Distinct Random Sampling from a Distributed Stream

Size: px
Start display at page:

Download "Distinct Random Sampling from a Distributed Stream"

Transcription

1 Ditinct Random Sampling from a Ditributed Stream Yung-Yu Chung and Srikanta Tirthapura 2215 Coover Hall, Iowa State Univerity, Ame, IA, USA {ychung,nt}@iatate.edu Abtract We conider continuou maintenance of a random ample of ditinct element from a maive data tream, whoe input element are oberved at multiple ditributed ite that communicate via a central coordinator. At any point, when a query i received at the coordinator, it repond with a random ample from the et of all ditinct element oberved at the different ite o far. We preent the firt algorithm for ditinct random ampling from a ditributed tream. We alo preent a lower bound on the expected number of meage that mut be tranmitted by any ditributed algorithm, howing that our algorithm i meage optimal to within a factor of four. We preent extenion to liding window, and experimental reult howing the performance of our algorithm on real-world data et. Keyword-Ditinct Sampling, Random Sampling, Ditributed Stream Mining, Ditinct Element I. INTRODUCTION A random ample i a flexible and general ynopi of data, and can be ued to anwer aggregate querie on data with provable probabilitic guarantee on the quality of the anwer. A ditinct random ample (commonly known a a ditinct ample ) i choen from the et of all ditinct element in data. A ditinct ample ha the property that the probability of an element incluion in the ample i independent of the frequency of the element, a long a it occur within the data. Thi property i ueful in computing an important cla of aggregate that depend on the et of ditinct element in data. For intance, a ditinct ample can be ued in anwering the querie: what i the number of ditinct IP addree oberved within a tream of IP packet during a given hour, or how many ditinct viitor have ued a web ervice, who come from a particular country?, or what i the average age of the ditinct uer acceing thi webite? In contrat with a ditinct ample, a imple random ample i choen from the et of all occurrence within the data. The greater the frequency of an element in data, the more likely it incluion within a imple random ample. A imple random ample i not uited to etimate anwer to the above querie, while a ditinct ample i. A imple random ample favour the heavy-hitter within data, while a ditinct ample doe not. In cae where the ditribution of element i kewed, i.e. a few ditinct element account for a majority of the data element, a imple random ample will likely ignore element with low frequency and only contain element of high frequency (the heavy hitter ). In contrat, a ditinct random ample i equally likely to contain low frequency a well a high frequency element. Due to thi property, a ditinct ample of a data tream ha found application in databae query proceing [1], network monitoring [2], and in detecting anomalou network traffic [3]. Ditinct ampling i a fundamental primitive in current day big data ytem. Query optimizer in commercial relational databae ytem maintain a ditinct ample of individual databae column to help in chooing efficient query proceing trategie [4]. In tream databae ytem, ampling play a fundamental role a a data filtering method to reduce proceing load, and typically a variety of ampling method are implemented, including ditinct ampling [5]. Sytem uch a BlinkDB [6] ue ampling to proce querie quickly, in exchange for approximate accuracy guarantee, and have been ucceful in providing dramatic peedup for common aggregate querie. Thee ytem operate on ten of terabyte of data, pread over hundred of machine [6], and ditributed ampling i a key component of their query execution trategy. We conider the maintenance of a ditinct ample from a tream whoe element are oberved at multiple ditributed ite that communicate via a network. We preent a memory and communication-efficient olution uited to today tream proceing framework uch a IBM Infophere Stream [7], Apache Spark [8] and Apache Storm [9]. For the purpoe of decribing and analyzing the algorithm, we work in the continuou ditributed monitoring model [10] [13], where a coordinator node i required to continuouly maintain an aggregate function over the union of all tream oberved at the different ditributed ite. For ditributed proceing of large tream, the bottleneck i often the network, rather than proceor peed, and hence we focu on minimizing the meage complexity of the treaming algorithm. Sampling from a tream ha a long hitory of reearch, tarting from the popular reervoir ampling algorithm due to Waterman (ee Algorithm R from [14]) that ha been known ince the Thi include work on peeding up reervoir ampling [14], weighted reervoir ampling [15], ampling over a liding window and tream evolution [16] [19], and ditributed random ampling [10], [12]. While there i prior work on ditinct ampling from a ingle tream [1], [20] [22], thee do not extend to continuou ditributed monitoring.

2 A. Contribution We make the following contribution. Let k denote the number of ditributed ite, the deired ample ize, and d the total number of ditinct element in the entire ditributed tream. We preent a ditributed algorithm for continuouly maintaining a ditinct ample from a ditributed tream whoe expected meage complexity i O ( ) k ln d. When a query i poed at the central coordinator, it repond with a ditinct ample of ize min{, d} from the et of all element een o far in the tream. We preent a matching lower bound howing that any algorithm for continuouly maintaining a ditinct ample from the ditributed tream mut have a meage complexity of Ω ( ) k ln d. In other word, the meage complexity of our algorithm i optimal to within a contant factor. We preent an extenion to upport querie over a liding window of the ditributed data tream. We how thi algorithm i alo near-optimal in it meage complexity through demontrating an appropriate lower bound. We implemented our algorithm and evaluated it performance over real-world dataet. We found that our algorithm are eay to implement and provide a meage complexity that i typically much better than the wort cae bound that we derive. Comparion with Simple Random Sampling: It i intereting to compare the meage complexity of ditributed ditinct ampling (DDS) to that of ditributed imple random ampling (DRS). From previou work [10], [12], it i known that the meage complexity of DRS for a ample ize for n total element ditributed over k ite increae a approximately max{k, } log(n/) 1 In contrat, the meage cot of DDS increae a k log(d/), where k i the number of ite, d the total number of ditinct element, and the ample ize. The meage complexity of DDS i inherently larger than that of DRS if d (the number of ditinct element in the tream) i comparable to n (the total number of element in the tream). To undertand the reaon for thi difference, note that in DRS, when n element have already been received in the ytem, the probability of a new element being elected into the ample decreae a /n. In cae of DDS, the probability of a new element being elected into the ample decreae a /d, where d i the number of ditinct element received by the ytem. Due to the poibility of the ame element appearing at different ite at a imilar time, many ite may be fooled into ending meage to the coordinator after imultaneouly oberving the ame element, and hence more ( ) 1 k log(n/) The exact expreion i Θ if < k/8 and Θ( log(n/)) log(k/) if k/8. Thi meage complexity i tight up to a contant factor, due to matching lower bound. meage may be communicated between the ite and the coordinator in DDS. Our lower bound how that thi i not an artifact of an inferior algorithm for DDS. Intead, greater coordination i inherently neceary for DDS than for DRS. B. Prior and Related Work There i a growing literature on algorithm for continuou ditributed monitoring, tarting from the baic countdown problem [23], frequency moment [11], [23], entropy [24], heavy-hitter and quantile [25], geometric method for monitoring [26], [27], and often, matching lower bound [11]. For a recent urvey on reult in thi model, ee [13]. A urvey of ampling on tream appear in [28]. The cloet prior work i the work on algorithm and lower bound for ditributed random ampling [10], [12]. While the problem are different, our algorithm for DDS on an infinite window i eentially an adaptation of the DRS algorithm from [12] with a few important difference: one i that our algorithm ue ampling baed on a randomly choen hah function while the algorithm for DRS ample baed on independent random bit. Further, our algorithm employ additional data tructure to prevent duplicate meage from a ite to the coordinator due to oberving the ame element, thi i not neceary for imple random ampling. The analyi of the meage complexity i different for the DRS and the DDS algorithm. The lower bound argument for the two problem (DRS and DDS) are very different from each other. Our analyi of the lower bound on meage cot of any algorithm for DDS ue method, that to our knowledge, have not been applied to the continuou ditributed treaming model. A related model of ditributed tream wa conidered in [20], [29]. In thi model, the coordinator wa not required to continuouly maintain an aggregate, but intead, when the query wa poed to the coordinator, the ite would be contacted and the query reult would be contructed. In their model, the coordinator could be aid to be reactive, wherea in the model conidered in thi paper (the continuou ditributed treaming model), the coordinator i pro-active. Roadmap. We firt preent the model and problem definition in Section II, the algorithm and lower bound for infinite window in Section III, an overview of the algorithm for liding window in Section IV, and experimental reult in Section V. Due to pace contraint, we omit ome of the proof, and alo omit detail of the liding window algorithm and experimental reult. Thee detail, including all the proof, can be found in the full verion of the paper [30]. II. MODEL We conider a ytem with k ite, numbered from 1 to k. Each ite i monitor a local tream of element, which

3 are not all necearily ditinct. There i an integer time aociated with each obervation, and time i non-decreaing within a tream. At time t, let S i (t) denote the tream oberved by ite i o far, and let S(t) = k i=1 S i(t) denote the tream oberved by the ytem o far. Let D(t) denote the et of ditinct element in S(t), n(t) the number of element in S(t), and d(t) the number of ditinct element in S(t). There i a coordinator node, different from the other k ite, to whom all querie are poed. The different ite a well a the coordinator are aumed to be ynchronized in time, and we aume that the meage delay from the ite to the coordinator are all equal, a in a ynchronou ditributed ytem. Thee aumption, which have alo been ued in previou work in the continuou ditributed treaming model [12], [23], [31], allow u to focu on communication efficiency. We conider two verion of ditinct ampling, baed on the cope of the data that the query addree. In the infinite window cae, at every time intant t, the coordinator mut maintain a random ample of ize min{, d} from D(t). In the liding window cae [32], we are given a window ize w a a parameter, and the uer i intereted in all element that have arrived in the mot recent w time interval. In particular, let Si w (t) denote the tream that ha arrived at ite i at time t w + 1, t w + 2,..., t, and let S w (t) = k i=1 S i(t). Let D w (t) denote the et of ditinct element in in S w (t). The goal i for the coordinator to maintain at all time t, a random ample of ize min{, d} from the element in D w (t). Our meaure of performance i the total number of meage ent between the coordinator and the ite, and we aim to minimize thi number. Since each meage in our algorithm i of a mall ize, thi i alo a meaure of the number of byte tranmitted, in our cae. The ize 2 of local data tream S i, the order of arrival, the number of ditinct element in the local tream, and the interleaving of the tream at different ite, can all be arbitrary, and the algorithm cannot make any aumption about thee. III. INFINITE WINDOW We firt conider the cae of infinite window, when the ample ha to be choen from the et of all ditinct element een o far in the tream. Let h : U [0, 1] be a hah function that aign a real number in the range [0, 1] to each element in U. For different input, it i aumed that the output of h are mutually independent random variable. For et S, let h(s) denote {h(x) x S}. Note that for any time t, h(s(t)) = h(d(s(t))). The baic ampling trategy i a follow. The ditinct ample at time t i the et of element from S(t) that yield the mallet element in h(s(t)). It i clear that the above yield a ditinct ample from S(t). To ee thi, take any ubet T D(S(t)) of ize 2 The meage ize in our algorithm i contant, auming that each tream element can be tored in a contant number of byte. ; the probability that the element in T will yield the mallet value in h(d(s(t))) i exactly the probability that in a random permutation of h(d(s(t))), the element in T are ordered before the ret, which i 1/ ( ) D(t). Our ditributed algorithm for maintaining the above ample i a follow. Let u(t) denote the value of the th mallet element in h(s(t)). The coordinator alway ha the current value of u(t). Each ite i maintain a tate variable u i (t), which i it local view of u(t). u i (t) i initialized to 1 and i updated a follow. Whenever ite i oberve an item e uch that h(e) < u i (t), e and h(e) are ent to the coordinator, who update u(t). If h(e) indeed changed the value of u(t), then e i elected into the ample at the coordinator, replacing a current element in the ample (unle fewer than ditinct element have been een o far). In turn, the coordinator end a meage back to i to refreh the value of u i (t). The algorithm at ite i i preented in Algorithm 1 and at the coordinator i in Algorithm 2. Algorithm 1: Infinite Window: Algorithm at Site i Initialization: Receive hah function h from the coordinator; u i 1, P i ; repeat if receive element e from S i then if h(e) < u i and e P i then Inert e into P i ; Send e to the Coordinator; if receive element u from the coordinator then u i u; Dicard all element e in P i uch that h(e ) > u i ; until Forever; Lemma 1. When queried at time t, the coordinator return a random ample of ize min{, d} elected without replacement from D(t). Proof: Firt note that at time t, variable u i track u i (t) at ite i, and u track u(t) at the coordinator. For each ite i we claim u i u at all time. Thi can be proved uing induction. For the bae cae, note that initially, u i = u. There are two type of event to conider for the inductive tep: (1) when u u updated, and (2) when u i i updated. Each time u i updated, it only become maller (i.e. u i non-increaing), which maintain the invariant. Each time u i i updated, it i et to the current value of u, which alo maintain the invariant. Auming that hahe of different element are ditinct, we claim that the value of u equal the lth mallet hah value een o far, where l = min{, d}. We claim that every

4 Algorithm 2: Infinite Window: Algorithm at the Coordinator /* P i the random ample, and variable u ha the current value of u(t). */ Initialization: P, u 1; repeat if receive e from ite i then if h(e) < u then If e P, then inert e into P ; if P > then Dicard element e = argmax{h(e) e P } from P ; Update u max{h(e) e P }; Send u to ite i; if a query for a random ample arrive then Send P ; until Forever; time a new element e arrive at ite i uch that h(e) < u, e will be ent to the coordinator. To ee thi, conider an element e uch that h(e) < u, oberved at ite i. We have u u i, and h(e) < u; thee together imply that h(e) < u i. From the algorithm at the ite, it i clear that e will be ent to the coordinator. Thu, the coordinator can inductively maintain the l element from D(t) with the mallet hah value. It i clear thi contitute a random ample of ize l choen without replacement from D(t). A. Infinite Window: Analyi We preent an analyi of the meage and pace complexitie of our algorithm. In Section III-A1 we preent an upper bound on the meage complexity of our algorithm and in Section III-A2 we preent a lower bound. 1) Upper Bound: Conider the execution of the algorithm until the end of time tep t. Let d = D(t) be the total number of ditinct element that were oberved in the tream. Let d i denote the total number of ditinct element in S i (t). Clearly, d i d. Let Y i denote the total number of meage that are ent by node i (note that the number of meage ent by ite i equal the number of meage received). Let Y denote the total number of meage ent in the ytem. In the following, when the context i clear, we ue S i to mean S i (t), D i to mean D i (t), and o on. Lemma 2. An element e oberved at ite i multiple time will not lead to more than one meage tranmiion from i to the coordinator, and thi tranmiion can only happen the firt time ite i oberve e. Proof: Suppoe that e i oberved by ite i at time t 1, t 2,... where t 1 < t 2 <.... The firt time it i oberved at time t 1, uppoe that a meage wa not ent to the coordinator. In thi cae, h(e) u i (t 1 ). Since u i i nonincreaing, when e i oberved at a future time, ay t 2, it will alo be true that h(e) u i (t 2 ), and no further meage are ent to the coordinator. Next, uppoe that a meage wa ent to the coordinator when e wa oberved at time t 1. In thi cae e i inerted into P i at time t 1. When e i oberved again at time t 2 (or later), either e i till in P i, or it mut be true that u i (t 2 ) < h(e), ince otherwie, e would not have been dicarded from P i. Hence, e will not be ent to the coordinator in thi cae, according to Algorithm 1. For j = 1... d i, let e j i be the jth new ditinct element in the local tream S i. For j = 1... d i, let Y j i be a random variable equal to 1 if ite i communicated with the coordinator upon receiving e j i, and 0 otherwie. We have Y i = d i j=1 Y j i. [ ] Lemma 3. For ite i, and j = 1... d i, Pr Y j i = 1 j, if j >. Proof: Let Z j i denote the event that h(e j i ) i among the mallet element in {h(e q i ) q = 1... j}. Note that if Y j i = 1, then Z j i mut be true. Note that the convere i not necearily true; it i poible that Z j i i true, but Y j i = 0, for example, e j i may have already been oberved by a ite other than i, and it value may have been incorporated [ ] into u, and hence into u i. It i eay to ee that Pr Z j i = [ ] [ ] j for j >. Since Pr Y j i = 1 Pr Z j i, the lemma follow. Lemma 4. E [Y i ] + (H di H ) Proof: Uing linearity of expectation on Y i, we get: d i E [Y i ] = = j=1 j=1 + [ E Y j i ] d i = j=1 [ ] Pr Y j i = 1 d i j=+1 where we have ued Lemma 3. + [ ] Pr Y j i = 1 d i j=+1 [ ] Pr Y j i = 1 j = + (H d i H ) Lemma 5. Let Y denote the number of meage tranmitted by the ditributed algorithm during an execution when d ditinct element are oberved overall. E [Y ] 2k + 2k (H d H ) 2k ( 1 + ln ( )) d Proof: Note that for each meage ent by a ite, there i exactly one meage ent by the coordinator, and hence we have Y = 2 k i=1 Y i. The proof follow from Lemma 4 combined with the obervation d i d.

5 We remark that uing Lemma 4, it i poible to get a tighter upper bound for E [Y ] in cae when the number of ditinct element oberved at individual ite i much maller than the number of ditinct element overall. Obervation 1. E [Y ] 2k + 2 k (H di H ) 2k + 2 i=1 k ( ) di ln i=1 The following theorem ummarize the performance of the algorithm for infinite window. Theorem 1. Let d,, and k repectively denote the total number of ditinct element in the ditributed tream, ample ize, and the number of ite. There i an algorithm that continuouly maintain a ditinct ample of a ditributed tream, whoe expected total number of meage i no more than 2k ln ( ) de, memory conumption per ite i O(), memory conumption at the coordinator i O(), and proceing time per element i O(1). Proof: The meage complexity follow from Lemma 5. The memory at the coordinator i alo clear from an inpection of Algorithm 2. For the memory at ite i, note that a oon a an element e i inerted into P i, e i alo ent to the coordinator, and in turn the ite receive the current value of u, et u i u, and dicard all element e from P i uch that h(e) > u. Globally, there are no more than ditinct element e uch that h(e) u, hence even within ite i, there are no more than element in P i. 2) Lower Bound: Given any ditributed algorithm A that continuouly maintain a ditinct ample of the tream of ize, we contruct an input which caue A to end at leat a certain minimum number of meage, in expectation. Suppoe that the element were all choen from the et [m] = {1, 2,..., m} for m >> k. Note that the probability pace here i the one from which the random ample i choen; every algorithm need acce to random bit that define thi probability pace, uch that the ample choen i uniformly choen from the et of all ditinct element oberved o far. Lemma 6. Suppoe a et D of ditinct element of ize d ha already been oberved by the ytem o far, after ome round of computation. For any algorithm A, and any ite i = 1... k, there i an element e D i uch that upon receiving any element e [m] e D i D in the next round, ite i will end a meage to the coordinator with probability at leat 2(d+1). Proof: We ue proof by contradiction. Note that et D ha already been oberved by the ytem o far. Suppoe that there were two ditinct element e i, e i [m] D uch that upon e i, the probability that i ent a meage to the coordinator wa le than 2(d+1), and alo upon e i, the probability that i ent a meage to the coordinator wa le than 2(d+1). Conider the following two input in the next round. In one input I 1, ite i i given e i and the other ite do not receive any element. In the other input I 2, ite i i given e i and the other ite do not receive an element. In I1, at the end of thi round, there i a probability of d+1 that e i belong to the random ample. In I 2, at the end of thi round, there i a probability of d+1 that e i belong to the random ample. Thu, with probability at leat d+1, the ample at the coordinator after oberving I 1 i different from the ample after oberving I 2. However, in thi round, the coordinator oberve a change in execution between I 1 or I 2 only if either (1) ite i end a meage to the coordinator in I 1, or (2) ite i end a meage to the coordinator in I 2. The probability that at leat one of the above event happen i le than d+1, uing the union bound. Further, the behavior of the other ite ha an identical ditribution on both input, ince they did not receive any element. Thu, we have that the content of the random ample at the end of the round are different with probability at leat, but the probability that the meage oberved d+1 d+1 by the coordinator are different i le than. Thi lead to a non-zero probability that the random ample at the coordinator i incorrect at the end of thi round, and hence a contradiction. Hence, there can be at mot one element e i uch that upon receiving an element e [m] e i D, the probability of ending a meage to the coordinator i at leat 2(d+1), and the lemma follow. Lemma 7. Suppoe the et of ditinct element oberved o far i D, and let d = D, and d. For any algorithm A, there exit another round of input I(D) uch that after oberving I(D), the ite will end at leat an expected k 2(d+1) element to the coordinator. Proof: The input I(D) i contructed a follow. Let e be any element uch that e ( D ( k i=1 {ed i })). Element e i given to every ite in thi round. Uing Lemma 6, we get that each ite i = 1... k will end a meage to the coordinator with probability at leat follow. 2(d+1). The lemma Theorem 2. For any algorithm A, there exit an input ditributed tream, I A with d ditinct element uch that the expected number of meage ent by the algorithm upon receiving I A i at leat k 2 (H d H + 1) k 2 ln ( ) de. Proof: Input I A i contructed a follow. Let D 0 =. The input in the firt round i I(D 0 ). For i 0, let the et of all ditinct element oberved till (and including) round i be D i. Note that the ize of D i i exactly i. For i = 0... (d 1), the input in round i + 1 i I(D i ). From Lemma 7, in round i >, the expected number of k meage ent to the coordinator i at leat 2(i+1). Summing

6 thi over all round + 1,..., d, we get the number of k(h meage to be at leat d H ) 2. For round 1 till, imilar method yield that the expected number of meage ent in each round mut be at leat k 2, for the above input. Thu the expected total number of meage ent by thi algorithm i at leat k 2 (H d H + 1). B. Sampling With Replacement Thu far, we have conidered ditinct ampling without replacement. In ampling with replacement, the different ample are all choen independently and randomly from the et of ditinct element oberved o far, D(t). A poible olution to ditinct ampling with replacement i to repeat parallel copie of the ingle element ampling algorithm, each copy uing a different hah function. The correctne of thi cheme i trivial, and the meage cot i time the cot of a ingle element ampling algorithm, which i O(k log de). We can do better, uing the following obervation. When a node i receive an element e, it end only a ingle copy of e to the coordinator if it wa ampled by one or more of the parallel copie (of the ingle element ampling algorithm). We ketch the analyi of thi algorithm. For ite i and the jth ditinct element received by the ite, let Y j i be a random variable equal to 1 if the ite communicated with the coordinator upon receiving thi element, and 0 otherwie. The probability that the element i ampled ( by ) any of the. different copie of the algorithm i j Hence, ). E[Y j i (1 ] = 1 1 j If Yi denote the total number of meage ent by ite i during execution, we have uing linearity of expectation that E[Y i ] = ( ( ) ) d i j= j. Simplifying thi expreion (detail can be found in the full verion of the paper), we get E[Y i ] ln(d i e/). Thu the expected total number of meage in the ytem i bounded by O(k log(de/)), even for the algorithm for ampling with replacement. We alo note the following reduction from ditinct ampling without replacement to ditinct ampling with replacement. From a ditinct ample with replacement of ize lightly greater than, we elect a random ample without replacement. It i poible to prove that the reult i indeed a ditinct ample without replacement from the original dataet. Thu, the lower bound of Ω(k log( d )) applie to both ampling with and without replacement, and our method lead to an algorithm for ampling with replacement with optimal meage complexity. IV. SLIDING WINDOW In the liding window etting, a ample i deired over only a window of mot recent element of the tream. We aume that time i divided into lot that are numbered conecutively in an increaing equence. It i aumed that time i ynchronized acro the ite o that when ite 1 i oberving element in lot t, other ite are alo oberving element in the ame lot. Let w > 0 be an integer denoting the ize of the window. The problem i a follow: when a query i iued to the coordinator at lot t, it return a random ample of ize choen without replacement from the et of all ditinct element oberved in lot (t w + 1) till t, both endpoint incluive. We ue the term lot and time interchangeably in the following dicuion. Algorithm Idea: At time t, let S i (t, w) denote the element that have arrived in lot (t w +1) till t at ite i. Let S(t, w) denote the (multiet) union of S i (t, w), i = 1... k. Let D i (t, w) denote the et of ditinct element in S i (t, w), and D(t, w) the et of ditinct element in S(t, w). For a et of element E, let h(e) = {h(e) e E}. The highlevel algorithm idea i to chooe the element with the mallet hah value from D(t, w). Let u(t, w) denote the -th mallet hah value from h(d(t, w)). Implementation of the Idea: The challenge in implementing the liding window cenario, when compared with the infinite window cenario, i that u(t, w) i not monotonically decreaing. A element expire from the window, the value of u(t, w) may increae, and an algorithm ha to keep track of thi at both the coordinator and the ite. One poible method i a follow. Each ite i, at time t, keep track of the et of element with the mallet hah value, choen from S i (t, w). Note that thi alo form a ditinct ample from S i (t, w). Whenever thi ample change, the coordinator i informed. Since the coordinator ha all element with the mallet hah value from each ite, it can maintain element with the globally mallet hah value from D(t, w), and hence a ditinct ample from S(t, w). The meage complexity of thi algorithm depend on how often the local ample at the individual ite change. We can reduce the communication cot of the above algorithm by incorporating feedback from the coordinator to the ite a follow. Similar to infinite window, the coordinator maintain a variable u, which ha the value of u(t, w) at all time. When the coordinator replie back to a ite, it convey the current value of u. However, note that u may increae at the coordinator, due to element expiring from the window, and thi need alo to be conveyed to the ite (note that thi wa not neceary for the infinite window cae). One way to achieve thi i that each time u increae, the coordinator broadcat the new value of u to all node. While thi algorithm i correct, a broadcat baed method can be expenive, ince it communicate with node which may not be actively receiving meage. We intead ue an alternate approach, where the coordinator replie back with the value of u to a ite only when the ite communicate with the coordinator. Each ite i alo need to maintain a local liding window ample, coniting of the mallet hah value within D i (t, w). It i known that in general, maintaining the kth mallet element within a liding window require pace

7 linear in the window ize in the wort cae [32]. However, in our cae, we do not need to maintain the minima over arbitrary number, but need to do o over a equence of random number. Hence, we can ue the idea from priority ampling imilar to [17] to ignificantly reduce the pace conumption at each ite. For element e, e and time t, t, we ay that tuple (e, t) dominate tuple (e, t ) at ite i if t > t and h(e) < h(e ). Let τ i (e) denote the mot recent time when e wa oberved at ite i. For element e, e, we ay e dominate e at ite i if (e, τ i (e)) dominate (e, τ i (e )). Each ite i ha a data tructure T i coniting of all element that could potentially be included within the random ample of ditinct element either now, or in the future. T i can be maintained efficiently uing a Treap [33]. We omit detailed decription of the algorithm due to pace contraint, and preent the main reult on the performance of the algorithm. Theorem 3. There i a ditributed algorithm that continuouly maintain a ditinct random ample of ize over a time-baed liding window of ize w, with k ite and T total time tep. Suppoe M = min i,t { D i (t, w) }. The expected number of meage tranmiion between the ite and the coordinator i bounded by 2kT/M. We alo prove the following lower bound on the meage complexity of any algorithm for continuou ditinct ampling over a liding window. In particular, we how that the linear dependence of the meage complexity on T i neceary. Lemma 8. Suppoe there i only one ite and a coordinator. There exit an input tream S uch that for any ditributed algorithm A that can continuouly maintain a ditinct random ample at the coordinator over T tep, the expected number of meage ent to the coordinator i at leat Ω ( T w + log ) w. Lemma 9. The expected pace complexity of the liding window algorithm at ite i at time t i O( log D i (t, w) ) word. V. EXPERIMENTS We preent an experimental evaluation of our propoed algorithm. We ued two dataet. The firt i an OC48 Internet Trace Dataet [34], which ha anonymou traffic trace taken at a US wet coat OC48 peering link for a large ISP in 2002 and To generate an element, we conider the concatenation of the ender IP addre and the receiver IP addre. The other i the Enron Dataet [35], where an element i contructed by concatenating the ender addre and the receiver addre. A ummary of the data i hown in Table I. Each data point preented i the average of 20 independent run. We implemented the algorithm in Java, uing the MurmurHah [36] hah function. Due to pace contraint, we only preent the # Element # Ditinct OC48 42,268,510 4,337,768 Enron 1,557, ,330 Table I THE NUMBER OF ELEMENTS AND DISTINCT ELEMENTS IN OC48 IP AND ENRON DATASETS (a) OC48 IP Dataet Figure 1. The number of meage a a function of number of element under two different method of data ditribution, flooding and random, for 10 ite and a ample ize of 5. experimental reult for the OC48 dataet. The reult for the Enron dataet can be found in the full verion of the paper [30]. Our theoretical analyi wa for the wort cae, when the input data can be ditributed in an arbitrary manner by the adverary. Here we preent experimental reult for the infinite window cae, including the impact of data ditribution, of ample ize, and a comparion with an alternate, natural algorithm that i baed on a broadcat from the coordinator to the ite. Due to pace contraint, we do not preent the reult of the liding window experiment here. They can be found in the full verion. Impact of Data Ditribution: We examine the performance of our algorithm under different ditribution method. In the firt method, called flooding, each incoming element i aigned to every ite. In the econd method, called random, an incoming element i ent to a ingle ite, choen uniformly at random. In the third method, roundrobin, each element i ent to a ingle ite, and the element are aigned to ite in a round-robin manner, i.e. the j-th element i monitored by ite (j mod k)+1. The reult for random and round-robin turned out to be nearly identical, o we only preent the reult for random ditribution. Figure 1 how the number of meage a the function of the number of element for 10 ite and a ample ize of 5. From Figure 1, we oberve that at the beginning, the number of meage increae quickly, ince the ample i changing often. A more element are oberved, the rate of change of the ample decreae and fewer meage are ent. It i clear

8 Figure ite. (a) OC48 IP Dataet The number of meage a a function of the ample ize for (a) OC48 IP Dataet Figure 4. The number of meage ent by Algorithm Broadcat and our propoed method for 10 ite and a ample ize of 5. (a) OC48 IP Dataet Figure 3. The number of meage a function of the number of ite k for a ample ize of 20. that the number of meage under flooding i ignificantly larger than the number of meage under random, though the total number of ditinct element een in both input i the ame. Thi ( cenario i explained by our tighter upper bound 2k 1 + ) k di i=1 ln (ee Obervation 1), which how the influence of the number of ditinct element oberved at the individual ite on the total meage ent. Impact of Sample Size: Figure 2 how the number of meage a a function of the ample ize for 50 ite; the meage complexity increae almot linearly with the ample ize, though the lope are different for different method of data ditribution. Figure 3 how the number of meage a a function of the number of ite k for a ample ize of 20. For flooding, the number of meage increae linearly with the number of ite. However, for random ditribution, the number of meage i much maller than in cae of flooding, and i almot independent of the number of ite. Comparion with Other Algorithm: To our knowledge, there are no prior publihed method for ditinct ampling on a ditributed tream. We compare the performance of our algorithm with another natural algorithm, which we call Algorithm Broadcat. The difference between Algorithm Broadcat and our propoed method i that Algorithm Broadcat will broadcat the current value of u (which ha the value of u(t) at time t) to all ite whenever there i an update to u. Thi verion ha the advantage that fewer meage are ent from the ite to the coordinator, ince the u i are alway in ync with the coordinator. However, thi ha the downide of requiring a broadcat each time u change. In Figure 4, we preent a comparion between the two algorithm for 10 ite and a ample ize of 5. It i clear that Broadcat require ignificantly more meage than our algorithm; thi ugget that typically it i not worth keeping the different ite ynchronized with repect to the value of u. A lazy approach of refrehing u when neceary reult in fewer meage. We alo note that the meage cot of Algorithm Broadcat i linear in the number of ite k, and the ample ize. However, the lope of the Broadcat algorithm i coniderably higher. We how the comparion of the two algorithm a a function of the ample ize in Figure 6. Similar reult are oberved with the dependence on the number of ite. Impact of Skew: We next conider the influence of the non-uniformity in the ize of the tream oberved at different ite. Here, we contruct a ditributed input which i kewed toward a ingle ite that dominate over the other ite in term of the number of ditinct element that it oberve. Each input element i ent to only one ite; but intead of dealing it randomly or in a round-robin fahion, we end the element to ite 1 with a probability that i a factor α time the probability that a ite other than 1 i choen. We call thi factor a the kew. For example, if the kew i 200, then ite 1 i 200 time more likely to receive an element than another ite. Figure 7 how the relation

9 between number of meage tranmiion and the kew for different algorithm with 20 ite and a ample ize of 20. The number of meage tranmitted reduce a the kew increae; thi can be anticipated through Obervation 1 that preent an analyi in term of the number of ditinct element oberved at each ite. Note that the higher the kew, the cloer thi get to centralized tream monitoring. (a) OC48 IP Dataet Figure 5. The number of meage ent by Algorithm Broadcat and our propoed method on different ample ize with 50 ite. (a) OC48 IP Dataet Figure 6. The number of meage ent by Algorithm Broadcat and our propoed method on different number of ite (k) with a ample ize of 20. VI. CONCLUSION We preent new meage-efficient algorithm for ditinct random ampling on a ditributed data tream. Our algorithm are practical, eay to implement, and have nearoptimal expected meage complexity. Surpriingly, the expected meage complexity of maintaining a ditinct random ample from a ditributed manner i inherently greater than that of maintaining a imple random ample from a ditributed tream. Our analyi of the algorithm ha ued the ynchronou model of a ditributed ytem, ince it i clearer to analyze the meage complexity of a ditributed computation within a ynchronou ytem than within an aynchronou ytem. However, our algorithm are not dependent on there being ynchronou lock-tep execution of proceor within the ytem. In particular, the paradigm of chooing the element with the mallet hah value a the random ample continue to work in the preence of timing inconitencie, even if meage are delayed or reordered during tranmiion. Some direction for future work are a follow. Our lower bound for liding window are not tight, and it i intereting to obtain an optimal algorithm for DDS on liding window either through a better lower bound or a better algorithm. It i alo intereting to derive ditributed ampling algorithm for more complex tructure on data, uch a tructure on graph. REFERENCES [1] P. B. Gibbon, Ditinct ampling for highly-accurate anwer to ditinct value querie and event report, in Proceeding of the 27th International Conference on Very Large Data Bae (VLDB), 2001, pp [2] G. Cormode and M. Garofalaki, Sketching tream through the net: ditributed approximate query tracking, in Proceeding of the 31t International Conference on Very Large Data Bae. VLDB Endowment, 2005, pp [3] S. Venkataraman, D. X. Song, P. B. Gibbon, and A. Blum, New treaming algorithm for fat detection of uperpreader, in Proceeding of the Network and Ditributed Sytem Security Sympoium, NDSS, (a) OC48 IP Dataet Figure 7. The number of meage ent by Algorithm Broadcat and our propoed method, a a function of the kew, for 20 ite and a ample ize of 20. [4] A. Poddar, One pa ditinct ampling, file.wordpre.com/2011/12/one-pa-ditinct-ampling.pdf. [5] T. Johnon, S. Muthukrihnan, and I. Rozenbaum, Sampling algorithm in a tream operator, in Proceeding of the ACM SIGMOD International Conference on Management of Data, 2005, pp

10 [6] S. Agarwal, B. Mozafari, A. Panda, H. Milner, S. Madden, and I. Stoica, Blinkdb: querie with bounded error and bounded repone time on very large data, in Eighth Euroy Conference, 2013, pp [7] IBM Corporation, Infophere tream, com/oftware/product/en/infophere-tream, acceed Jan [8] M. Zaharia, T. Da, H. Li, T. Hunter, S. Shenker, and I. Stoica, Dicretized tream: Fault-tolerant treaming computation at cale, in Proc. ACM Sympoium on Operating Sytem Principle (SOSP), 2013, pp [9] Apache Software Foundation, Storm: Ditributed and faulttolerant realtime computation, project/torm.html. [10] G. Cormode, S. Muthukrihnan, K. Yi, and Q. Zhang, Continuou ampling from ditributed tream, Journal of the ACM, vol. 59, no. 2, pp. 10:1 10:25. [11] D. P. Woodruff and Q. Zhang, Tight bound for ditributed functional monitoring, in Proceeding of the Forty-fourth Annual ACM Sympoium on Theory of Computing, er. STOC 12, pp [12] S. Tirthapura and D. P. Woodruff, Optimal random ampling from ditributed tream reviited, in Proc. International Sympoium on Ditributed Computing (DISC), 2011, pp [13] G. Cormode, The continuou ditributed monitoring model, SIGMOD Record, vol. 42, no. 1, pp. 5 14, May [14] J. S. Vitter, Random ampling with a reervoir, ACM Tranaction on Mathematical Software, vol. 11, no. 1, pp , [15] P. S. Efraimidi and P. G. Spiraki, Weighted random ampling with a reervoir, Information Proceing Letter, vol. 97, no. 5, pp , [16] V. Braverman, R. Otrovky, and C. Zaniolo, Optimal ampling from liding window, Journal of Computer and Sytem Science, vol. 78, pp , [17] B. Babcock, M. Datar, and R. Motwani, Sampling from a moving window over treaming data, in Proceeding of the Thirteenth Annual ACM-SIAM Sympoium on Dicrete Algorithm, 2002, pp [18] R. Gemulla and W. Lehner, Sampling time-baed liding window in bounded pace, in Proceeding of the 2008 ACM SIGMOD International Conference on Management of Data, er. SIGMOD 08, pp [19] C. Aggarwal, On biaed reervoir ampling in the preence of tream evolution, in Proceeding of the 32nd International Conference on Very Large Databae, 2006, pp [20] P. Gibbon and S. Tirthapura, Etimating imple function on the union of data tream, in Proc. ACM Symp. on Parallel Algorithm and Architecture (SPAA), 2001, pp [21] S. Ganguly, M. Garofalaki, and R. Ratogi, Tracking etexpreion cardinalitie over continuou update tream, The VLDB Journal, vol. 13, no. 4, pp , [22] G. Frahling, P. Indyk, and C. Sohler, Sampling in dynamic data tream and application, in Proceeding of the Twentyfirt Annual Sympoium on Computational Geometry, er. SCG 05. ACM, pp [23] G. Cormode, S. Muthukrihnan, and K. Yi, Algorithm for ditributed functional monitoring, in Proceeding of the Nineteenth Annual ACM-SIAM Sympoium on Dicrete Algorithm, 2008, pp [24] C. Arackaparambil, J. Brody, and A. Chakrabarti, Functional monitoring without monotonicity, in Proceeding of the 36th International Colloquium on Automata, Language and Programming: Part I, 2009, pp [25] K. Yi and Q. Zhang, Optimal tracking of ditributed heavy hitter and quantile, Algorithmica, vol. 65, no. 1, pp , [26] I. Sharfman, A. Schuter, and D. Keren, A geometric approach to monitoring threhold function over ditributed data tream, ACM Tranaction on Databae Sytem, vol. 32, no. 4, [27] N. Giatrako, A. Deligiannaki, M. N. Garofalaki, I. Sharfman, and A. Schuter, Prediction-baed geometric monitoring over ditributed data tream, in Proceeding of the 2012 ACM SIGMOD International Conference on Management of Data, er. SIGMOD 12, pp [28] B. Lahiri and S. Tirthapura, Stream ampling, in Encyclopedia of Databae Sytem, 2009, pp [29] P. B. Gibbon and S. Tirthapura, Ditributed tream algorithm for liding window, in Proceeding of the Fourteenth Annual ACM Sympoium on Parallel Algorithm and Architecture, er. SPAA 02, pp [30] Y.-Y. Chung and S. Tirthapura, Ditinct random ampling from a ditributed tream, nt/ pub/dd.pdf, Oct [31] G. Cormode, S. Muthukrihnan, K. Yi, and Q. Zhang, Optimal ampling from ditributed tream, in Proceeding of the Twenty-ninth ACM SIGMOD-SIGACT-SIGART Sympoium on Principle of Databae Sytem, 2010, pp [32] M. Datar, A. Gioni, P. Indyk, and R. Motwani, Maintaining tream tatitic over liding window, SIAM Journal on Computing, vol. 31, no. 6, pp , [33] R. Seidel and C. R. Aragon, Randomized earch tree, in Algorithmica, vol. 16, no Springer-Verlag, 1996, pp [34] CAIDA OC48 Trace Project, The caida ucd oc48 internet trace dataet - ( ), M=CAIDA+OC48+Trace+Project, [35] CALO Project, Enron dataet, edu/ enron/. [36] V. Holub, Murmur hah 2.0, holub/ w/javamurmurhah/murmurhah.java.

Theoretical Computer Science. Optimal algorithms for online scheduling with bounded rearrangement at the end

Theoretical Computer Science. Optimal algorithms for online scheduling with bounded rearrangement at the end Theoretical Computer Science 4 (0) 669 678 Content lit available at SciVere ScienceDirect Theoretical Computer Science journal homepage: www.elevier.com/locate/tc Optimal algorithm for online cheduling

More information

Social Studies 201 Notes for November 14, 2003

Social Studies 201 Notes for November 14, 2003 1 Social Studie 201 Note for November 14, 2003 Etimation of a mean, mall ample ize Section 8.4, p. 501. When a reearcher ha only a mall ample ize available, the central limit theorem doe not apply to the

More information

Codes Correcting Two Deletions

Codes Correcting Two Deletions 1 Code Correcting Two Deletion Ryan Gabry and Frederic Sala Spawar Sytem Center Univerity of California, Lo Angele ryan.gabry@navy.mil fredala@ucla.edu Abtract In thi work, we invetigate the problem of

More information

Preemptive scheduling on a small number of hierarchical machines

Preemptive scheduling on a small number of hierarchical machines Available online at www.ciencedirect.com Information and Computation 06 (008) 60 619 www.elevier.com/locate/ic Preemptive cheduling on a mall number of hierarchical machine György Dóa a, Leah Eptein b,

More information

Chapter 2 Sampling and Quantization. In order to investigate sampling and quantization, the difference between analog

Chapter 2 Sampling and Quantization. In order to investigate sampling and quantization, the difference between analog Chapter Sampling and Quantization.1 Analog and Digital Signal In order to invetigate ampling and quantization, the difference between analog and digital ignal mut be undertood. Analog ignal conit of continuou

More information

Lecture 7: Testing Distributions

Lecture 7: Testing Distributions CSE 5: Sublinear (and Streaming) Algorithm Spring 014 Lecture 7: Teting Ditribution April 1, 014 Lecturer: Paul Beame Scribe: Paul Beame 1 Teting Uniformity of Ditribution We return today to property teting

More information

List coloring hypergraphs

List coloring hypergraphs Lit coloring hypergraph Penny Haxell Jacque Vertraete Department of Combinatoric and Optimization Univerity of Waterloo Waterloo, Ontario, Canada pehaxell@uwaterloo.ca Department of Mathematic Univerity

More information

Problem Set 8 Solutions

Problem Set 8 Solutions Deign and Analyi of Algorithm April 29, 2015 Maachuett Intitute of Technology 6.046J/18.410J Prof. Erik Demaine, Srini Devada, and Nancy Lynch Problem Set 8 Solution Problem Set 8 Solution Thi problem

More information

CHAPTER 8 OBSERVER BASED REDUCED ORDER CONTROLLER DESIGN FOR LARGE SCALE LINEAR DISCRETE-TIME CONTROL SYSTEMS

CHAPTER 8 OBSERVER BASED REDUCED ORDER CONTROLLER DESIGN FOR LARGE SCALE LINEAR DISCRETE-TIME CONTROL SYSTEMS CHAPTER 8 OBSERVER BASED REDUCED ORDER CONTROLLER DESIGN FOR LARGE SCALE LINEAR DISCRETE-TIME CONTROL SYSTEMS 8.1 INTRODUCTION 8.2 REDUCED ORDER MODEL DESIGN FOR LINEAR DISCRETE-TIME CONTROL SYSTEMS 8.3

More information

into a discrete time function. Recall that the table of Laplace/z-transforms is constructed by (i) selecting to get

into a discrete time function. Recall that the table of Laplace/z-transforms is constructed by (i) selecting to get Lecture 25 Introduction to Some Matlab c2d Code in Relation to Sampled Sytem here are many way to convert a continuou time function, { h( t) ; t [0, )} into a dicrete time function { h ( k) ; k {0,,, }}

More information

Lecture 9: Shor s Algorithm

Lecture 9: Shor s Algorithm Quantum Computation (CMU 8-859BB, Fall 05) Lecture 9: Shor Algorithm October 7, 05 Lecturer: Ryan O Donnell Scribe: Sidhanth Mohanty Overview Let u recall the period finding problem that wa et up a a function

More information

GNSS Solutions: What is the carrier phase measurement? How is it generated in GNSS receivers? Simply put, the carrier phase

GNSS Solutions: What is the carrier phase measurement? How is it generated in GNSS receivers? Simply put, the carrier phase GNSS Solution: Carrier phae and it meaurement for GNSS GNSS Solution i a regular column featuring quetion and anwer about technical apect of GNSS. Reader are invited to end their quetion to the columnit,

More information

7.2 INVERSE TRANSFORMS AND TRANSFORMS OF DERIVATIVES 281

7.2 INVERSE TRANSFORMS AND TRANSFORMS OF DERIVATIVES 281 72 INVERSE TRANSFORMS AND TRANSFORMS OF DERIVATIVES 28 and i 2 Show how Euler formula (page 33) can then be ued to deduce the reult a ( a) 2 b 2 {e at co bt} {e at in bt} b ( a) 2 b 2 5 Under what condition

More information

Chapter Landscape of an Optimization Problem. Local Search. Coping With NP-Hardness. Gradient Descent: Vertex Cover

Chapter Landscape of an Optimization Problem. Local Search. Coping With NP-Hardness. Gradient Descent: Vertex Cover Coping With NP-Hardne Chapter 12 Local Search Q Suppoe I need to olve an NP-hard problem What hould I do? A Theory ay you're unlikely to find poly-time algorithm Mut acrifice one of three deired feature

More information

Competitive Analysis of Task Scheduling Algorithms on a Fault-Prone Machine and the Impact of Resource Augmentation

Competitive Analysis of Task Scheduling Algorithms on a Fault-Prone Machine and the Impact of Resource Augmentation Competitive Analyi of Tak Scheduling Algorithm on a Fault-Prone Machine and the Impact of Reource Augmentation Antonio Fernández Anta a, Chryi Georgiou b, Dariuz R. Kowalki c, Elli Zavou a,d,1 a Intitute

More information

RELIABILITY OF REPAIRABLE k out of n: F SYSTEM HAVING DISCRETE REPAIR AND FAILURE TIMES DISTRIBUTIONS

RELIABILITY OF REPAIRABLE k out of n: F SYSTEM HAVING DISCRETE REPAIR AND FAILURE TIMES DISTRIBUTIONS www.arpapre.com/volume/vol29iue1/ijrras_29_1_01.pdf RELIABILITY OF REPAIRABLE k out of n: F SYSTEM HAVING DISCRETE REPAIR AND FAILURE TIMES DISTRIBUTIONS Sevcan Demir Atalay 1,* & Özge Elmataş Gültekin

More information

Lecture 21. The Lovasz splitting-off lemma Topics in Combinatorial Optimization April 29th, 2004

Lecture 21. The Lovasz splitting-off lemma Topics in Combinatorial Optimization April 29th, 2004 18.997 Topic in Combinatorial Optimization April 29th, 2004 Lecture 21 Lecturer: Michel X. Goeman Scribe: Mohammad Mahdian 1 The Lovaz plitting-off lemma Lovaz plitting-off lemma tate the following. Theorem

More information

arxiv: v1 [math.mg] 25 Aug 2011

arxiv: v1 [math.mg] 25 Aug 2011 ABSORBING ANGLES, STEINER MINIMAL TREES, AND ANTIPODALITY HORST MARTINI, KONRAD J. SWANEPOEL, AND P. OLOFF DE WET arxiv:08.5046v [math.mg] 25 Aug 20 Abtract. We give a new proof that a tar {op i : i =,...,

More information

Lecture 8: Period Finding: Simon s Problem over Z N

Lecture 8: Period Finding: Simon s Problem over Z N Quantum Computation (CMU 8-859BB, Fall 205) Lecture 8: Period Finding: Simon Problem over Z October 5, 205 Lecturer: John Wright Scribe: icola Rech Problem A mentioned previouly, period finding i a rephraing

More information

Multicolor Sunflowers

Multicolor Sunflowers Multicolor Sunflower Dhruv Mubayi Lujia Wang October 19, 2017 Abtract A unflower i a collection of ditinct et uch that the interection of any two of them i the ame a the common interection C of all of

More information

Social Studies 201 Notes for March 18, 2005

Social Studies 201 Notes for March 18, 2005 1 Social Studie 201 Note for March 18, 2005 Etimation of a mean, mall ample ize Section 8.4, p. 501. When a reearcher ha only a mall ample ize available, the central limit theorem doe not apply to the

More information

New bounds for Morse clusters

New bounds for Morse clusters New bound for More cluter Tamá Vinkó Advanced Concept Team, European Space Agency, ESTEC Keplerlaan 1, 2201 AZ Noordwijk, The Netherland Tama.Vinko@ea.int and Arnold Neumaier Fakultät für Mathematik, Univerität

More information

Clustering Methods without Given Number of Clusters

Clustering Methods without Given Number of Clusters Clutering Method without Given Number of Cluter Peng Xu, Fei Liu Introduction A we now, mean method i a very effective algorithm of clutering. It mot powerful feature i the calability and implicity. However,

More information

A relationship between generalized Davenport-Schinzel sequences and interval chains

A relationship between generalized Davenport-Schinzel sequences and interval chains A relationhip between generalized Davenport-Schinzel equence and interval chain The MIT Faculty ha made thi article openly available. Pleae hare how thi acce benefit you. Your tory matter. Citation A Publihed

More information

Alternate Dispersion Measures in Replicated Factorial Experiments

Alternate Dispersion Measures in Replicated Factorial Experiments Alternate Diperion Meaure in Replicated Factorial Experiment Neal A. Mackertich The Raytheon Company, Sudbury MA 02421 Jame C. Benneyan Northeatern Univerity, Boton MA 02115 Peter D. Krau The Raytheon

More information

Bogoliubov Transformation in Classical Mechanics

Bogoliubov Transformation in Classical Mechanics Bogoliubov Tranformation in Claical Mechanic Canonical Tranformation Suppoe we have a et of complex canonical variable, {a j }, and would like to conider another et of variable, {b }, b b ({a j }). How

More information

Convex Hulls of Curves Sam Burton

Convex Hulls of Curves Sam Burton Convex Hull of Curve Sam Burton 1 Introduction Thi paper will primarily be concerned with determining the face of convex hull of curve of the form C = {(t, t a, t b ) t [ 1, 1]}, a < b N in R 3. We hall

More information

ON THE APPROXIMATION ERROR IN HIGH DIMENSIONAL MODEL REPRESENTATION. Xiaoqun Wang

ON THE APPROXIMATION ERROR IN HIGH DIMENSIONAL MODEL REPRESENTATION. Xiaoqun Wang Proceeding of the 2008 Winter Simulation Conference S. J. Maon, R. R. Hill, L. Mönch, O. Roe, T. Jefferon, J. W. Fowler ed. ON THE APPROXIMATION ERROR IN HIGH DIMENSIONAL MODEL REPRESENTATION Xiaoqun Wang

More information

Sampling and the Discrete Fourier Transform

Sampling and the Discrete Fourier Transform Sampling and the Dicrete Fourier Tranform Sampling Method Sampling i mot commonly done with two device, the ample-and-hold (S/H) and the analog-to-digital-converter (ADC) The S/H acquire a CT ignal at

More information

A NEW LOAD MODEL OF THE PEDESTRIANS LATERAL ACTION

A NEW LOAD MODEL OF THE PEDESTRIANS LATERAL ACTION A NEW LOAD MODEL OF THE PEDESTRIANS LATERAL ACTION Fiammetta VENUTI PhD Politecnico di Torino Torino, IT Luca Bruno Aociate Profeor Politecnico di Torino Torino, IT Summary Thi paper propoe a new load

More information

Control Systems Analysis and Design by the Root-Locus Method

Control Systems Analysis and Design by the Root-Locus Method 6 Control Sytem Analyi and Deign by the Root-Locu Method 6 1 INTRODUCTION The baic characteritic of the tranient repone of a cloed-loop ytem i cloely related to the location of the cloed-loop pole. If

More information

UNIT 15 RELIABILITY EVALUATION OF k-out-of-n AND STANDBY SYSTEMS

UNIT 15 RELIABILITY EVALUATION OF k-out-of-n AND STANDBY SYSTEMS UNIT 1 RELIABILITY EVALUATION OF k-out-of-n AND STANDBY SYSTEMS Structure 1.1 Introduction Objective 1.2 Redundancy 1.3 Reliability of k-out-of-n Sytem 1.4 Reliability of Standby Sytem 1. Summary 1.6 Solution/Anwer

More information

μ + = σ = D 4 σ = D 3 σ = σ = All units in parts (a) and (b) are in V. (1) x chart: Center = μ = 0.75 UCL =

μ + = σ = D 4 σ = D 3 σ = σ = All units in parts (a) and (b) are in V. (1) x chart: Center = μ = 0.75 UCL = Our online Tutor are available 4*7 to provide Help with Proce control ytem Homework/Aignment or a long term Graduate/Undergraduate Proce control ytem Project. Our Tutor being experienced and proficient

More information

IEOR 3106: Fall 2013, Professor Whitt Topics for Discussion: Tuesday, November 19 Alternating Renewal Processes and The Renewal Equation

IEOR 3106: Fall 2013, Professor Whitt Topics for Discussion: Tuesday, November 19 Alternating Renewal Processes and The Renewal Equation IEOR 316: Fall 213, Profeor Whitt Topic for Dicuion: Tueday, November 19 Alternating Renewal Procee and The Renewal Equation 1 Alternating Renewal Procee An alternating renewal proce alternate between

More information

Online Parallel Scheduling of Non-uniform Tasks: Trading Failures for Energy

Online Parallel Scheduling of Non-uniform Tasks: Trading Failures for Energy Online Parallel Scheduling of Non-uniform Tak: Trading Failure for Energy Antonio Fernández Anta a, Chryi Georgiou b, Dariuz R. Kowalki c, Elli Zavou a,d,1 a Intitute IMDEA Network b Univerity of Cypru

More information

Beyond Cut-Set Bounds - The Approximate Capacity of D2D Networks

Beyond Cut-Set Bounds - The Approximate Capacity of D2D Networks Beyond Cut-Set Bound - The Approximate Capacity of DD Network Avik Sengupta Hume Center for National Security and Technology Department of Electrical and Computer Engineering Virginia Tech, Blackburg,

More information

Chapter 4. The Laplace Transform Method

Chapter 4. The Laplace Transform Method Chapter 4. The Laplace Tranform Method The Laplace Tranform i a tranformation, meaning that it change a function into a new function. Actually, it i a linear tranformation, becaue it convert a linear combination

More information

Technical Appendix: Auxiliary Results and Proofs

Technical Appendix: Auxiliary Results and Proofs A Technical Appendix: Auxiliary Reult and Proof Lemma A. The following propertie hold for q (j) = F r [c + ( ( )) ] de- ned in Lemma. (i) q (j) >, 8 (; ]; (ii) R q (j)d = ( ) q (j) + R q (j)d ; (iii) R

More information

TRIPLE SOLUTIONS FOR THE ONE-DIMENSIONAL

TRIPLE SOLUTIONS FOR THE ONE-DIMENSIONAL GLASNIK MATEMATIČKI Vol. 38583, 73 84 TRIPLE SOLUTIONS FOR THE ONE-DIMENSIONAL p-laplacian Haihen Lü, Donal O Regan and Ravi P. Agarwal Academy of Mathematic and Sytem Science, Beijing, China, National

More information

SPREAD: An Adaptive Scheme for Redundant and Fair Storage in Dynamic Heterogeneous Storage Systems

SPREAD: An Adaptive Scheme for Redundant and Fair Storage in Dynamic Heterogeneous Storage Systems SPREAD: An Adaptive Scheme for Redundant and Fair Storage in Dynamic Heterogeneou Storage Sytem Mario Mene Chritian Scheideler Abtract In thi paper we tudy the problem of deigning an adaptive hah table

More information

Unbounded solutions of second order discrete BVPs on infinite intervals

Unbounded solutions of second order discrete BVPs on infinite intervals Available online at www.tjna.com J. Nonlinear Sci. Appl. 9 206), 357 369 Reearch Article Unbounded olution of econd order dicrete BVP on infinite interval Hairong Lian a,, Jingwu Li a, Ravi P Agarwal b

More information

Stochastic Optimization with Inequality Constraints Using Simultaneous Perturbations and Penalty Functions

Stochastic Optimization with Inequality Constraints Using Simultaneous Perturbations and Penalty Functions Stochatic Optimization with Inequality Contraint Uing Simultaneou Perturbation and Penalty Function I-Jeng Wang* and Jame C. Spall** The John Hopkin Univerity Applied Phyic Laboratory 11100 John Hopkin

More information

CHAPTER 4 DESIGN OF STATE FEEDBACK CONTROLLERS AND STATE OBSERVERS USING REDUCED ORDER MODEL

CHAPTER 4 DESIGN OF STATE FEEDBACK CONTROLLERS AND STATE OBSERVERS USING REDUCED ORDER MODEL 98 CHAPTER DESIGN OF STATE FEEDBACK CONTROLLERS AND STATE OBSERVERS USING REDUCED ORDER MODEL INTRODUCTION The deign of ytem uing tate pace model for the deign i called a modern control deign and it i

More information

EFFECT ON PERSISTENCE OF INTRA-SPECIFIC COMPETITION IN COMPETITION MODELS

EFFECT ON PERSISTENCE OF INTRA-SPECIFIC COMPETITION IN COMPETITION MODELS Electronic Journal of Differential Equation, Vol. 2007(2007, No. 25, pp. 0. ISSN: 072-669. URL: http://ejde.math.txtate.edu or http://ejde.math.unt.edu ftp ejde.math.txtate.edu (login: ftp EFFECT ON PERSISTENCE

More information

Factor Analysis with Poisson Output

Factor Analysis with Poisson Output Factor Analyi with Poion Output Gopal Santhanam Byron Yu Krihna V. Shenoy, Department of Electrical Engineering, Neurocience Program Stanford Univerity Stanford, CA 94305, USA {gopal,byronyu,henoy}@tanford.edu

More information

Avoiding Forbidden Submatrices by Row Deletions

Avoiding Forbidden Submatrices by Row Deletions Avoiding Forbidden Submatrice by Row Deletion Sebatian Wernicke, Jochen Alber, Jen Gramm, Jiong Guo, and Rolf Niedermeier Wilhelm-Schickard-Intitut für Informatik, niverität Tübingen, Sand 13, D-72076

More information

White Rose Research Online URL for this paper: Version: Accepted Version

White Rose Research Online URL for this paper:   Version: Accepted Version Thi i a repoitory copy of Identification of nonlinear ytem with non-peritent excitation uing an iterative forward orthogonal leat quare regreion algorithm. White Roe Reearch Online URL for thi paper: http://eprint.whiteroe.ac.uk/107314/

More information

SMALL-SIGNAL STABILITY ASSESSMENT OF THE EUROPEAN POWER SYSTEM BASED ON ADVANCED NEURAL NETWORK METHOD

SMALL-SIGNAL STABILITY ASSESSMENT OF THE EUROPEAN POWER SYSTEM BASED ON ADVANCED NEURAL NETWORK METHOD SMALL-SIGNAL STABILITY ASSESSMENT OF THE EUROPEAN POWER SYSTEM BASED ON ADVANCED NEURAL NETWORK METHOD S.P. Teeuwen, I. Erlich U. Bachmann Univerity of Duiburg, Germany Department of Electrical Power Sytem

More information

A BATCH-ARRIVAL QUEUE WITH MULTIPLE SERVERS AND FUZZY PARAMETERS: PARAMETRIC PROGRAMMING APPROACH

A BATCH-ARRIVAL QUEUE WITH MULTIPLE SERVERS AND FUZZY PARAMETERS: PARAMETRIC PROGRAMMING APPROACH Mathematical and Computational Application Vol. 11 No. pp. 181-191 006. Aociation for Scientific Reearch A BATCH-ARRIVA QEE WITH MTIPE SERVERS AND FZZY PARAMETERS: PARAMETRIC PROGRAMMING APPROACH Jau-Chuan

More information

A Provably Secure Scheme for Remote User Authentication

A Provably Secure Scheme for Remote User Authentication A Provably Secure Scheme for Remote Uer Authentication Fuw-Yi Yang 1, Su-Hui Chiu 2 1 Department of Computer Science and Information Engineering, Chaoyang Univerity of Technology Taichung County 41349,

More information

Gain and Phase Margins Based Delay Dependent Stability Analysis of Two- Area LFC System with Communication Delays

Gain and Phase Margins Based Delay Dependent Stability Analysis of Two- Area LFC System with Communication Delays Gain and Phae Margin Baed Delay Dependent Stability Analyi of Two- Area LFC Sytem with Communication Delay Şahin Sönmez and Saffet Ayaun Department of Electrical Engineering, Niğde Ömer Halidemir Univerity,

More information

COHOMOLOGY AS A LOCAL-TO-GLOBAL BRIDGE

COHOMOLOGY AS A LOCAL-TO-GLOBAL BRIDGE COHOMOLOGY AS A LOCAL-TO-GLOBAL BRIDGE LIVIU I. NICOLAESCU ABSTRACT. I dicu low dimenional incarnation of cohomology and illutrate how baic cohomological principle lead to a proof of Sperner lemma. CONTENTS.

More information

Microblog Hot Spot Mining Based on PAM Probabilistic Topic Model

Microblog Hot Spot Mining Based on PAM Probabilistic Topic Model MATEC Web of Conference 22, 01062 ( 2015) DOI: 10.1051/ matecconf/ 2015220106 2 C Owned by the author, publihed by EDP Science, 2015 Microblog Hot Spot Mining Baed on PAM Probabilitic Topic Model Yaxin

More information

THE SPLITTING SUBSPACE CONJECTURE

THE SPLITTING SUBSPACE CONJECTURE THE SPLITTING SUBSPAE ONJETURE ERI HEN AND DENNIS TSENG Abtract We anwer a uetion by Niederreiter concerning the enumeration of a cla of ubpace of finite dimenional vector pace over finite field by proving

More information

[Saxena, 2(9): September, 2013] ISSN: Impact Factor: INTERNATIONAL JOURNAL OF ENGINEERING SCIENCES & RESEARCH TECHNOLOGY

[Saxena, 2(9): September, 2013] ISSN: Impact Factor: INTERNATIONAL JOURNAL OF ENGINEERING SCIENCES & RESEARCH TECHNOLOGY [Saena, (9): September, 0] ISSN: 77-9655 Impact Factor:.85 IJESRT INTERNATIONAL JOURNAL OF ENGINEERING SCIENCES & RESEARCH TECHNOLOGY Contant Stre Accelerated Life Teting Uing Rayleigh Geometric Proce

More information

CHAPTER 6. Estimation

CHAPTER 6. Estimation CHAPTER 6 Etimation Definition. Statitical inference i the procedure by which we reach a concluion about a population on the bai of information contained in a ample drawn from that population. Definition.

More information

Lecture 10 Filtering: Applied Concepts

Lecture 10 Filtering: Applied Concepts Lecture Filtering: Applied Concept In the previou two lecture, you have learned about finite-impule-repone (FIR) and infinite-impule-repone (IIR) filter. In thee lecture, we introduced the concept of filtering

More information

SOME RESULTS ON INFINITE POWER TOWERS

SOME RESULTS ON INFINITE POWER TOWERS NNTDM 16 2010) 3, 18-24 SOME RESULTS ON INFINITE POWER TOWERS Mladen Vailev - Miana 5, V. Hugo Str., Sofia 1124, Bulgaria E-mail:miana@abv.bg Abtract To my friend Kratyu Gumnerov In the paper the infinite

More information

FUNDAMENTALS OF POWER SYSTEMS

FUNDAMENTALS OF POWER SYSTEMS 1 FUNDAMENTALS OF POWER SYSTEMS 1 Chapter FUNDAMENTALS OF POWER SYSTEMS INTRODUCTION The three baic element of electrical engineering are reitor, inductor and capacitor. The reitor conume ohmic or diipative

More information

Optimal Random Sampling from Distributed Streams Revisited

Optimal Random Sampling from Distributed Streams Revisited Optimal Random Sampling from Distributed Streams Revisited Srikanta Tirthapura 1 and David P. Woodruff 2 1 Dept. of ECE, Iowa State University, Ames, IA, 50011, USA. snt@iastate.edu 2 IBM Almaden Research

More information

LDPC Convolutional Codes Based on Permutation Polynomials over Integer Rings

LDPC Convolutional Codes Based on Permutation Polynomials over Integer Rings LDPC Convolutional Code Baed on Permutation Polynomial over Integer Ring Marco B. S. Tavare and Gerhard P. Fettwei Vodafone Chair Mobile Communication Sytem, Dreden Univerity of Technology, 01062 Dreden,

More information

STOCHASTIC GENERALIZED TRANSPORTATION PROBLEM WITH DISCRETE DISTRIBUTION OF DEMAND

STOCHASTIC GENERALIZED TRANSPORTATION PROBLEM WITH DISCRETE DISTRIBUTION OF DEMAND OPERATIONS RESEARCH AND DECISIONS No. 4 203 DOI: 0.5277/ord30402 Marcin ANHOLCER STOCHASTIC GENERALIZED TRANSPORTATION PROBLEM WITH DISCRETE DISTRIBUTION OF DEMAND The generalized tranportation problem

More information

Ecient Parallel Algorithms for Computing All Pair. University of Kentucky. Duke University

Ecient Parallel Algorithms for Computing All Pair. University of Kentucky. Duke University Ecient Parallel Algorithm for Computing All Pair Shortet Path in Directed Graph 1 Yijie Han 2 * Victor Y. Pan 3 ** John H. Reif*** *Department of Computer Science Univerity of Kentucky Lexington, KY 40506

More information

Optimal Coordination of Samples in Business Surveys

Optimal Coordination of Samples in Business Surveys Paper preented at the ICES-III, June 8-, 007, Montreal, Quebec, Canada Optimal Coordination of Sample in Buine Survey enka Mach, Ioana Şchiopu-Kratina, Philip T Rei, Jean-Marc Fillion Statitic Canada New

More information

Memoryle Strategie in Concurrent Game with Reachability Objective Λ Krihnendu Chatterjee y Luca de Alfaro x Thoma A. Henzinger y;z y EECS, Univerity o

Memoryle Strategie in Concurrent Game with Reachability Objective Λ Krihnendu Chatterjee y Luca de Alfaro x Thoma A. Henzinger y;z y EECS, Univerity o Memoryle Strategie in Concurrent Game with Reachability Objective Krihnendu Chatterjee, Luca de Alfaro and Thoma A. Henzinger Report No. UCB/CSD-5-1406 Augut 2005 Computer Science Diviion (EECS) Univerity

More information

Exploiting A Support-based Upper Bound of Pearson s Correlation Coefficient for Efficiently Identifying Strongly Correlated Pairs

Exploiting A Support-based Upper Bound of Pearson s Correlation Coefficient for Efficiently Identifying Strongly Correlated Pairs Exploiting A Support-baed Upper Bound of Pearon Correlation Coefficient for Efficiently Identifying Strongly Correlated Pair Hui Xiong Computer Science Univerity of Minneota huix@c.umn.edu Shahi Shekhar

More information

CDMA Signature Sequences with Low Peak-to-Average-Power Ratio via Alternating Projection

CDMA Signature Sequences with Low Peak-to-Average-Power Ratio via Alternating Projection CDMA Signature Sequence with Low Peak-to-Average-Power Ratio via Alternating Projection Joel A Tropp Int for Comp Engr and Sci (ICES) The Univerity of Texa at Autin 1 Univerity Station C0200 Autin, TX

More information

On the Stability Region of Congestion Control

On the Stability Region of Congestion Control On the Stability Region of Congetion Control Xiaojun Lin and Ne B. Shroff School of Electrical and Computer Engineering Purdue Univerity, Wet Lafayette, IN 47906 {linx,hroff}@ecn.purdue.edu Abtract It

More information

One Class of Splitting Iterative Schemes

One Class of Splitting Iterative Schemes One Cla of Splitting Iterative Scheme v Ciegi and V. Pakalnytė Vilniu Gedimina Technical Univerity Saulėtekio al. 11, 2054, Vilniu, Lithuania rc@fm.vtu.lt Abtract. Thi paper deal with the tability analyi

More information

Source slideplayer.com/fundamentals of Analytical Chemistry, F.J. Holler, S.R.Crouch. Chapter 6: Random Errors in Chemical Analysis

Source slideplayer.com/fundamentals of Analytical Chemistry, F.J. Holler, S.R.Crouch. Chapter 6: Random Errors in Chemical Analysis Source lideplayer.com/fundamental of Analytical Chemitry, F.J. Holler, S.R.Crouch Chapter 6: Random Error in Chemical Analyi Random error are preent in every meaurement no matter how careful the experimenter.

More information

Integration Testing - Looking for a Solution to Testing Concurrent Components and Systems

Integration Testing - Looking for a Solution to Testing Concurrent Components and Systems Integration Teting - Looking for a Solution to Teting Concurrent Component and Sytem Integration Teting 23. TAV 1 Andrej Pietchker, PhD Siemen AG, CT SE 1 andrej.pietchker@iemen.com ph: +49 (89) 636 55130

More information

S_LOOP: SINGLE-LOOP FEEDBACK CONTROL SYSTEM ANALYSIS

S_LOOP: SINGLE-LOOP FEEDBACK CONTROL SYSTEM ANALYSIS S_LOOP: SINGLE-LOOP FEEDBACK CONTROL SYSTEM ANALYSIS by Michelle Gretzinger, Daniel Zyngier and Thoma Marlin INTRODUCTION One of the challenge to the engineer learning proce control i relating theoretical

More information

The Use of MDL to Select among Computational Models of Cognition

The Use of MDL to Select among Computational Models of Cognition The Ue of DL to Select among Computational odel of Cognition In J. yung, ark A. Pitt & Shaobo Zhang Vijay Balaubramanian Department of Pychology David Rittenhoue Laboratorie Ohio State Univerity Univerity

More information

Approximating discrete probability distributions with Bayesian networks

Approximating discrete probability distributions with Bayesian networks Approximating dicrete probability ditribution with Bayeian network Jon Williamon Department of Philoophy King College, Str and, London, WC2R 2LS, UK Abtract I generalie the argument of [Chow & Liu 1968]

More information

Quantitative Information Leakage. Lecture 9

Quantitative Information Leakage. Lecture 9 Quantitative Information Leakage Lecture 9 1 The baic model: Sytem = Information-Theoretic channel Secret Information Obervable 1 o1... Sytem... m on Input Output 2 Toward a quantitative notion of leakage

More information

Primitive Digraphs with the Largest Scrambling Index

Primitive Digraphs with the Largest Scrambling Index Primitive Digraph with the Larget Scrambling Index Mahmud Akelbek, Steve Kirkl 1 Department of Mathematic Statitic, Univerity of Regina, Regina, Sakatchewan, Canada S4S 0A Abtract The crambling index of

More information

OVERFLOW PROBABILITY IN AN ATM QUEUE WITH SELF-SIMILAR INPUT TRAFFIC

OVERFLOW PROBABILITY IN AN ATM QUEUE WITH SELF-SIMILAR INPUT TRAFFIC Copyright by IEEE OVERFLOW PROBABILITY IN AN ATM QUEUE WITH SELF-SIMILAR INPUT TRAFFIC Bori Tybakov Intitute for Problem in Information Tranmiion Ruian Academy of Science Mocow, Ruia e-mail: bt@ippi ac

More information

Connectivity in large mobile ad-hoc networks

Connectivity in large mobile ad-hoc networks Weiertraß-Intitut für Angewandte Analyi und Stochatik Connectivity in large mobile ad-hoc network WOLFGANG KÖNIG (WIAS und U Berlin) joint work with HANNA DÖRING (Onabrück) and GABRIEL FARAUD (Pari) Mohrentraße

More information

Chapter 4: Applications of Fourier Representations. Chih-Wei Liu

Chapter 4: Applications of Fourier Representations. Chih-Wei Liu Chapter 4: Application of Fourier Repreentation Chih-Wei Liu Outline Introduction Fourier ranform of Periodic Signal Convolution/Multiplication with Non-Periodic Signal Fourier ranform of Dicrete-ime Signal

More information

Comparing Means: t-tests for Two Independent Samples

Comparing Means: t-tests for Two Independent Samples Comparing ean: t-tet for Two Independent Sample Independent-eaure Deign t-tet for Two Independent Sample Allow reearcher to evaluate the mean difference between two population uing data from two eparate

More information

Reliability Analysis of Embedded System with Different Modes of Failure Emphasizing Reboot Delay

Reliability Analysis of Embedded System with Different Modes of Failure Emphasizing Reboot Delay International Journal of Applied Science and Engineering 3., 4: 449-47 Reliability Analyi of Embedded Sytem with Different Mode of Failure Emphaizing Reboot Delay Deepak Kumar* and S. B. Singh Department

More information

On the Isomorphism of Fractional Factorial Designs 1

On the Isomorphism of Fractional Factorial Designs 1 journal of complexity 17, 8697 (2001) doi:10.1006jcom.2000.0569, available online at http:www.idealibrary.com on On the Iomorphim of Fractional Factorial Deign 1 Chang-Xing Ma Department of Statitic, Nankai

More information

DYNAMIC MODELS FOR CONTROLLER DESIGN

DYNAMIC MODELS FOR CONTROLLER DESIGN DYNAMIC MODELS FOR CONTROLLER DESIGN M.T. Tham (996,999) Dept. of Chemical and Proce Engineering Newcatle upon Tyne, NE 7RU, UK.. INTRODUCTION The problem of deigning a good control ytem i baically that

More information

Suggested Answers To Exercises. estimates variability in a sampling distribution of random means. About 68% of means fall

Suggested Answers To Exercises. estimates variability in a sampling distribution of random means. About 68% of means fall Beyond Significance Teting ( nd Edition), Rex B. Kline Suggeted Anwer To Exercie Chapter. The tatitic meaure variability among core at the cae level. In a normal ditribution, about 68% of the core fall

More information

Math Skills. Scientific Notation. Uncertainty in Measurements. Appendix A5 SKILLS HANDBOOK

Math Skills. Scientific Notation. Uncertainty in Measurements. Appendix A5 SKILLS HANDBOOK ppendix 5 Scientific Notation It i difficult to work with very large or very mall number when they are written in common decimal notation. Uually it i poible to accommodate uch number by changing the SI

More information

arxiv: v4 [math.co] 21 Sep 2014

arxiv: v4 [math.co] 21 Sep 2014 ASYMPTOTIC IMPROVEMENT OF THE SUNFLOWER BOUND arxiv:408.367v4 [math.co] 2 Sep 204 JUNICHIRO FUKUYAMA Abtract. A unflower with a core Y i a family B of et uch that U U Y for each two different element U

More information

Moment of Inertia of an Equilateral Triangle with Pivot at one Vertex

Moment of Inertia of an Equilateral Triangle with Pivot at one Vertex oment of nertia of an Equilateral Triangle with Pivot at one Vertex There are two wa (at leat) to derive the expreion f an equilateral triangle that i rotated about one vertex, and ll how ou both here.

More information

Control of Delayed Integrating Processes Using Two Feedback Controllers R MS Approach

Control of Delayed Integrating Processes Using Two Feedback Controllers R MS Approach Proceeding of the 7th WSEAS International Conference on SYSTEM SCIENCE and SIMULATION in ENGINEERING (ICOSSSE '8) Control of Delayed Integrating Procee Uing Two Feedback Controller R MS Approach LIBOR

More information

CONTROL SYSTEMS, ROBOTICS AND AUTOMATION Vol. VIII Decoupling Control - M. Fikar

CONTROL SYSTEMS, ROBOTICS AND AUTOMATION Vol. VIII Decoupling Control - M. Fikar DECOUPLING CONTROL M. Fikar Department of Proce Control, Faculty of Chemical and Food Technology, Slovak Univerity of Technology in Bratilava, Radlinkého 9, SK-812 37 Bratilava, Slovakia Keyword: Decoupling:

More information

A Constraint Propagation Algorithm for Determining the Stability Margin. The paper addresses the stability margin assessment for linear systems

A Constraint Propagation Algorithm for Determining the Stability Margin. The paper addresses the stability margin assessment for linear systems A Contraint Propagation Algorithm for Determining the Stability Margin of Linear Parameter Circuit and Sytem Lubomir Kolev and Simona Filipova-Petrakieva Abtract The paper addree the tability margin aement

More information

Design By Emulation (Indirect Method)

Design By Emulation (Indirect Method) Deign By Emulation (Indirect Method he baic trategy here i, that Given a continuou tranfer function, it i required to find the bet dicrete equivalent uch that the ignal produced by paing an input ignal

More information

Acceptance sampling uses sampling procedure to determine whether to

Acceptance sampling uses sampling procedure to determine whether to DOI: 0.545/mji.203.20 Bayeian Repetitive Deferred Sampling Plan Indexed Through Relative Slope K.K. Sureh, S. Umamahewari and K. Pradeepa Veerakumari Department of Statitic, Bharathiar Univerity, Coimbatore,

More information

The Impact of Imperfect Scheduling on Cross-Layer Rate. Control in Multihop Wireless Networks

The Impact of Imperfect Scheduling on Cross-Layer Rate. Control in Multihop Wireless Networks The mpact of mperfect Scheduling on Cro-Layer Rate Control in Multihop Wirele Network Xiaojun Lin and Ne B. Shroff Center for Wirele Sytem and Application (CWSA) School of Electrical and Computer Engineering,

More information

Multi-dimensional Fuzzy Euler Approximation

Multi-dimensional Fuzzy Euler Approximation Mathematica Aeterna, Vol 7, 2017, no 2, 163-176 Multi-dimenional Fuzzy Euler Approximation Yangyang Hao College of Mathematic and Information Science Hebei Univerity, Baoding 071002, China hdhyywa@163com

More information

Flag-transitive non-symmetric 2-designs with (r, λ) = 1 and alternating socle

Flag-transitive non-symmetric 2-designs with (r, λ) = 1 and alternating socle Flag-tranitive non-ymmetric -deign with (r, λ = 1 and alternating ocle Shenglin Zhou, Yajie Wang School of Mathematic South China Univerity of Technology Guangzhou, Guangdong 510640, P. R. China lzhou@cut.edu.cn

More information

MATEMATIK Datum: Tid: eftermiddag. A.Heintz Telefonvakt: Anders Martinsson Tel.:

MATEMATIK Datum: Tid: eftermiddag. A.Heintz Telefonvakt: Anders Martinsson Tel.: MATEMATIK Datum: 20-08-25 Tid: eftermiddag GU, Chalmer Hjälpmedel: inga A.Heintz Telefonvakt: Ander Martinon Tel.: 073-07926. Löningar till tenta i ODE och matematik modellering, MMG5, MVE6. Define what

More information

PIPELINING AND PARALLEL PROCESSING. UNIT 4 Real time Signal Processing

PIPELINING AND PARALLEL PROCESSING. UNIT 4 Real time Signal Processing PIPELINING AND PARALLEL PROCESSING UNIT 4 Real time Signal Proceing Content Introduction Pipeling of FIR Digital Filter Parallel proceing Low power Deign FIR Digital Filter A FIR Filter i defined a follow:

More information

Fit a Spread Estimator in Small Memory

Fit a Spread Estimator in Small Memory Fit a Spread Etimator in Small Memory MyungKeun Yoon Tao Li Shigang Chen Jih-Kwon Peir Department of Computer & Information Science & Engineering Univerity of Florida, Gaineville, FL 36, USA Email: {myoon,

More information

Proactive Serving Decreases User Delay Exponentially: The Light-tailed Service Time Case

Proactive Serving Decreases User Delay Exponentially: The Light-tailed Service Time Case Proactive Serving Decreae Uer Delay Exponentially: The Light-tailed Service Time Cae Shaoquan Zhang, Longbo Huang, Minghua Chen, and Xin Liu Abtract In online ervice ytem, the delay experienced by uer

More information

Real-Time Identification of Sliding Friction Using LabVIEW FPGA

Real-Time Identification of Sliding Friction Using LabVIEW FPGA Real-Time Identification of Sliding Friction Uing LabVIEW FPGA M. Laine Mear, Jeannie S. Falcon, IEEE, and Thoma R. Kurfe, IEEE Abtract Friction i preent in all mechanical ytem, and can greatly affect

More information