, the condton q (t) > 0 mples that ^f 0 (t) =1 +(1, ), and hence that (33) holds. 13

14 Notce that q (0) 1, so f we choose T = T k +1=(1, ); (35) then (34) follows from (33). Ths establshes the bass of the nducton. For the nducton step, suppose (33) and (34) hold for = k, 1;k, 2;:::;r+1. We wll now prove that (33) and (34) also hold for = r. Consder a regular pont t T r+1. If r p + 1, then the condton q r (t) > 0must mply ^f 0 r(t) =1, =r+1 > r +(1, ); (36) whch gves q 0 r,(1, ). To see ths note that, n an unscaled sample path, one and only one customer of types k, 1;k, 2;:::;r can be served n a tme slot. Thus, (36) follows from observatons (a) and (b 1 ) and the nductve hypothess, whch asserts that the customers of each of the types = k, 1;:::;r+1are served at exactly the correspondng rate for all t>t. (We omt routne ; -techncaltes smlar to those used n the proof of Lemma 3.) If r p, then by applyng observatons (a) and (b 2 ), we get =1, so q 0 r,(1, ) agan holds. r,1 ^f 0 r(t) 1,, p =r+1 =k,r+1, p =r+1, r + r = r +(1, ) Now f we observe that q r (T r+1 ) 1+ r T r+1, and set, n analogy wth (35), T r =(1+ r T r+1 )=(1, ); then we see that (34) follows from (33). The nductve step and hence the proof of Lemma 4 and Theorem 3 s complete. 5 Moment Convergence It s shown n [8] that condton (1) mples not only stablty, but also very strong momentexstence and convergence propertes. For example, Theorem 4 below follows drectly from (1) and Theorem 6.2 n [8] (whch can easly be adjusted for our dscrete-tme case). 14

15 Theorem 4 Suppose the are symmetrc, < 1 holds, and the nput processes are..d. sequences wth nte (p+1)-st moments (p 1 s an nteger). Let (1) have the statonary dstrbuton of the Markov chan under ether Frst Ft or Best Ft. Then and for any ntal state (0), Ek(1)k p < 1 lm t!1 Ek(t)kp = Ek(1)k p 6 Connecton to Average-Case Analyss Our Frst Ft stablty analyss s closely related to the average-case analyss of Frst Ft bn packng under dscrete dstrbutons [4]. We can recast the average-case model nto our settng as follows. Suppose the ntal state conssts of a queue of n customers wth customer types beng a sequence of ndependent samples from a gven dstrbuton on f1;:::;k, 1g; and assume there are no new arrvals. A formula for the expected total wasted bandwdth n the packng of n customers s the objectve of the average-case analyss. The problem s dcult, so vrtually all of the results to-date descrbe large-n asymptotc behavor. In partcular, t has been shown that for certan customer-type dstrbutons, the expected total wasted bandwdth s o(n). Ths secton demonstrates how the Frst Ft average-case result can easly be obtaned from the propertes of the ud lmt of a stochastc system lke the one consdered n prevous sectons. Further results of ths type are dscussed n the next secton. Consder an nnte sequence of..d. customer types 1 ; 2 ;:::, takng values n the set f1;:::;g, wth the dstrbuton f ;;:::;g, P =1. The reason for adoptng our arrval rate notaton for the customer-type dstrbuton n an average-case model wll be clear when the theorem below lnks up the average-case and stochastc analyss. Dene a sequence of systems (just lke those analyzed n prevous sectons) ndexed by n = 1; 2;:::. The n th system has an ntal state consstng of customers of types 1 ;:::; n watng n queue n the order lsted. Suppose there are no new arrvals after tme 0. Note that n ths settng the ntal state s random. For the n th system, dene U (n) to be the tme slot n whch the last customer of the ntal state s served under Frst Ft, and dene W (n) = U (n), n =k ; W (n) s the total bandwdth (or server capacty) wasted by the Frst Ft packng process. Theorem 5 Under Frst Ft and a symmetrc dstrbuton f g = ; 1 k, 1; 15

16 the followng holds wth probablty 1: lm W (n) =n =0; (37) n!1 lm U (n) =n =1=2 : (38) n!1 Remark. Snce the random varables W (n) =n and U (n) =n are bounded above unformly n n, the probablty 1 convergence mples convergence of mean values. Proof Frst, t s clear that the propertes (37) and (38) are equvalent, snce 1 lm n!1 n n k = 1 2 ; wth probablty 1.Also, W (n) 0 obvously holds, and therefore lm nf n!1 U (n) =n 1=2; so t wll suce to show that lm sup n!1 U (n) =n 1=2 : (39) For every ndex n, consder a moded system n whch new arrvals after tme 0 do occur; the nput (say Posson) ow oftype arrvals has ntensty. By the denton of Frst Ft, such a modcaton can not change the random varables W (n) and U (n), because t has no eect on the servce of ntal customers. The prmary reason for consderng the moded system s to comply wth the formulatons of the results n prevous sectons, to ease the `reuse' of those results. We observe that our sequence of processes, wth ndex n, satses the condtons of Lemmas 1 and 3, except for the fact that the ntal state s now random. But snce the ntal states are drawn from a sequence of..d. random varables, the functonal strong law of large numbers apples to the sequence of ntal states. We can conclude that: Except for property (13), Lemma 1 and Lemma 3 are vald for our sequence of moded processes. Moreover, they are vald wth T 1 = T 2 =0and a lmtng set of functons (a ud process) s such that f (t) = (t, (,1));, 1 t 0; 8 (40) Indeed, t follows that f (t) = (t, (,1)); t,1; 8 (41) and hence that (0) =,1 for any. The only property of the constant T 1 requred n earler proofs was that each functon f () be strctly lnear wth slope n the nterval [ (T 1 ); 1); wth T 1 = 0, we stll have ths property. The only property of the constant T 2 requred n 16

17 the earler proofs was that T 2 T 1 and (T 2 ) ::: 1 (T 2 ). Agan, wth T 2 =0,we stll have ths property. Applyng the results of Lemma 1 and Lemma 3, we get 0 (t) 1= = 2 at any regular pont t 0. In fact, from the conservaton law (8), we see that equaltymust hold: 0 (t) =1= =2, at any regular pont t 0. Then the followng must also hold: (t) = (t) =:::= 1 (t) =,1+2t; t 0; (42) because an nequalty (t) < (t) for any xed t and would contradct property (8). We are now n poston to prove (39). Let us x a small >0. It follows from (42) that any lmtng set of functons s s such that, for all, ^f ((1, )=2) = (1, ) < = f (0): Ths means that, wth probablty 1 for any, the sequence of scaled processes that for all n, except perhaps for values n some nte set, (n) ^f () ssuch (1, 2) Ths n turn means that, n the unscaled systems: (n) ^f ((1, )=2) <f (n) (0) < (1 + ) : (a) n the rst b(n=2)(1, )c tme slots, no server bandwdth was wasted and only ntal customers were served; (b) U (n) (n=2)(1, )+nk 3. Therefore, wth probablty 1, lm sup n!1 U (n) =n (1, )=2+3k : Snce >0 can be chosen arbtrarly small, we get (39), whch concludes the proof. 7 Dscusson Our proofs of Theorems 2and3verfy that, for arrval processes wthn the broad framework gven n Secton 2, the stablty of the system under consderaton depends essentally on the nput ow ntenstes and s nsenstve to the precse probablstc structure of the nput ows. Other specal cases of nterest to whch the result of Theorem 2 s easly extended, are sets of dvsble bandwdth demands. If h=k and j=k; j >hare any two demands n a dvsble subset of f1=k; 2=k;:::; 1g, then h dvdes j and j n turn dvdes k. The specal case f1=2 a ; 1=2 a,1 ;:::;1=2; 1g for some postve nteger a s of nterest n computer applcatons. We leave to the nterested reader an easy adaptaton of the ud approach to a proof that <1 s sucent for stablty under Frst Ft and Best Ft ndependent of the relatve szes of the arrval rates. 17

18 In heavy congeston, Frst Ft typcally matches demands =k wth ther complements (k, )=k, thus wastng no bandwdth and allowng < 1 to be sucent as well as necessary for stablty. For ths matchng to occur, the queue must reorder tself dynamcally nto decreasng order by sze. In an attempt to nd other examples where Frst Ft has a smlar self-organzng property, consder any dstrbuton that yelds perfect packngs respectng the arrval rates,.e., examples for whch there are ntegers n; n 1 ;:::;n such that, for some ; 0 < < 1, we have = n =n; 1 k, 1; and we can pack n 1 type-1 customers, n 2 type-2 customers,...,and n type-(k, 1) customers nto n tme unts wth no avalable bandwdth left over. It s easy to dene an algorthm that sutably restrcts the demand-type conguratons allowed n each tme unt so as to guarantee that < 1 s sucent for stablty. (What makes ths algorthmc technque mpractcal, of course, s that t requres advance knowledge of the arrval rates.) Typcally, however, t s not possble to make the same stablty clam about Frst Ft. As a smple example, take k = 7 and let the only nonzero arrval rates be 2 =2 3. A greedy algorthm that, whenever possble, packs 2 type-2 customers and 1 type-3 customer n a tme slot wll be stable as long as < 1. However, t can be shown that, durng perods of congeston under Frst Ft, a postve fracton of the tme slots wll be packed wth 2 type-3 customers or 3 type-2 customers, wastng 1/7 of the bandwdth n each case. A formal proof that Frst Ft s unstable for values of n an nterval [1, ; 1]; >0ssketched n the appendx. Examples lke those above suggest that the class of bandwdth-demand dstrbutons for whch < 1 s sucent for stablty under Frst Ft or Best Ft s lkely to be relatvely small. More dentve statements of ths knd present nterestng drectons for further research. We establshed n Secton 6 that the large-n packng process n the average-case model and the queueng process under heavy congeston n the stochastc model are descrbed by essentally the same ud process. Ths connecton makes the followng conjecture qute plausble. Consder an average-case model wth a xed customer-type dstrbuton fb g; and a famly of stochastc models wth ths same customer-type dstrbuton; then for each model n the famly = b ; where the total arrval rate = P ndexes the famly of stochastc models. We conjecture that, for the Frst Ft algorthm, the expected total wasted bandwdth s o(n) n the average-case model f and only f <1 s sucent for stablty n the famly of stochastc models. Expected total wasted bandwdth s known to be O( p nk) for Frst Ft packng [4] and customer types ndependently and unformly dstrbuted on f1;:::;k, 1g. Ths result and our Theorem 5 on the more general symmetrc dstrbutons lends support for the conjecture. Further support s provded by recent results of Albers and Mtzenmacher [1] who showed that, n the average-case model, the expected wasted bandwdth under Frst Ft s O(1) when b 1 = = b k,2 =1=(k, 2). The ud approach shows easly that Frst Ft s stable n the correspondng stochastc model wth ntenstes 1 = = k,2 and = 0, and wth <1. In fact, the proof of Theorem 2 s easly generalzed to prove the same result for any set of ntenstes satsfyng: () 1 s arbtrary, () for some gven nteger m; 2 m<k=2, we have the symmetry = k, for all ; m k, m, and () all other ntenstes are 18

19 0. Acknowledgments We gratefully acknowledge Anja Feldmann and Nabl Kahale for havng helped to ntate the research culmnatng n ths paper. Appendx Proposton 1 Consder Frst Ft bandwdth packng wth k =7and Posson nput ows of only types 2 and 3 customers, wth ntenstes satsfyng 2 =2 3 > 0. There exsts a >0 such that the system wth >1, s unstable. Proof sketch We consder rst a smpler, `saturated' verson of the system n whch new arrvals are generated f and only f there s room for more customers n the current tme slot. In other words, when packng the current tme slot, f the entre queue has been scanned and 2/7 of the tme slot s stll empty, the queue s mmedately extended by new arrvals, wth the numbers of new customers of types 2 and 3 beng Posson dstrbuted wth means 2 and 3. These arrvals are mmedately avalable for further packng. The process s repeated as necessary untl the slot s at most 1/7 empty. The queue state just after each nteger tme s a dscrete-tme countable Markov chan. Its ergodcty s easly vered. It s also easy to see that the average per-slot rates 2 and 3 at whch customers of types 2 and 3 are served are such that 3 = 2 =2 < 1. Also, the Markov chan can be vewed as a regeneratve process wth an empty queue beng a regeneraton state. All moments of the regeneraton cycle are nte. Let us now return to our orgnal system wth 2 and 3 such that 3 < 3 = 2 =2 < 1, whch means 3 <<1. Ths system s unstable. To prove ths, we consder an ntal state formed by arrvals wthout servce (packng s suspended) for M tme slots, wth M large. When the packng starts, the packng process s ndstngushable from the packng process n the saturated system, untl the tme slot when the last ntal customer s reached (.e., scanned for the rst tme). It wll take approxmately M tme slots, wth = 3 = 3 > 1; for the packng process to reach the last ntal customer. By that tme, the queue s longer than the ntal queue; t s extended by new arrvals durng approxmately M tme slots. The packng process wll then take approxmately 2 M slots to reach the end of that queue. Ths contnues, wth the maxmum queue length growng wthout bound. Usng routne large-devaton estmates, t s a smple matter to convert the above observatons nto a rgorous argument that, wth postve probablty, the queue length tends to nnty (see, for example, the nstablty example n [11]). References [1] S. Albers and M. Mtzenmacher. Average-Case Analyss of Frst Ft and Random Ft Bn Packng. Proc. Nnth Ann. ACM-SIAM Symp. Dsc. Alg., (1998), pp

20 [2] H. Chen. Flud Approxmatons and Stablty of Multclass Queueng Networks: Workconservng Dscplnes. Annals of Appled Probablty, Vol. 5, (1995), pp [3] E. G. Coman, Jr., A. Feldmann, N. Kahale, B. Poonen. Computng Call Admsson Capactes n Lnear Networks. Bell Labs, Lucent Technologes, Murray Hll, NJ (1997), submtted for publcaton. [4] E. G. Coman, Jr., D. S. Johnson, P. W. Shor, and R. R. Weber. Bn packng wth dscrete tem szes, Part II: Tght bounds on Frst Ft. Random Structures and Algorthms, 10:69{ 101, [5] E. G. Coman, Jr., M. R. Garey, and D. S. Johnson. Approxmaton Algorthms for Bn Packng: A Survey. n Approxmaton Algorthms for NP-Complete Problems, D. S. Hochbaum (ed.). PWS Publshng Co. (1995), pp [6] E. G. Coman, Jr., S. Haln, A. Jean-Mare, and Ph. Robert. Stochastc Analyss of a Slotted, FIFO Communcaton Channel. IEEE Trans. Inf. Th., 39(1993), pp. 1555{1566. [7] J. G. Da. On the Postve Harrs Recurrence for Open Multclass Queueng Networks: A Uned Approach Va Flud Lmt Models. Annals of Appled Probablty, Vol. 5, (1995), pp [8] J. G. Da and S. P. Meyn. Stablty and Convergence of Moments for Open Multclass Queueng Networks Va Flud Lmt Models. IEEE Transactons on Automatc Control, Vol. 40, (1995), pp [9] C. Kpns and Ph. Robert. A Dynamc Storage Process. Stoch. Proc. and Ther Applc., 34(1990), pp. 155{169. [10] V.A. Malyshev and M.V. Menshkov. Ergodcty, Contnuty, and Analytcty of Countable Markov Chans. Transactons of Moscow Mathematcal Socety, Vol. 39, (1979), pp [11] A.N. Rybko and A.L. Stolyar. Ergodcty of Stochastc Processes Descrbng the Operaton of Open Queueng Networks Problems of Informaton Transmsson, Vol. 28, (1992), pp [12] A.L. Stolyar. On the Stablty of Multclass Queueng Networks: A Relaxed Sucent Condton va Lmtng Flud Processes Markov Processes and Related Felds, 1(4), 1995, pp

Recurrence. 1 Definitions and main statements

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

More information

Luby s Alg. for Maximal Independent Sets using Pairwise Independence

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

More information

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

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

More information

1 Example 1: Axis-aligned rectangles

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

More information

What is Candidate Sampling

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

More information

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

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

More information

Product-Form Stationary Distributions for Deficiency Zero Chemical Reaction Networks

Product-Form Stationary Distributions for Deficiency Zero Chemical Reaction Networks Bulletn of Mathematcal Bology (21 DOI 1.17/s11538-1-9517-4 ORIGINAL ARTICLE Product-Form Statonary Dstrbutons for Defcency Zero Chemcal Reacton Networks Davd F. Anderson, Gheorghe Cracun, Thomas G. Kurtz

More information

BERNSTEIN POLYNOMIALS

BERNSTEIN POLYNOMIALS On-Lne Geometrc Modelng Notes BERNSTEIN POLYNOMIALS Kenneth I. Joy Vsualzaton and Graphcs Research Group Department of Computer Scence Unversty of Calforna, Davs Overvew Polynomals are ncredbly useful

More information

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

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

More information

Project Networks With Mixed-Time Constraints

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

More information

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

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

More information

An Alternative Way to Measure Private Equity Performance

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

More information

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

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

More information

DEFINING %COMPLETE IN MICROSOFT PROJECT

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

More information

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

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

More information

8 Algorithm for Binary Searching in Trees

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

More information

CALL ADMISSION CONTROL IN WIRELESS MULTIMEDIA NETWORKS

CALL ADMISSION CONTROL IN WIRELESS MULTIMEDIA NETWORKS CALL ADMISSION CONTROL IN WIRELESS MULTIMEDIA NETWORKS Novella Bartoln 1, Imrch Chlamtac 2 1 Dpartmento d Informatca, Unverstà d Roma La Sapenza, Roma, Italy novella@ds.unroma1.t 2 Center for Advanced

More information

The OC Curve of Attribute Acceptance Plans

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

More information

Generalizing the degree sequence problem

Generalizing the degree sequence problem Mddlebury College March 2009 Arzona State Unversty Dscrete Mathematcs Semnar The degree sequence problem Problem: Gven an nteger sequence d = (d 1,...,d n ) determne f there exsts a graph G wth d as ts

More information

General Auction Mechanism for Search Advertising

General Auction Mechanism for Search Advertising General Aucton Mechansm for Search Advertsng Gagan Aggarwal S. Muthukrshnan Dávd Pál Martn Pál Keywords game theory, onlne auctons, stable matchngs ABSTRACT Internet search advertsng s often sold by an

More information

NPAR TESTS. One-Sample Chi-Square Test. Cell Specification. Observed Frequencies 1O i 6. Expected Frequencies 1EXP i 6

NPAR TESTS. One-Sample Chi-Square Test. Cell Specification. Observed Frequencies 1O i 6. Expected Frequencies 1EXP i 6 PAR TESTS If a WEIGHT varable s specfed, t s used to replcate a case as many tmes as ndcated by the weght value rounded to the nearest nteger. If the workspace requrements are exceeded and samplng has

More information

v a 1 b 1 i, a 2 b 2 i,..., a n b n i.

v a 1 b 1 i, a 2 b 2 i,..., a n b n i. SECTION 8.4 COMPLEX VECTOR SPACES AND INNER PRODUCTS 455 8.4 COMPLEX VECTOR SPACES AND INNER PRODUCTS All the vector spaces we have studed thus far n the text are real vector spaces snce the scalars are

More information

Value Driven Load Balancing

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

More information

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

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

More information

CHOLESTEROL REFERENCE METHOD LABORATORY NETWORK. Sample Stability Protocol

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

More information

Lecture 3: Force of Interest, Real Interest Rate, Annuity

Lecture 3: Force of Interest, Real Interest Rate, Annuity Lecture 3: Force of Interest, Real Interest Rate, Annuty Goals: Study contnuous compoundng and force of nterest Dscuss real nterest rate Learn annuty-mmedate, and ts present value Study annuty-due, and

More information

How To Solve A Problem In A Powerline (Powerline) With A Powerbook (Powerbook)

How To Solve A Problem In A Powerline (Powerline) With A Powerbook (Powerbook) MIT 8.996: Topc n TCS: Internet Research Problems Sprng 2002 Lecture 7 March 20, 2002 Lecturer: Bran Dean Global Load Balancng Scrbe: John Kogel, Ben Leong In today s lecture, we dscuss global load balancng

More information

Support Vector Machines

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

More information

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

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

More information

J. Parallel Distrib. Comput.

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

More information

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

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

More information

A Probabilistic Theory of Coherence

A Probabilistic Theory of Coherence A Probablstc Theory of Coherence BRANDEN FITELSON. The Coherence Measure C Let E be a set of n propostons E,..., E n. We seek a probablstc measure C(E) of the degree of coherence of E. Intutvely, we want

More information

Extending Probabilistic Dynamic Epistemic Logic

Extending Probabilistic Dynamic Epistemic Logic Extendng Probablstc Dynamc Epstemc Logc Joshua Sack May 29, 2008 Probablty Space Defnton A probablty space s a tuple (S, A, µ), where 1 S s a set called the sample space. 2 A P(S) s a σ-algebra: a set

More information

We are now ready to answer the question: What are the possible cardinalities for finite fields?

We are now ready to answer the question: What are the possible cardinalities for finite fields? Chapter 3 Fnte felds We have seen, n the prevous chapters, some examples of fnte felds. For example, the resdue class rng Z/pZ (when p s a prme) forms a feld wth p elements whch may be dentfed wth the

More information

Calculating the high frequency transmission line parameters of power cables

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

More information

Stochastic epidemic models revisited: Analysis of some continuous performance measures

Stochastic epidemic models revisited: Analysis of some continuous performance measures Stochastc epdemc models revsted: Analyss of some contnuous performance measures J.R. Artalejo Faculty of Mathematcs, Complutense Unversty of Madrd, 28040 Madrd, Span A. Economou Department of Mathematcs,

More information

Chapter 4 ECONOMIC DISPATCH AND UNIT COMMITMENT

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

More information

PSYCHOLOGICAL RESEARCH (PYC 304-C) Lecture 12

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

More information

Distributed Optimal Contention Window Control for Elastic Traffic in Wireless LANs

Distributed Optimal Contention Window Control for Elastic Traffic in Wireless LANs Dstrbuted Optmal Contenton Wndow Control for Elastc Traffc n Wreless LANs Yalng Yang, Jun Wang and Robn Kravets Unversty of Illnos at Urbana-Champagn { yyang8, junwang3, rhk@cs.uuc.edu} Abstract Ths paper

More information

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

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

More information

The Cox-Ross-Rubinstein Option Pricing Model

The Cox-Ross-Rubinstein Option Pricing Model Fnance 400 A. Penat - G. Pennacc Te Cox-Ross-Rubnsten Opton Prcng Model Te prevous notes sowed tat te absence o arbtrage restrcts te prce o an opton n terms o ts underlyng asset. However, te no-arbtrage

More information

Online Advertisement, Optimization and Stochastic Networks

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

More information

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

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

More information

The University of Texas at Austin. Austin, Texas 78712. December 1987. Abstract. programs in which operations of dierent processes mayoverlap.

The University of Texas at Austin. Austin, Texas 78712. December 1987. Abstract. programs in which operations of dierent processes mayoverlap. Atomc Semantcs of Nonatomc Programs James H. Anderson Mohamed G. Gouda Department of Computer Scences The Unversty of Texas at Austn Austn, Texas 78712 December 1987 Abstract We argue that t s possble,

More information

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

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

More information

Sngle Snk Buy at Bulk Problem and the Access Network

Sngle Snk Buy at Bulk Problem and the Access Network A Constant Factor Approxmaton for the Sngle Snk Edge Installaton Problem Sudpto Guha Adam Meyerson Kamesh Munagala Abstract We present the frst constant approxmaton to the sngle snk buy-at-bulk network

More information

Analysis of Energy-Conserving Access Protocols for Wireless Identification Networks

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

More information

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

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

More information

A Lyapunov Optimization Approach to Repeated Stochastic Games

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

More information

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

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

More information

n + d + q = 24 and.05n +.1d +.25q = 2 { n + d + q = 24 (3) n + 2d + 5q = 40 (2)

n + d + q = 24 and.05n +.1d +.25q = 2 { n + d + q = 24 (3) n + 2d + 5q = 40 (2) MATH 16T Exam 1 : Part I (In-Class) Solutons 1. (0 pts) A pggy bank contans 4 cons, all of whch are nckels (5 ), dmes (10 ) or quarters (5 ). The pggy bank also contans a con of each denomnaton. The total

More information

INVESTIGATION OF VEHICULAR USERS FAIRNESS IN CDMA-HDR NETWORKS

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

More information

Logistic Regression. Lecture 4: More classifiers and classes. Logistic regression. Adaboost. Optimization. Multiple class classification

Logistic Regression. Lecture 4: More classifiers and classes. Logistic regression. Adaboost. Optimization. Multiple class classification Lecture 4: More classfers and classes C4B Machne Learnng Hlary 20 A. Zsserman Logstc regresson Loss functons revsted Adaboost Loss functons revsted Optmzaton Multple class classfcaton Logstc Regresson

More information

REGULAR MULTILINEAR OPERATORS ON C(K) SPACES

REGULAR MULTILINEAR OPERATORS ON C(K) SPACES REGULAR MULTILINEAR OPERATORS ON C(K) SPACES FERNANDO BOMBAL AND IGNACIO VILLANUEVA Abstract. The purpose of ths paper s to characterze the class of regular contnuous multlnear operators on a product of

More information

Section 5.4 Annuities, Present Value, and Amortization

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

More information

The literature on many-server approximations provides significant simplifications toward the optimal capacity

The literature on many-server approximations provides significant simplifications toward the optimal capacity Publshed onlne ahead of prnt November 13, 2009 Copyrght: INFORMS holds copyrght to ths Artcles n Advance verson, whch s made avalable to nsttutonal subscrbers. The fle may not be posted on any other webste,

More information

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

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

More information

In some supply chains, materials are ordered periodically according to local information. This paper investigates

In some supply chains, materials are ordered periodically according to local information. This paper investigates MANUFACTURING & SRVIC OPRATIONS MANAGMNT Vol. 12, No. 3, Summer 2010, pp. 430 448 ssn 1523-4614 essn 1526-5498 10 1203 0430 nforms do 10.1287/msom.1090.0277 2010 INFORMS Improvng Supply Chan Performance:

More information

Cross-Selling in a Call Center with a Heterogeneous Customer Population

Cross-Selling in a Call Center with a Heterogeneous Customer Population OPERATIONS RESEARCH Vol. 57, No. 2, March Aprl 29, pp. 299 313 ssn 3-364X essn 1526-5463 9 572 299 nforms do 1.1287/opre.18.568 29 INFORMS Cross-Sellng n a Call Center wth a Heterogeneous Customer Populaton

More information

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

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

More information

Embedding lattices in the Kleene degrees

Embedding lattices in the Kleene degrees F U N D A M E N T A MATHEMATICAE 62 (999) Embeddng lattces n the Kleene degrees by Hsato M u r a k (Nagoya) Abstract. Under ZFC+CH, we prove that some lattces whose cardnaltes do not exceed ℵ can be embedded

More information

This paper concerns the evaluation and analysis of order

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

More information

Basic Queueing Theory M/M/* Queues. Introduction

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

More information

Whch one should I mtate? Karl H. Schlag Projektberech B Dscusson Paper No. B-365 March, 996 I wsh to thank Avner Shaked for helpful comments. Fnancal support from the Deutsche Forschungsgemenschaft, Sonderforschungsberech

More information

On the Interaction between Load Balancing and Speed Scaling

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

More information

Multi-Resource Fair Allocation in Heterogeneous Cloud Computing Systems

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

More information

The Greedy Method. Introduction. 0/1 Knapsack Problem

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

More information

On the Interaction between Load Balancing and Speed Scaling

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

More information

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

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

More information

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

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

More information

PERRON FROBENIUS THEOREM

PERRON FROBENIUS THEOREM PERRON FROBENIUS THEOREM R. CLARK ROBINSON Defnton. A n n matrx M wth real entres m, s called a stochastc matrx provded () all the entres m satsfy 0 m, () each of the columns sum to one, m = for all, ()

More information

7.5. Present Value of an Annuity. Investigate

7.5. Present Value of an Annuity. Investigate 7.5 Present Value of an Annuty Owen and Anna are approachng retrement and are puttng ther fnances n order. They have worked hard and nvested ther earnngs so that they now have a large amount of money on

More information

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

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

More information

Calculation of Sampling Weights

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

More information

THE DISTRIBUTION OF LOAN PORTFOLIO VALUE * Oldrich Alfons Vasicek

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

More information

Complete Fairness in Secure Two-Party Computation

Complete Fairness in Secure Two-Party Computation Complete Farness n Secure Two-Party Computaton S. Dov Gordon Carmt Hazay Jonathan Katz Yehuda Lndell Abstract In the settng of secure two-party computaton, two mutually dstrustng partes wsh to compute

More information

APPLICATION OF PROBE DATA COLLECTED VIA INFRARED BEACONS TO TRAFFIC MANEGEMENT

APPLICATION OF PROBE DATA COLLECTED VIA INFRARED BEACONS TO TRAFFIC MANEGEMENT APPLICATION OF PROBE DATA COLLECTED VIA INFRARED BEACONS TO TRAFFIC MANEGEMENT Toshhko Oda (1), Kochro Iwaoka (2) (1), (2) Infrastructure Systems Busness Unt, Panasonc System Networks Co., Ltd. Saedo-cho

More information

Conferencing protocols and Petri net analysis

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

More information

How To Calculate The Accountng Perod Of Nequalty

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

More information

OPTIMAL INVESTMENT POLICIES FOR THE HORSE RACE MODEL. Thomas S. Ferguson and C. Zachary Gilstein UCLA and Bell Communications May 1985, revised 2004

OPTIMAL INVESTMENT POLICIES FOR THE HORSE RACE MODEL. Thomas S. Ferguson and C. Zachary Gilstein UCLA and Bell Communications May 1985, revised 2004 OPTIMAL INVESTMENT POLICIES FOR THE HORSE RACE MODEL Thomas S. Ferguson and C. Zachary Glsten UCLA and Bell Communcatons May 985, revsed 2004 Abstract. Optmal nvestment polces for maxmzng the expected

More information

Exhaustive Regression. An Exploration of Regression-Based Data Mining Techniques Using Super Computation

Exhaustive Regression. An Exploration of Regression-Based Data Mining Techniques Using Super Computation Exhaustve Regresson An Exploraton of Regresson-Based Data Mnng Technques Usng Super Computaton Antony Daves, Ph.D. Assocate Professor of Economcs Duquesne Unversty Pttsburgh, PA 58 Research Fellow The

More information

Simple Interest Loans (Section 5.1) :

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

More information

SDN: Systemic Risks due to Dynamic Load Balancing

SDN: Systemic Risks due to Dynamic Load Balancing SDN: Systemc Rsks due to Dynamc Load Balancng Vladmr Marbukh IRTF SDN Abstract SDN acltates dynamc load balancng Systemc benets o dynamc load balancng: - economc: hgher resource utlzaton, hgher revenue,..

More information

On Lockett pairs and Lockett conjecture for π-soluble Fitting classes

On Lockett pairs and Lockett conjecture for π-soluble Fitting classes On Lockett pars and Lockett conjecture for π-soluble Fttng classes Lujn Zhu Department of Mathematcs, Yangzhou Unversty, Yangzhou 225002, P.R. Chna E-mal: ljzhu@yzu.edu.cn Nanyng Yang School of Mathematcs

More information

Lecture 3: Annuity. Study annuities whose payments form a geometric progression or a arithmetic progression.

Lecture 3: Annuity. Study annuities whose payments form a geometric progression or a arithmetic progression. Lecture 3: Annuty Goals: Learn contnuous annuty and perpetuty. Study annutes whose payments form a geometrc progresson or a arthmetc progresson. Dscuss yeld rates. Introduce Amortzaton Suggested Textbook

More information

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

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

More information

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

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

More information

Forecasting the Direction and Strength of Stock Market Movement

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

More information

Optimal resource capacity management for stochastic networks

Optimal resource capacity management for stochastic networks Submtted for publcaton. Optmal resource capacty management for stochastc networks A.B. Deker H. Mlton Stewart School of ISyE, Georga Insttute of Technology, Atlanta, GA 30332, ton.deker@sye.gatech.edu

More information

Time Value of Money Module

Time Value of Money Module Tme Value of Money Module O BJECTIVES After readng ths Module, you wll be able to: Understand smple nterest and compound nterest. 2 Compute and use the future value of a sngle sum. 3 Compute and use the

More information

Dynamic Online-Advertising Auctions as Stochastic Scheduling

Dynamic Online-Advertising Auctions as Stochastic Scheduling Dynamc Onlne-Advertsng Auctons as Stochastc Schedulng Isha Menache and Asuman Ozdaglar Massachusetts Insttute of Technology {sha,asuman}@mt.edu R. Srkant Unversty of Illnos at Urbana-Champagn rsrkant@llnos.edu

More information

Cross-Selling in a Call Center with a Heterogeneous Customer Population

Cross-Selling in a Call Center with a Heterogeneous Customer Population OPERATIONS RESEARCH Vol. 57, No. 2, March Aprl 2009, pp. 299 313 ssn 0030-364X essn 1526-5463 09 5702 0299 nforms do 10.1287/opre.1080.0568 2009 INFORMS Cross-Sellng n a Call Center wth a Heterogeneous

More information

Traffic State Estimation in the Traffic Management Center of Berlin

Traffic State Estimation in the Traffic Management Center of Berlin Traffc State Estmaton n the Traffc Management Center of Berln Authors: Peter Vortsch, PTV AG, Stumpfstrasse, D-763 Karlsruhe, Germany phone ++49/72/965/35, emal peter.vortsch@ptv.de Peter Möhl, PTV AG,

More information

Dscrete-Tme Approxmatons of the Holmstrom-Mlgrom Brownan-Moton Model of Intertemporal Incentve Provson 1 Martn Hellwg Unversty of Mannhem Klaus M. Schmdt Unversty of Munch and CEPR Ths verson: May 5, 1998

More information

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

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

More information

Dynamics of heterogeneous peer-to-peer networks

Dynamics of heterogeneous peer-to-peer networks Dynamcs of heterogeneous peer-to-peer networks Fernando Pagann, Andrés Ferragut and Martín Zubeldía Unversdad ORT Uruguay Abstract Dynamc models of populaton n peer-to-peer fle sharng systems have focused

More information

1 OPTIMIZATION ISSUES IN WEB

1 OPTIMIZATION ISSUES IN WEB 1 OPTIMIZATIO ISSUES I WEB SEARCH EGIES Zhen Lu 1 and Phlppe an 2 1 IBM Research Hawthorne, Y 10532, USA zhenl@us.bm.com 2 IRIA B.P. 93, 06902, Sopha Antpols Cedex, France Phlppe.an@nra.fr Abstract: Crawlers

More information

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

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

More information

INSTITUT FÜR INFORMATIK

INSTITUT FÜR INFORMATIK INSTITUT FÜR INFORMATIK Schedulng jobs on unform processors revsted Klaus Jansen Chrstna Robene Bercht Nr. 1109 November 2011 ISSN 2192-6247 CHRISTIAN-ALBRECHTS-UNIVERSITÄT ZU KIEL Insttut für Informat

More information

Linear Circuits Analysis. Superposition, Thevenin /Norton Equivalent circuits

Linear Circuits Analysis. Superposition, Thevenin /Norton Equivalent circuits Lnear Crcuts Analyss. Superposton, Theenn /Norton Equalent crcuts So far we hae explored tmendependent (resste) elements that are also lnear. A tmendependent elements s one for whch we can plot an / cure.

More information