Arnaud Legrand, Olivier Beaumont, Loris Marchal, Yves Robert

Size: px
Start display at page:

Download "Arnaud Legrand, Olivier Beaumont, Loris Marchal, Yves Robert"

Transcription

1 Arnaud Legrand, Olivier Beaumont, Loris Marchal, Yves Robert École Nor 46 Allée d I Télé Téléc Adresse é

2 when using a general gr comes from the decompo are used concurrently to that a concrete scheduli asymptotically optimal, solutions). Nous nous intéressons ic d une application comple grille de calcul. De tel de données à travers le le débit d une telle diffu nombre de messages do pour le parallélisme de d modélisée par un graphe ont des vitesses différent avec des topologies restr gés), et montrons que ce Nous montrons commen construire un ordonnanc Keywords: Scheduling Mots-clés: Ordonnanceme

3 For the first two problems, th For the third problem, the obje steady-state operation, i.e. the a In the case of the atomic br the source) would receive the m implemented using a spanning t complex: the idea is to use severa fractions of the total message. A packets, which are sent in a pipe for an illustration with two-dime The series of broadcasts prob Desprez et al. [24], but with a sources successively broadcast o communications. Here, we assum this is closer to a master-slave pa slaves in a pipelined fashion, for in problem instances. The series of broadcasts rese the latter using an algorithm for of the atomic messages (packets) to the pipelined broadcast proble function of the total message len In this paper, we re-visit the in the context of heterogeneous broadcasting with processors com capacities, and/or different start

4 in Section 7. Finally, we state so The target architectural platform (V,E,c), as illustrated in Figure paths. Let p = V be the number role: it initially holds all the dat are destination nodes which mus There are several scenarios for this paper, we concentrate on the receive data from one of its neigh any given time-step, there are at in emission and the other in rece Each edge e j,k : P j P k is communicate one unit-size mess the pipelined broadcast problem) reverse direction, from P k to P j, communication link between P j and P k are neighbors in the comm precisely: if P j sends a unit-size 2 Framework P k cannot initiate another r a send operation), P j cannot initiate another

5 Series of broadcasts In the se (potentially infinite) sequence of of the link capacities c j,k. The o to maximize the throughput, i.e We work out a little example in Pipelined broadcast In the p a large message of total size L. T The time to send a packet of s start-up costs in the definition o optimization problem, which we to find the number and size of th so that the total execution time 3 Comparing topolog We start this section with an exam We compare the best throughpu (DAG), or the full topology wit broadcast tree is a NP-hard prob 3.1 Working out an exam Optimal solution Consider the simple example of t can be achieved on this network

6 at t = 1, and message m 2 faster (c 1,3 = 1/2), one tim every odd-numbered time-s P 2 operates in a similar fash to P 4. Overall, the steady-state is by the source processor eve to 1. Period 1 Time 1 2 Link P s P 1 m 1 P s P 2 m P 1 P 2 m P 1 P 3 P 2 P 1 P 2 P 4 Figure 2: An op We further use the example and of DAGs over spanning trees

7 and P 2 are children of the source In the first case, where P 1 and tree of Figure 3. The labels on represent the (average) number unit. The value 1/2 means that o because of the one-port constrain can be achieved using this tree. In the second case, one of t Without loss of generality, we ass leads to the spanning tree of Fig on one hand, the one-port constr transfer a message to its childre every 3 time-steps. We can inde the figure. Overall, this is the be this network. We choose a less restrictive assum instead of a broadcast tree, out entry vertex, namely the source tree? The answer is positive. The spanning trees: the DAG shown (P 1,P 2 ) is replaced by the edge of Figure 5. Because the first br achieve a throughput at least 1/ Broadcast DAGs

8 As a conclusion, we point ou strongly depends upon the graph example shows, restricting to tre of 2 3 ), and restricting to DAGs is of 1 instead of 4 5 ). It turns out that computing t when restricting to DAGs than w fore, we give the solution for DA general graphs (Section 5). Befor result: we show that finding the b for the Series problem is NP-Co 3.2 Finding the best broa The problem of determining th follows: Definition 1 (BEST-BROAD G = (G,V,c) representing a netw throughput TP. The associated decision prob Definition 2 (BEST-BROAD G = (V,E,c) representing a ne broadcast tree T such that TP Theorem 1. BEST-BROADCA

9 TREE, which is known to be NP TRAINT-SPANNING-TREE is integer K V d, does there exist than K? We construct the follow we let G = (V,E,c) where V = V the un-oriented edge (j,k) is in source at random, i.e. any node all edges, except for the edges o the bound D = 1. Clearly, the in of I 1. We show that I 2 admits a Assume first that I 1 admits We orient T d so as to trans T is a broadcast tree in G, all nodes in T have an outat most K (because s has n its children in T is j V,(P s,p j ) E T 1 The time spent by any oth j V,(P i,p j ) E T 1 c i,j = Therefore TP D, and I 2

10 j V,(P s,p j ) E T TP where δ + (i) is number of c δ + (s) K. Similarly for from i has weight 1/(K 1 degree of i is at most K (o arborescence T rooted at s than or equal to K, hence 4 Series of broadcast In this section we provide an algo optimization problem, in the rest 4.1 Linear program for D In this section, we assume the n that all nodes are reachable from n j,k is the (fractional) num P k during one time-unit, t j,k is the fraction of time time-unit.

11 knowns): SSBPDAG(G), Steady-S Maximize TP, subject to P j, P k t j,k = n j, P j, P k 0 t j,k P j, P k,(p j,p P j, P j with j s, P k,(p k,p P k,(p k,p We make a digression to pro optimal solution of the Series p 4.2 The edge coloring lem Given a set of processors operat set of communications within a p is that Equations 4 and 5 are sat P j, P j, P k,(p j P k,(p k However, it is not obvious that t because only independent comm scheduled simultaneously. Fortun

12 Proof. Because G is a DAG, we each vertex j, such that h(j) < each node is reachable from s an H be the maximal height of any We define q as the least comm TP) in the solution of the linea broadcast of q TP messages f schedule can be pipelined to reac We build a schedule by indu messages at time q h(i) and th We state the following result, wh can be found in polynomial time Theorem 2. The solution of the to the Series problem on a DAG ber of broadcasts that can be init the corresponding optimal period The source already has the Consider a node P j of heig than or equal to h 1 have to Equation (6), in the solu P j

13 constraints are satisfied for Each sender P k will have t a node P j receives during t period [(h(p j ) 1) q,h(p can be performed in time q P l,(p k,p l ) E and h( Since the one-port constrai sult presented above, we ca these communications in ti We now prove that this sche done by pipelining the schedule constraints are still satisfied, dur the time spent by node P j r P k,(p j,p k ) E

14 Link P P P P P Once pipelined, it gives the f Period 0 Link P s P 1 [0,3 P s P 2 P 1 P 2 P 1 P 3 P 2 P 4 The last step is to use the receptions or emissions never ove Period 1 Time Link P s P P s P 2 P 1 P 2 P 1 P 3 P 2 P 4

15 size of the initial data. The sketc In what follows, we give a set o solution at steady-state. We norm to each processor every T timethis is the dual problem of Section broadcast per time-unit. Howeve of messages that transit along ed But things get more complicated For any node P j, we denote such that (P j,p k ) E; similarly P k such that (P k,p j ) E. Since we deal with broadcas But because of the pipelining, sev 1. We express the conditions t to the Series problem by m provides a lower bound for 2. From the solution of the li used to broadcast the diffe enables us to reach the low 3. From the set of trees, we to write the code of the br initial data. 5.2 Lower bound

16 This constraint reads: for number of messages destin the number of same type m in steady-state operation, i Link occupation The following sages that are transferred of messages that transit on for each i, the fraction x j,k i main difficulty is that the m be partly the same, since t fore, the constraint n j,k = be too pessimistic. Since o consider that all the messag set, namely the largest one occupation time t j,k of the We also need to write down incoming and outgoing com be the time spent by P j f

17 Steady-State Broa Minimize T, subject to i, i s, j, P j P s and P (P j,p k ) E, (P j,p k ) E, j, j, j,k, j, j, 5.3 Weighted broadcast t The solution of the linear prog needed to broadcast one unit-siz be achieved, because of the assu edge are all sub-sets of the large is possible to find a set of broadc Branching theorem. Unfortunat exponential in the problem size. Branching theorem, that produc

18 Therefore, by the Max-flow, Min G between P s and P i is at least disconnect P s and P i. Since the a Lemma 3. κ(g,p 0 ) N Proof. Suppose that κ(g,p 0 ) = and Fulkerson, for each P i, ther denote the value of this flow y j,k i construction), and let us denote between P s and P i. Then, i, i, j, P j P s and P j (P j,p k ) E, (P j,p k ) E, j, j, Therefore, there would exist a N N T < T, which is a contradic Finally we have the following Theorem 4. κ(g,p 0 ) = N.

19 where χ T j,k (T l) = 1 if (P j,p k ) Moreover, the trees can be found m Lemma 4. l λ l = κ(g,p 0 ) = Proof. The proof of this lemma formation of G into a multi-grap Lemma 5. The set of trees can Proof. The number of trees is b be encoded in size of order V ( and maxn j,k, can be encoded in above. Therefore, the weighted versi time a set of weighted trees, wh use these trees in order to broad overall number of messages that Moreover, since the overall w platform, we have:

20 V = V out j,k where p = V is the number of and Pk in is weighted by the qua overall amount of data transitin the communications, we use the chapter 20]): Theorem 6. Let G M = (V,E,m ings M 1,...,M km, with integer w where χ M j,k (M i) = 1 if (P j,p k µ i = m Moreover, the matchings can be i Lemma 6. i µ i NT. Proof. By Equation(17), m j,k m j,k c j,k j and m j,k c j,k k

21 P b P e (c) Graph of the Nn j,k = max i P s P a P c P d P b P e (e) Second broadcast tree, λ Figure 6: Example where m j,k < message is 5 time-units, due to by the least common multiple N on each edge. Figures 6(d) and 6 figure, each of them with a weigh trees. On the edge (P c,p e ), we tree, so m c,e = 1, whereas Nn c,e through this edge in the optimal made when we use trees.

22 by (19), and by ( Let us denote by In the following, we exhibit NsT, and Ns messages are bro throughput 1/T. Let m l j (q) be the set of messa the q-th period. The sketch of the Step i (P j,p k ), i M (j,k) l, (P j,p k ) T l end end We prove the correctness of t

23 M(r Total number of messages re sages are sent along the ed different, and are different overall number of messages Therefore, during one period new different messages. Therefo one period is 1 T, hence its optim sum of the duration of the differe as claimed in the proof of Lemm 5.5 Asymptotic optimalit In this section, we prove that the cally, no scheduling algorithm (e a given time-frame than ours, up to the formal statement of this r

24 P j, P k,(p k,p j ) E P i, P j P i,p s, T k,j K ( P k,(p j,p k ) forwarded by P j to P i ) P i,p i P s,opt(g,k) = the source to each node) P i,p i P s,opt(g,k) = node) Let x i,j k = Nj,k i opt(g,k), n j,k = m P k N out (P j ) t j,k. Let T(K) = op All the equations of the linear pr value. This concludes the proof. Again, this lemma states tha There remains to bound the lo periodic solution, to come up w state operation. Consider the fol Solve the linear program fo the period T such that eve

25 Clean-up during the J last final destination. No proce be evenly divisible by T). The number of messages b steady(g,k) = (r + 1) T Clearly, the initialization and tation, using parallel routing and do not need to refine the previou Theorem 7. The previous sched totically optimal: Proof. Using the previous lemm algorithm, we have steady(g, K) result because I, J, and TP(G) 6 Pipelined broadcas In the pipelined broadcast proble of total size L, which can be spl model must include start-up ove linear in the packet size, the bes

26 T pipe (L) β Proof. Let w = max j,k j,k c j,k so tha units to broadcast 1 TP T unit-s SSB(G), and T the integer period messages of length ν +w. As sho and to quit the steady-state mo time-units to broadcast ν messag Conversely, Lemma 7 shows into account, so that we have a fo 7 Experiments In this section, we work out a random generator of topology [6 topology is represented on Figur Figure 7(b) shows the result represent communications, and t in its list, it means that Nx i,j k = through edge (i,j) to reach P k. 152 time-units. From these communications, Figure 8, where both the logical

27 2 2 Figure 7 2 2(5) 2(6)

28 the top of tree. Some theoretica have been developed for the prob A more complex model is int message into account, but also t time needed to receive the messa proportional to the length of the Yet another model of commu the message between any process T i,j and a part depending on the m the two processors, B i,j. Since th total communication time betwe proposed for the broadcast and t All previous models assume t a given processor can send dat overlapping this operation with o Other collective communicati have been studied in the context others. 9 Conclusion In this paper, we have studied Our major objective was to max mode, when a large number of sa single large message is split into p the best throughput may well re

29 (a) First (b) Secon logical tree

30 [2] M. Banikazemi, J. Sampathk nication modeling of heterog zation of collective operation pages IEEE Comp [3] D. Bertsimas and D. Gamar ing and packet routing. Jou [4] P.B. Bhat, C.S. Raghavendr for distributed heterogeneou 59(2): , [5] P.B. Bhat, C.S. Raghavend in distributed heterogeneou Distributed Computing Syste [6] Kenneth L. Calvert, Matt topology. IEEE Communic [7] T. H. Cormen, C. E. Leisers Press, [8] M. R. Garey and D. S. John NP-Completeness. W. H. Fr [9] N.G. Hall, W.-P. Liu, and 32(14): , 1998.

31 [17] P. Liu and T.-H. Sheng. B systems. In SPAA 2000, 1 Architectures, pages [23] M. Snir, S. W. Otto, S. H complete reference. The MIT [18] P. Liu and D.-W. Wang. Re In 14th International Para IEEE Computer Society Pre [19] B. Lowekamp and A. Begue on heterogeneous networks. Symposium (IPDPS 96). IE [20] J.A. Moore and M.J. Quinn gray codes. IEEE Trans. Pa [21] F. Ooshita, S. Matsumae, an cluster systems. In Proceed mance Computing Systems a [22] A. Schrijver. Combinatorial rithms and Combinatorics. [24] Y-C. Tseng, S-Y. Wang, and efficient pipelining and overl 61, 1995.

32

Scheduling divisible loads with return messages on heterogeneous master-worker platforms

Scheduling divisible loads with return messages on heterogeneous master-worker platforms Scheduling divisible loads with return messages on heterogeneous master-worker platforms Olivier Beaumont, Loris Marchal, Yves Robert Laboratoire de l Informatique du Parallélisme École Normale Supérieure

More information

Steady-state scheduling of task graphs on heterogeneous computing platforms

Steady-state scheduling of task graphs on heterogeneous computing platforms Laboratoire de l Informatique du Parallélisme École Normale Supérieure de Lyon Unité Mixte de Recherche CNRS-INRIA-ENS LYON n o 5668 Steady-state scheduling of task graphs on heterogeneous computing platforms

More information

Scheduling divisible loads with return messages on heterogeneous master-worker platforms

Scheduling divisible loads with return messages on heterogeneous master-worker platforms Scheduling divisible loads with return messages on heterogeneous master-worker platforms Olivier Beaumont 1, Loris Marchal 2, and Yves Robert 2 1 LaBRI, UMR CNRS 5800, Bordeaux, France Olivier.Beaumont@labri.fr

More information

Overlay networks maximizing throughput

Overlay networks maximizing throughput Overlay networks maximizing throughput Olivier Beaumont, Lionel Eyraud-Dubois, Shailesh Kumar Agrawal Cepage team, LaBRI, Bordeaux, France IPDPS April 20, 2010 Outline Introduction 1 Introduction 2 Complexity

More information

Assessing the impact and limits of steady-state scheduling for mixed task and data parallelism on heterogeneous platforms

Assessing the impact and limits of steady-state scheduling for mixed task and data parallelism on heterogeneous platforms Laboratoire de l Informatique du Parallélisme École Normale Supérieure de Lyon Unité Mixte de Recherche CNRS-INRIA-ENS LYON-UCBL n o 5668 Assessing the impact and limits of steady-state scheduling for

More information

Divisible Load Scheduling

Divisible Load Scheduling Divisible Load Scheduling Henri Casanova 1,2 1 Associate Professor Department of Information and Computer Science University of Hawai i at Manoa, U.S.A. 2 Visiting Associate Professor National Institute

More information

SPI. Laboratoire de l Informatique du Parallélisme

SPI. Laboratoire de l Informatique du Parallélisme Laboratoire de l Informatique du Parallélisme Ecole Normale Superieure de Lyon Unite de recherche associee au CNRS n o 198 SPI More on Scheduling Block-Cyclic Array Redistribution 1 Frederic Desprez,Stephane

More information

FIFO scheduling of divisible loads with return messages under the one-port model

FIFO scheduling of divisible loads with return messages under the one-port model INSTITUT NATIONAL DE RECHERCHE EN INFORMATIQUE ET EN AUTOMATIQUE FIFO scheduling of divisible loads with return messages under the one-port model Olivier Beaumont Loris Marchal Veronika Rehn Yves Robert

More information

On the Complexity of Mapping Pipelined Filtering Services on Heterogeneous Platforms

On the Complexity of Mapping Pipelined Filtering Services on Heterogeneous Platforms On the Complexity of Mapping Pipelined Filtering Services on Heterogeneous Platforms Anne Benoit, Fanny Dufossé and Yves Robert LIP, École Normale Supérieure de Lyon, France {Anne.Benoit Fanny.Dufosse

More information

Scheduling multiple bags of tasks on heterogeneous master-worker platforms: centralized versus distributed solutions

Scheduling multiple bags of tasks on heterogeneous master-worker platforms: centralized versus distributed solutions Laboratoire de l Informatique du Parallélisme École Normale Supérieure de Lyon Unité Mixte de Recherche CNRS-INRIA-ENS LYON-UCBL n o 5668 Scheduling multiple bags of tasks on heterogeneous master-worker

More information

IN this paper, we deal with the master-slave paradigm on a

IN this paper, we deal with the master-slave paradigm on a IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS, VOL. 15, NO. 4, APRIL 2004 1 Scheduling Strategies for Master-Slave Tasking on Heterogeneous Processor Platforms Cyril Banino, Olivier Beaumont, Larry

More information

Lecture 14 - P v.s. NP 1

Lecture 14 - P v.s. NP 1 CME 305: Discrete Mathematics and Algorithms Instructor: Professor Aaron Sidford (sidford@stanford.edu) February 27, 2018 Lecture 14 - P v.s. NP 1 In this lecture we start Unit 3 on NP-hardness and approximation

More information

CS264: Beyond Worst-Case Analysis Lecture #11: LP Decoding

CS264: Beyond Worst-Case Analysis Lecture #11: LP Decoding CS264: Beyond Worst-Case Analysis Lecture #11: LP Decoding Tim Roughgarden October 29, 2014 1 Preamble This lecture covers our final subtopic within the exact and approximate recovery part of the course.

More information

Network Flows. CTU FEE Department of control engineering. March 28, 2017

Network Flows. CTU FEE Department of control engineering. March 28, 2017 Network Flows Zdeněk Hanzálek, Přemysl Šůcha hanzalek@fel.cvut.cz CTU FEE Department of control engineering March 28, 2017 Z. Hanzálek (CTU FEE) Network Flows March 28, 2017 1 / 44 Table of contents 1

More information

APTAS for Bin Packing

APTAS for Bin Packing APTAS for Bin Packing Bin Packing has an asymptotic PTAS (APTAS) [de la Vega and Leuker, 1980] For every fixed ε > 0 algorithm outputs a solution of size (1+ε)OPT + 1 in time polynomial in n APTAS for

More information

- Well-characterized problems, min-max relations, approximate certificates. - LP problems in the standard form, primal and dual linear programs

- Well-characterized problems, min-max relations, approximate certificates. - LP problems in the standard form, primal and dual linear programs LP-Duality ( Approximation Algorithms by V. Vazirani, Chapter 12) - Well-characterized problems, min-max relations, approximate certificates - LP problems in the standard form, primal and dual linear programs

More information

Equitable and semi-equitable coloring of cubic graphs and its application in batch scheduling

Equitable and semi-equitable coloring of cubic graphs and its application in batch scheduling Equitable and semi-equitable coloring of cubic graphs and its application in batch scheduling Hanna Furmańczyk, Marek Kubale Abstract In the paper we consider the problems of equitable and semi-equitable

More information

The partial inverse minimum cut problem with L 1 -norm is strongly NP-hard. Elisabeth Gassner

The partial inverse minimum cut problem with L 1 -norm is strongly NP-hard. Elisabeth Gassner FoSP Algorithmen & mathematische Modellierung FoSP Forschungsschwerpunkt Algorithmen und mathematische Modellierung The partial inverse minimum cut problem with L 1 -norm is strongly NP-hard Elisabeth

More information

Computational Complexity

Computational Complexity Computational Complexity Algorithm performance and difficulty of problems So far we have seen problems admitting fast algorithms flow problems, shortest path, spanning tree... and other problems for which

More information

Computing the throughput of replicated workflows on heterogeneous platforms

Computing the throughput of replicated workflows on heterogeneous platforms Laboratoire de l Informatique du Parallélisme École Normale Supérieure de Lyon Unité Mixte de Recherche CNRS-INRIA-ENS LYON-UCBL n o 5668 Computing the throughput of replicated workflows on heterogeneous

More information

Duality of LPs and Applications

Duality of LPs and Applications Lecture 6 Duality of LPs and Applications Last lecture we introduced duality of linear programs. We saw how to form duals, and proved both the weak and strong duality theorems. In this lecture we will

More information

Representations of All Solutions of Boolean Programming Problems

Representations of All Solutions of Boolean Programming Problems Representations of All Solutions of Boolean Programming Problems Utz-Uwe Haus and Carla Michini Institute for Operations Research Department of Mathematics ETH Zurich Rämistr. 101, 8092 Zürich, Switzerland

More information

Elisabeth Gassner Introduction THE PARTIAL INVERSE MINIMUM CUT PROBLEM WITH L 1 -NORM IS STRONGLY NP-HARD

Elisabeth Gassner Introduction THE PARTIAL INVERSE MINIMUM CUT PROBLEM WITH L 1 -NORM IS STRONGLY NP-HARD RAIRO-Oper. Res. 44 (010) 41 49 DOI: 10.1051/ro/010017 RAIRO Operations Research www.rairo-ro.org THE PARTIAL INVERSE MINIMUM CUT PROBLEM WITH L 1 -NORM IS STRONGLY NP-HARD Elisabeth Gassner 1 Abstract.

More information

Mathematical Programs Linear Program (LP)

Mathematical Programs Linear Program (LP) Mathematical Programs Linear Program (LP) Integer Program (IP) Can be efficiently solved e.g., by Ellipsoid Method Cannot be efficiently solved Cannot be efficiently solved assuming P NP Combinatorial

More information

Selected partial inverse combinatorial optimization problems with forbidden elements. Elisabeth Gassner

Selected partial inverse combinatorial optimization problems with forbidden elements. Elisabeth Gassner FoSP Algorithmen & mathematische Modellierung FoSP Forschungsschwerpunkt Algorithmen und mathematische Modellierung Selected partial inverse combinatorial optimization problems with forbidden elements

More information

MAXIMAL VERTEX-CONNECTIVITY OF S n,k

MAXIMAL VERTEX-CONNECTIVITY OF S n,k MAXIMAL VERTEX-CONNECTIVITY OF S n,k EDDIE CHENG, WILLIAM A. LINDSEY, AND DANIEL E. STEFFY Abstract. The class of star graphs is a popular topology for interconnection networks. However it has certain

More information

Scheduling multiple bags of tasks on heterogeneous master- worker platforms: centralized versus distributed solutions

Scheduling multiple bags of tasks on heterogeneous master- worker platforms: centralized versus distributed solutions Scheduling multiple bags of tasks on heterogeneous master- worker platforms: centralized versus distributed solutions Olivier Beaumont, Larry Carter, Jeanne Ferrante, Arnaud Legrand, Loris Marchal, Yves

More information

Communicating the sum of sources in a 3-sources/3-terminals network

Communicating the sum of sources in a 3-sources/3-terminals network Communicating the sum of sources in a 3-sources/3-terminals network Michael Langberg Computer Science Division Open University of Israel Raanana 43107, Israel Email: mikel@openu.ac.il Aditya Ramamoorthy

More information

Robust Network Codes for Unicast Connections: A Case Study

Robust Network Codes for Unicast Connections: A Case Study Robust Network Codes for Unicast Connections: A Case Study Salim Y. El Rouayheb, Alex Sprintson, and Costas Georghiades Department of Electrical and Computer Engineering Texas A&M University College Station,

More information

THE COMMUNICATIONS COMPLEXITY HIERARCHY IN DISTRIBUTED COMPUTING. J. B. Sidney + and J. Urrutia*

THE COMMUNICATIONS COMPLEXITY HIERARCHY IN DISTRIBUTED COMPUTING. J. B. Sidney + and J. Urrutia* THE COMMUNICATIONS COMPLEXITY HIERARCHY IN DISTRIBUTED COMPUTING J. B. Sidney + and J. Urrutia* 1. INTRODUCTION Since the pioneering research of Cook [1] and Karp [3] in computational complexity, an enormous

More information

The impact of heterogeneity on master-slave on-line scheduling

The impact of heterogeneity on master-slave on-line scheduling Laboratoire de l Informatique du Parallélisme École Normale Supérieure de Lyon Unité Mixte de Recherche CNRS-INRIA-ENS LYON-UCBL n o 5668 The impact of heterogeneity on master-slave on-line scheduling

More information

Non-preemptive multiprocessor scheduling of strict periodic systems with precedence constraints

Non-preemptive multiprocessor scheduling of strict periodic systems with precedence constraints Non-preemptive multiprocessor scheduling of strict periodic systems with precedence constraints Liliana Cucu, Yves Sorel INRIA Rocquencourt, BP 105-78153 Le Chesnay Cedex, France liliana.cucu@inria.fr,

More information

Discrete Optimization 2010 Lecture 1 Introduction / Algorithms & Spanning Trees

Discrete Optimization 2010 Lecture 1 Introduction / Algorithms & Spanning Trees Discrete Optimization 2010 Lecture 1 Introduction / Algorithms & Spanning Trees Marc Uetz University of Twente m.uetz@utwente.nl Lecture 1: sheet 1 / 43 Marc Uetz Discrete Optimization Outline 1 Introduction

More information

Divisible load theory

Divisible load theory Divisible load theory Loris Marchal November 5, 2012 1 The context Context of the study Scientific computing : large needs in computation or storage resources Need to use systems with several processors

More information

Geometric Steiner Trees

Geometric Steiner Trees Geometric Steiner Trees From the book: Optimal Interconnection Trees in the Plane By Marcus Brazil and Martin Zachariasen Part 3: Computational Complexity and the Steiner Tree Problem Marcus Brazil 2015

More information

SCHEDULING STRATEGIES FOR MIXED DATA AND TASK PARALLELISM ON HETEROGENEOUS CLUSTERS

SCHEDULING STRATEGIES FOR MIXED DATA AND TASK PARALLELISM ON HETEROGENEOUS CLUSTERS Parallel Processing Letters, Vol. 3, No. 2 (2003) 225 244 c World Scientific Publishing Company SCHEDULING STRATEGIES FOR MIXED DATA AND TASK PARALLELISM ON HETEROGENEOUS CLUSTERS O. BEAUMONT LaBRI, UMR

More information

Lecture 8 Network Optimization Algorithms

Lecture 8 Network Optimization Algorithms Advanced Algorithms Floriano Zini Free University of Bozen-Bolzano Faculty of Computer Science Academic Year 2013-2014 Lecture 8 Network Optimization Algorithms 1 21/01/14 Introduction Network models have

More information

Multi-criteria approximation schemes for the resource constrained shortest path problem

Multi-criteria approximation schemes for the resource constrained shortest path problem Noname manuscript No. (will be inserted by the editor) Multi-criteria approximation schemes for the resource constrained shortest path problem Markó Horváth Tamás Kis Received: date / Accepted: date Abstract

More information

Scheduling unit processing time arc shutdown jobs to maximize network flow over time: complexity results

Scheduling unit processing time arc shutdown jobs to maximize network flow over time: complexity results Scheduling unit processing time arc shutdown jobs to maximize network flow over time: complexity results Natashia Boland Thomas Kalinowski Reena Kapoor Simranjit Kaur Abstract We study the problem of scheduling

More information

Branch-and-Bound for the Travelling Salesman Problem

Branch-and-Bound for the Travelling Salesman Problem Branch-and-Bound for the Travelling Salesman Problem Leo Liberti LIX, École Polytechnique, F-91128 Palaiseau, France Email:liberti@lix.polytechnique.fr March 15, 2011 Contents 1 The setting 1 1.1 Graphs...............................................

More information

Flows and Cuts. 1 Concepts. CS 787: Advanced Algorithms. Instructor: Dieter van Melkebeek

Flows and Cuts. 1 Concepts. CS 787: Advanced Algorithms. Instructor: Dieter van Melkebeek CS 787: Advanced Algorithms Flows and Cuts Instructor: Dieter van Melkebeek This lecture covers the construction of optimal flows and cuts in networks, their relationship, and some applications. It paves

More information

How to deal with uncertainties and dynamicity?

How to deal with uncertainties and dynamicity? How to deal with uncertainties and dynamicity? http://graal.ens-lyon.fr/ lmarchal/scheduling/ 19 novembre 2012 1/ 37 Outline 1 Sensitivity and Robustness 2 Analyzing the sensitivity : the case of Backfilling

More information

Degradable Agreement in the Presence of. Byzantine Faults. Nitin H. Vaidya. Technical Report #

Degradable Agreement in the Presence of. Byzantine Faults. Nitin H. Vaidya. Technical Report # Degradable Agreement in the Presence of Byzantine Faults Nitin H. Vaidya Technical Report # 92-020 Abstract Consider a system consisting of a sender that wants to send a value to certain receivers. Byzantine

More information

Resource Constrained Project Scheduling Linear and Integer Programming (1)

Resource Constrained Project Scheduling Linear and Integer Programming (1) DM204, 2010 SCHEDULING, TIMETABLING AND ROUTING Lecture 3 Resource Constrained Project Linear and Integer Programming (1) Marco Chiarandini Department of Mathematics & Computer Science University of Southern

More information

P P P NP-Hard: L is NP-hard if for all L NP, L L. Thus, if we could solve L in polynomial. Cook's Theorem and Reductions

P P P NP-Hard: L is NP-hard if for all L NP, L L. Thus, if we could solve L in polynomial. Cook's Theorem and Reductions Summary of the previous lecture Recall that we mentioned the following topics: P: is the set of decision problems (or languages) that are solvable in polynomial time. NP: is the set of decision problems

More information

10 Max-Flow Min-Cut Flows and Capacitated Graphs 10 MAX-FLOW MIN-CUT

10 Max-Flow Min-Cut Flows and Capacitated Graphs 10 MAX-FLOW MIN-CUT 10 Max-Flow Min-Cut 10.1 Flows and Capacitated Graphs Previously, we considered weighted graphs. In this setting, it was natural to thinking about minimizing the weight of a given path. In fact, we considered

More information

The Maximum Flow Problem with Disjunctive Constraints

The Maximum Flow Problem with Disjunctive Constraints The Maximum Flow Problem with Disjunctive Constraints Ulrich Pferschy Joachim Schauer Abstract We study the maximum flow problem subject to binary disjunctive constraints in a directed graph: A negative

More information

Chapter 34: NP-Completeness

Chapter 34: NP-Completeness Graph Algorithms - Spring 2011 Set 17. Lecturer: Huilan Chang Reference: Cormen, Leiserson, Rivest, and Stein, Introduction to Algorithms, 2nd Edition, The MIT Press. Chapter 34: NP-Completeness 2. Polynomial-time

More information

Design and Analysis of Algorithms

Design and Analysis of Algorithms CSE 0, Winter 08 Design and Analysis of Algorithms Lecture 8: Consolidation # (DP, Greed, NP-C, Flow) Class URL: http://vlsicad.ucsd.edu/courses/cse0-w8/ Followup on IGO, Annealing Iterative Global Optimization

More information

Parallel Performance Theory - 1

Parallel Performance Theory - 1 Parallel Performance Theory - 1 Parallel Computing CIS 410/510 Department of Computer and Information Science Outline q Performance scalability q Analytical performance measures q Amdahl s law and Gustafson-Barsis

More information

Minimal Delay Traffic Grooming for WDM Optical Star Networks

Minimal Delay Traffic Grooming for WDM Optical Star Networks Minimal Delay Traffic Grooming for WDM Optical Star Networks Hongsik Choi, Nikhil Grag, and Hyeong-Ah Choi Abstract All-optical networks face the challenge of reducing slower opto-electronic conversions

More information

1.1 P, NP, and NP-complete

1.1 P, NP, and NP-complete CSC5160: Combinatorial Optimization and Approximation Algorithms Topic: Introduction to NP-complete Problems Date: 11/01/2008 Lecturer: Lap Chi Lau Scribe: Jerry Jilin Le This lecture gives a general introduction

More information

MAS210 Graph Theory Exercises 5 Solutions (1) v 5 (1)

MAS210 Graph Theory Exercises 5 Solutions (1) v 5 (1) MAS210 Graph Theor Exercises 5 Solutions Q1 Consider the following directed network N. x 3 (3) v 1 2 (2) v 2 5 (2) 2(2) 1 (0) 3 (0) 2 (0) 3 (0) 3 2 (2) 2(0) v v 5 1 v 6 The numbers in brackets define an

More information

Week Cuts, Branch & Bound, and Lagrangean Relaxation

Week Cuts, Branch & Bound, and Lagrangean Relaxation Week 11 1 Integer Linear Programming This week we will discuss solution methods for solving integer linear programming problems. I will skip the part on complexity theory, Section 11.8, although this is

More information

NP-complete problems. CSE 101: Design and Analysis of Algorithms Lecture 20

NP-complete problems. CSE 101: Design and Analysis of Algorithms Lecture 20 NP-complete problems CSE 101: Design and Analysis of Algorithms Lecture 20 CSE 101: Design and analysis of algorithms NP-complete problems Reading: Chapter 8 Homework 7 is due today, 11:59 PM Tomorrow

More information

BBM402-Lecture 20: LP Duality

BBM402-Lecture 20: LP Duality BBM402-Lecture 20: LP Duality Lecturer: Lale Özkahya Resources for the presentation: https://courses.engr.illinois.edu/cs473/fa2016/lectures.html An easy LP? which is compact form for max cx subject to

More information

1 The independent set problem

1 The independent set problem ORF 523 Lecture 11 Spring 2016, Princeton University Instructor: A.A. Ahmadi Scribe: G. Hall Tuesday, March 29, 2016 When in doubt on the accuracy of these notes, please cross chec with the instructor

More information

The Algorithmic Aspects of the Regularity Lemma

The Algorithmic Aspects of the Regularity Lemma The Algorithmic Aspects of the Regularity Lemma N. Alon R. A. Duke H. Lefmann V. Rödl R. Yuster Abstract The Regularity Lemma of Szemerédi is a result that asserts that every graph can be partitioned in

More information

Chapter 3 Deterministic planning

Chapter 3 Deterministic planning Chapter 3 Deterministic planning In this chapter we describe a number of algorithms for solving the historically most important and most basic type of planning problem. Two rather strong simplifying assumptions

More information

Fault-Tolerant Consensus

Fault-Tolerant Consensus Fault-Tolerant Consensus CS556 - Panagiota Fatourou 1 Assumptions Consensus Denote by f the maximum number of processes that may fail. We call the system f-resilient Description of the Problem Each process

More information

cs/ee/ids 143 Communication Networks

cs/ee/ids 143 Communication Networks cs/ee/ids 143 Communication Networks Chapter 5 Routing Text: Walrand & Parakh, 2010 Steven Low CMS, EE, Caltech Warning These notes are not self-contained, probably not understandable, unless you also

More information

A comparison of sequencing formulations in a constraint generation procedure for avionics scheduling

A comparison of sequencing formulations in a constraint generation procedure for avionics scheduling A comparison of sequencing formulations in a constraint generation procedure for avionics scheduling Department of Mathematics, Linköping University Jessika Boberg LiTH-MAT-EX 2017/18 SE Credits: Level:

More information

CS264: Beyond Worst-Case Analysis Lecture #18: Smoothed Complexity and Pseudopolynomial-Time Algorithms

CS264: Beyond Worst-Case Analysis Lecture #18: Smoothed Complexity and Pseudopolynomial-Time Algorithms CS264: Beyond Worst-Case Analysis Lecture #18: Smoothed Complexity and Pseudopolynomial-Time Algorithms Tim Roughgarden March 9, 2017 1 Preamble Our first lecture on smoothed analysis sought a better theoretical

More information

Data redistribution algorithms for heterogeneous processor rings

Data redistribution algorithms for heterogeneous processor rings Data redistribution algorithms for heterogeneous processor rings Hélène Renard, Yves Robert, Frédéric Vivien To cite this version: Hélène Renard, Yves Robert, Frédéric Vivien. Data redistribution algorithms

More information

Technische Universität München, Zentrum Mathematik Lehrstuhl für Angewandte Geometrie und Diskrete Mathematik. Combinatorial Optimization (MA 4502)

Technische Universität München, Zentrum Mathematik Lehrstuhl für Angewandte Geometrie und Diskrete Mathematik. Combinatorial Optimization (MA 4502) Technische Universität München, Zentrum Mathematik Lehrstuhl für Angewandte Geometrie und Diskrete Mathematik Combinatorial Optimization (MA 4502) Dr. Michael Ritter Problem Sheet 1 Homework Problems Exercise

More information

Travelling Salesman Problem

Travelling Salesman Problem Travelling Salesman Problem Fabio Furini November 10th, 2014 Travelling Salesman Problem 1 Outline 1 Traveling Salesman Problem Separation Travelling Salesman Problem 2 (Asymmetric) Traveling Salesman

More information

Input-queued switches: Scheduling algorithms for a crossbar switch. EE 384X Packet Switch Architectures 1

Input-queued switches: Scheduling algorithms for a crossbar switch. EE 384X Packet Switch Architectures 1 Input-queued switches: Scheduling algorithms for a crossbar switch EE 84X Packet Switch Architectures Overview Today s lecture - the input-buffered switch architecture - the head-of-line blocking phenomenon

More information

Lecture 4 : Introduction to Low-density Parity-check Codes

Lecture 4 : Introduction to Low-density Parity-check Codes Lecture 4 : Introduction to Low-density Parity-check Codes LDPC codes are a class of linear block codes with implementable decoders, which provide near-capacity performance. History: 1. LDPC codes were

More information

Combinatorial optimization problems

Combinatorial optimization problems Combinatorial optimization problems Heuristic Algorithms Giovanni Righini University of Milan Department of Computer Science (Crema) Optimization In general an optimization problem can be formulated as:

More information

A Polynomial-Time Algorithm for Pliable Index Coding

A Polynomial-Time Algorithm for Pliable Index Coding 1 A Polynomial-Time Algorithm for Pliable Index Coding Linqi Song and Christina Fragouli arxiv:1610.06845v [cs.it] 9 Aug 017 Abstract In pliable index coding, we consider a server with m messages and n

More information

On the Min-Max-Delay Problem: NP-completeness, Algorithm, and Integrality Gap

On the Min-Max-Delay Problem: NP-completeness, Algorithm, and Integrality Gap On the Min-Max-Delay Problem: NP-completeness, Algorithm, and Integrality Gap Qingyu Liu, Lei Deng, Haibo Zeng, Minghua Chen Department of Electrical and Computer Engineering, Virginia Tech, USA Department

More information

In complexity theory, algorithms and problems are classified by the growth order of computation time as a function of instance size.

In complexity theory, algorithms and problems are classified by the growth order of computation time as a function of instance size. 10 2.2. CLASSES OF COMPUTATIONAL COMPLEXITY An optimization problem is defined as a class of similar problems with different input parameters. Each individual case with fixed parameter values is called

More information

NOTES ON LINEAR ODES

NOTES ON LINEAR ODES NOTES ON LINEAR ODES JONATHAN LUK We can now use all the discussions we had on linear algebra to study linear ODEs Most of this material appears in the textbook in 21, 22, 23, 26 As always, this is a preliminary

More information

Preliminaries. Graphs. E : set of edges (arcs) (Undirected) Graph : (i, j) = (j, i) (edges) V = {1, 2, 3, 4, 5}, E = {(1, 3), (3, 2), (2, 4)}

Preliminaries. Graphs. E : set of edges (arcs) (Undirected) Graph : (i, j) = (j, i) (edges) V = {1, 2, 3, 4, 5}, E = {(1, 3), (3, 2), (2, 4)} Preliminaries Graphs G = (V, E), V : set of vertices E : set of edges (arcs) (Undirected) Graph : (i, j) = (j, i) (edges) 1 2 3 5 4 V = {1, 2, 3, 4, 5}, E = {(1, 3), (3, 2), (2, 4)} 1 Directed Graph (Digraph)

More information

Lecture 5 January 16, 2013

Lecture 5 January 16, 2013 UBC CPSC 536N: Sparse Approximations Winter 2013 Prof. Nick Harvey Lecture 5 January 16, 2013 Scribe: Samira Samadi 1 Combinatorial IPs 1.1 Mathematical programs { min c Linear Program (LP): T x s.t. a

More information

Index coding with side information

Index coding with side information Index coding with side information Ehsan Ebrahimi Targhi University of Tartu Abstract. The Index Coding problem has attracted a considerable amount of attention in the recent years. The problem is motivated

More information

Graph coloring, perfect graphs

Graph coloring, perfect graphs Lecture 5 (05.04.2013) Graph coloring, perfect graphs Scribe: Tomasz Kociumaka Lecturer: Marcin Pilipczuk 1 Introduction to graph coloring Definition 1. Let G be a simple undirected graph and k a positive

More information

Chordal networks of polynomial ideals

Chordal networks of polynomial ideals Chordal networks of polynomial ideals Diego Cifuentes Laboratory for Information and Decision Systems Electrical Engineering and Computer Science Massachusetts Institute of Technology Joint work with Pablo

More information

Multiprocessor Scheduling of Age Constraint Processes

Multiprocessor Scheduling of Age Constraint Processes Multiprocessor Scheduling of Age Constraint Processes Lars Lundberg Department of Computer Science, University of Karlskrona/Ronneby, Soft Center, S-372 25 Ronneby, Sweden, email: Lars.Lundberg@ide.hk-r.se

More information

We say that a flow is feasible for G (or just feasible if the graph and capacity function in question are obvious) if

We say that a flow is feasible for G (or just feasible if the graph and capacity function in question are obvious) if CS787: Advanced Algorithms Lecture 4: Network Flow We devote this lecture to the network flow problem and the max-flow min-cut theorem. A number of other problems can be cast into a maximum flow or minimum

More information

A Short Proof of the VPN Tree Routing Conjecture on Ring Networks

A Short Proof of the VPN Tree Routing Conjecture on Ring Networks A Short Proof of the VPN Tree Routing Conjecture on Ring Networs Fabrizio Grandoni Voler Kaibel Gianpaolo Oriolo Martin Sutella January 15, 008 Abstract Only recently, Hurens, Keijsper, and Stougie proved

More information

Algorithms Exam TIN093 /DIT602

Algorithms Exam TIN093 /DIT602 Algorithms Exam TIN093 /DIT602 Course: Algorithms Course code: TIN 093, TIN 092 (CTH), DIT 602 (GU) Date, time: 21st October 2017, 14:00 18:00 Building: SBM Responsible teacher: Peter Damaschke, Tel. 5405

More information

FRACTIONAL PACKING OF T-JOINS. 1. Introduction

FRACTIONAL PACKING OF T-JOINS. 1. Introduction FRACTIONAL PACKING OF T-JOINS FRANCISCO BARAHONA Abstract Given a graph with nonnegative capacities on its edges, it is well known that the capacity of a minimum T -cut is equal to the value of a maximum

More information

NP-Completeness. NP-Completeness 1

NP-Completeness. NP-Completeness 1 NP-Completeness Reference: Computers and Intractability: A Guide to the Theory of NP-Completeness by Garey and Johnson, W.H. Freeman and Company, 1979. NP-Completeness 1 General Problems, Input Size and

More information

Maximum Flow Problem (Ford and Fulkerson, 1956)

Maximum Flow Problem (Ford and Fulkerson, 1956) Maximum Flow Problem (Ford and Fulkerson, 196) In this problem we find the maximum flow possible in a directed connected network with arc capacities. There is unlimited quantity available in the given

More information

Midterm Exam 2 Solutions

Midterm Exam 2 Solutions Algorithm Design and Analysis November 12, 2010 Pennsylvania State University CSE 565, Fall 2010 Professor Adam Smith Exam 2 Solutions Problem 1 (Miscellaneous). Midterm Exam 2 Solutions (a) Your friend

More information

Network Algorithms and Complexity (NTUA-MPLA) Reliable Broadcast. Aris Pagourtzis, Giorgos Panagiotakos, Dimitris Sakavalas

Network Algorithms and Complexity (NTUA-MPLA) Reliable Broadcast. Aris Pagourtzis, Giorgos Panagiotakos, Dimitris Sakavalas Network Algorithms and Complexity (NTUA-MPLA) Reliable Broadcast Aris Pagourtzis, Giorgos Panagiotakos, Dimitris Sakavalas Slides are partially based on the joint work of Christos Litsas, Aris Pagourtzis,

More information

PRAM lower bounds. 1 Overview. 2 Definitions. 3 Monotone Circuit Value Problem

PRAM lower bounds. 1 Overview. 2 Definitions. 3 Monotone Circuit Value Problem U.C. Berkeley CS273: Parallel and Distributed Theory PRAM lower bounds. Professor Satish Rao October 16, 2006 Lecturer: Satish Rao Last revised Scribe so far: Satish Rao cribbing from previous years lectures

More information

Isomorphism of free G-subflows (preliminary draft)

Isomorphism of free G-subflows (preliminary draft) Isomorphism of free G-subflows (preliminary draft) John D. Clemens August 3, 21 Abstract We show that for G a countable group which is not locally finite, the isomorphism of free G-subflows is bi-reducible

More information

Lecture 10: Mechanisms, Complexity, and Approximation

Lecture 10: Mechanisms, Complexity, and Approximation CS94 P9: Topics Algorithmic Game Theory November 8, 011 Lecture 10: Mechanisms, Complexity, and Approximation Lecturer: Christos Papadimitriou Scribe: Faraz Tavakoli It is possible to characterize a Mechanism

More information

Lecture 24 : Even more reductions

Lecture 24 : Even more reductions COMPSCI 330: Design and Analysis of Algorithms December 5, 2017 Lecture 24 : Even more reductions Lecturer: Yu Cheng Scribe: Will Wang 1 Overview Last two lectures, we showed the technique of reduction

More information

Causality and Time. The Happens-Before Relation

Causality and Time. The Happens-Before Relation Causality and Time The Happens-Before Relation Because executions are sequences of events, they induce a total order on all the events It is possible that two events by different processors do not influence

More information

A New Variation of Hat Guessing Games

A New Variation of Hat Guessing Games A New Variation of Hat Guessing Games Tengyu Ma 1, Xiaoming Sun 1, and Huacheng Yu 1 Institute for Theoretical Computer Science Tsinghua University, Beijing, China Abstract. Several variations of hat guessing

More information

On the intersection of infinite matroids

On the intersection of infinite matroids On the intersection of infinite matroids Elad Aigner-Horev Johannes Carmesin Jan-Oliver Fröhlich University of Hamburg 9 July 2012 Abstract We show that the infinite matroid intersection conjecture of

More information

Section 6 Fault-Tolerant Consensus

Section 6 Fault-Tolerant Consensus Section 6 Fault-Tolerant Consensus CS586 - Panagiota Fatourou 1 Description of the Problem Consensus Each process starts with an individual input from a particular value set V. Processes may fail by crashing.

More information

directed weighted graphs as flow networks the Ford-Fulkerson algorithm termination and running time

directed weighted graphs as flow networks the Ford-Fulkerson algorithm termination and running time Network Flow 1 The Maximum-Flow Problem directed weighted graphs as flow networks the Ford-Fulkerson algorithm termination and running time 2 Maximum Flows and Minimum Cuts flows and cuts max flow equals

More information

/633 Introduction to Algorithms Lecturer: Michael Dinitz Topic: Dynamic Programming II Date: 10/12/17

/633 Introduction to Algorithms Lecturer: Michael Dinitz Topic: Dynamic Programming II Date: 10/12/17 601.433/633 Introduction to Algorithms Lecturer: Michael Dinitz Topic: Dynamic Programming II Date: 10/12/17 12.1 Introduction Today we re going to do a couple more examples of dynamic programming. While

More information

Formal Verification of Mobile Network Protocols

Formal Verification of Mobile Network Protocols Dipartimento di Informatica, Università di Pisa, Italy milazzo@di.unipi.it Pisa April 26, 2005 Introduction Modelling Systems Specifications Examples Algorithms Introduction Design validation ensuring

More information

The min cost flow problem Course notes for Optimization Spring 2007

The min cost flow problem Course notes for Optimization Spring 2007 The min cost flow problem Course notes for Optimization Spring 2007 Peter Bro Miltersen February 7, 2007 Version 3.0 1 Definition of the min cost flow problem We shall consider a generalization of the

More information