Upper bounds for the forwarding indices of communication networks

The decision version of the forwarding index problem is, given a connected graph G and an integer ξ, to find a way of connecting each ordered pair of vertices by a path so that every vertex is an internal point of at most such paths. The optimization version of the problem is to find the smallest ξ for which a routing of this kind exists. Such a problem arises in the design of communication networks and distributed architectures. A model of parallel computation is represented by a network of processors, or machines processing and forwarding (synchronous) messages to each other, subject to physical constraints bearing on either the number of messages that can be processed by a single machine or the number of messages that can be sent through a connection. It was in this context that the problem was first introduced by Chung et al. (IEEE Trans. Inform. Theory 33 (1987) 224). The aim of this paper is to establish upper bounds for the optimal ξ as a function of the connectivity of the graph.     

Implementation of parallel branch-and-bound algorithms – experiences with the graph partitioning problem

Parallel processing is one of the essential concepts in the attempts to increase the computational power available for solving continuous and discrete optimization problems. In the case where an optimization algorithm is search-based, crucial issues of parallel distributed implementations are work-load distribution and granularity, i.e. how to distribute the search space among processors and how to control the amount of processing between interprocessor communication. The present paper compares distributed implementations of two branch-and-bound algorithms for the graph partitioning problem: Given an undirected graph with an even number of edges and weights assigned to each edge, partition the vertices into two subsets of equal size such that the sum of the costs of edges connecting vertices in different subsets is as small as possible. The problem is known to be NP-complete. The two branch-and-bound methods compared differ in design strategy: One is based on time-consuming bound calculations leading to tight bounds and thus a narrow search tree with few nodes, whereas the other employs an easy bound calculation scheme leading to a larger search tree. Both have been implemented on an iPSC-hypercube with 32 processors. We investigate the influence of the design strategy on the performance of the algorithms.     

Finding the needles in the haystack: efficient intelligence processing

As a result of communication technologies, the main intelligence challenge has shifted from collecting data to efficiently processing it so that relevant, and only relevant, information is passed on to intelligence analysts. We consider intelligence data intercepted on a social communication network. The social network includes both adversaries (eg terrorists) and benign participants. We propose a methodology for efficiently searching for relevant messages among the intercepted communications. Besides addressing a real and urgent problem that has attracted little attention in the open literature thus far, the main contributions of this paper are two-fold. First, we develop a novel knowledge accumulation model for intelligence processors, which addresses both the nodes of the social network (the participants) and its edges (the communications). Second, we propose efficient prioritization algorithms that utilize the processor’s accumulated knowledge. Our approach is based on methods from graphical models, social networks, random fields, Bayesian learning, and exploration/exploitation algorithms.     

Resource level minimization in the discrete–continuous scheduling

A discrete–continuous problem of non-preemptive task scheduling on identical parallel processors is considered. Tasks are described by means of a dynamic model, in which the speed of the task performance depends on the amount of a single continuously divisible renewable resource allotted to this task over time. An upper bound on the completion time of all the tasks is given. The criterion is to minimize the maximum resource consumption at each time instant, i.e., the resource level. This problem has been observed in many industrial applications, where a continuously divisible resource such as gas, fuel, electric, hydraulic or pneumatic power, etc., has to be distributed among the processing units over time, and it affects their productivity. The problem consists of two interrelated subproblems: task sequencing on processors (discrete subproblem) and resource allocation among the tasks (continuous subproblem). An optimal resource allocation algorithm for a given sequence of tasks is presented and computationally tested. Furthermore, approximation algorithms are proposed, and their theoretical and experimental worst-case performances are analyzed. Computer experiments confirmed the efficiency of all the algorithms.     

The Byzantine generals strike again

Can unanimity be achieved in an unreliable distributed system? This problem was named the “Byzantine Generals Problem” by L. Lamport, R. Shostak, and M. Pease (Technical Report 54, Computer Science Laboratory, SRI International, March 1980). The results obtained in the present paper prove that unanimity is achievable in any distributed system if and only if the number of faulty processors in the system is: (1) less than one-third of the total number of processors; and (2) less than one-half of the connectivity of the system's network. In cases where unanimity is achievable, algorithms for obtaining it are given. This result forms a complete characterization of networks in the light of the Byzantine Problem.     

LP-Based Heuristic Algorithms for Interconnecting Token Rings via Source Routing Bridges

We develop a method to determine the topology of a network that interconnects a number of token rings using source routing bridges. The purpose is to compute a topology that provides low response delays for network users at a minimal cost of bridge installations. We formulate this network design problem as a mixed binary integer linear program. We develop effective heuristic algorithms. The algorithms exploit the topology and routing solutions of the linear programming relaxation in a sophisticated manner which we believe is new in the literature. The model incorporates performance issues, such as network stability, bridge overflow, back pressure effect and broadcast storm, that are specific to the underlying communication technology. By formally incorporating these performance issues, we tighten the model formulation and improve the quality of the LP bound considerably. Computational results are reported for problems with up to 20 token rings and 190 potential bridge locations.     

Communication and time complexity of a distributed election protocol

we evaluate several variants of a standard election algorithm on a ring of processors. The performance measures of interest are the number of messages exchanged (communication complexity) and the execution time (time complexity). Classical models use a synchronism assumption according to which all processors start at the same time and message delays are constant.We attempt to capture the essential asynchronism of this class of algorithms by using probabilistic models. Two such models are discussed, one in discrete and one in continuous time. In each case, both the uni- and the bidirectional cases are studied and compared. We obtain expressions for the distributions of the number of exchanged messages and derive their asymptotic behavior whenn, the number of processors in the ring, grows large. The results show how the communication complexity actually depends on the speed of communcations on the ring, and what is the interest of having bidirectional communications.We also address in part the evaluation of the completion time of the algorithm. This time decomposes into astartup time and anexploration time. We show that the average of the startup time is of the order of logn.     

ATM network design for corporate networks

Asynchronous transfer mode (ATM) networks play an important role in support of distributed applications and modern corporate systems. The applications that use these architectures transfer high volumes of data in high-speed bursts and have stringent delay requirements. To address these needs, network managers seek cost efficient network solutions that provide network capacity sufficient to support high-speed applications.In this paper we present a new formulation and solution procedure for designing ATM networks to support corporate applications. Given the locations of the application servers and multiple clients we would like to design a minimum cost ATM network by which the clients can access the servers. Most of the previous work done in this area separates the problem into a routing problem (virtual circuit routing) and an assignment (virtual path assignment) problem. Thus, the solutions tend to be sub-optimal in the combined problem. This research optimizes virtual path (VP) assignment and virtual circuit (VC) routing simultaneously. We formulate the combined problem explicitly and develop an effective solution method. The solution method provides the designer with virtual paths and virtual circuits over which the actual communication takes place. Computational results are provided.     

Efficient diagonalization of oversized matrices on a distributed-memory multiprocessor

On parallel architectures, Jacobi methods for computing the singular value decomposition (SVD) and the symmetric eigenvalue decomposition (EVD) have been established as one of the most popular algorithms due to their excellent parallelism. Most of the Jacobi algorithms for distributed-memory architectures have been developed under the assumption that matrices can be distributed over the processors by square blocks of an even order or column blocks with an even number of columns. Obviously, there is a limit on the number of processors while we need to deal with problems of various sizes. We propose algorithms to diagonalize oversized matrices on a given distributed-memory multiprocessor with good load balancing and minimal message passing. Performances of the proposed algorithms vary greatly, depending on the relation between the problem size and the number of available processors. We give theoretical performance analyses which suggest the faster algorithm for a given problem size on a given distributed-memory multiprocessor. Finally, we present a new implementation for the convergence test of the algorithms on a distributed-memory multiprocessor and the implementation results of the algorithms on the NCUBE/seven hypercube architecture.This work was supported by National Science Foundation grant CCR-8813493. This work was partly done during the author's visit to the Mathematical Science Section, Engineering Physics and Mathematics Division, Oak Ridge National Laboratory, while participating in the Special Year on Numerical Linear Algebra, 1988, sponsored by the UTK Departments of Computer Science and Mathematics, and the ORNL Algebra sponsored by the UTK Departments of Computer Science and Mathematics, and the ORNL Mathematical Sciences Section, Engineering Physics and Mathematics Division.     

The Distributedk-Server Problem—A Competitive Distributed Translator fork-Server Algorithms

We consider thek-server problem23in a distributed setting. Given a network ofnprocessors andkidentical mobile servers, requests for service appear at the processors and a server must reach the request point. In addition to modeling problems in computer networks wherekidentical mobile resources are shared by the processors of the network, this models a realistic situation where the transfer of information is costly and there is no central control that governs the behavior of servers that move around to satisfy requests for service. We give a general translator to transform any deterministic global-control competitivek-server algorithm into a distributed competitive one. As consequences we get poly(k)-competitive distributed algorithms for the line, trees, and the ring. In contrast to the global-control case where there arek-server algorithms with competitive ratio that depends solely onk, we have a lower bound of Ω(max{k, (1/D) ·(log n/log log n)}) on the competitive ratio of any distributedk-server algorithm, where 1/Dis the ratio between the cost to transmit a message and the cost to move a server over the same distance. We also give a distributed version of the Harmonic randomizedk-server algorithm.     

Distributed Online Frequency Assignment in Cellular Networks

A cellular network is generally modeled as a subgraph of the triangular lattice. The distributed online frequency assignment problem can be abstracted as a multicoloring problem on a weighted graph, where the weight vector associated with the vertices models the number of calls to be served at the vertices and is assumed to change over time. In this paper, we develop a framework for studying distributed online frequency assignment in cellular networks. We present the first distributed online algorithms for this problem with proven bounds on their competitive ratios. We show a series of algorithms that use at each vertex information about increasingly larger neighborhoods of the vertex, and that achieve better competitive ratios. In contrast, we show lower bounds on the competitive ratios of some natural classes of online algorithms.     

Optimal Algorithms for Probabilistic Solitude Detection on Anonymous Rings

Probabilistic algorithms are developed for a basic problem in distributed computation, assuming anonymous, asynchronous, unidirectional rings of processors. The problem, known as Solitude Detection, requires that a nonempty subset of the processors, calledcontenders, determine whether or not there is exactly one contender. Monte Carlo algorithms are developed that err with probability bounded by a specified parameter and exhibit either message or processor termination. The algorithms transmit an optimal expected number of bits, to within a constant factor. Their bit complexities display a surprisingly rich dependence on the kind of termination exhibited and on the processors' knowledge of the size of the ring. Two probabilistic tools are isolated and then combined in various ways to achieve all our algorithms.     

On routing in VLSI design and communication networks

In this paper, we study the global routing problem in VLSI design and the multicast routing problem in communication networks. First we propose new and realistic models for both problems. In the global routing problem in VLSI design, we are given a lattice graph and subsets of the vertex set. The goal is to generate trees spanning these vertices in the subsets to minimize a linear combination of overall wirelength (edge length) and the number of bends of trees with respect to edge capacity constraints. In the multicast routing problem in communication networks, a graph is given to represent the network, together with subsets of the vertex set. We are required to find trees to span the given subsets and the overall edge length is minimized with respect to capacity constraints. Both problems are APX-hard. We present the integer linear programming (LP) formulation of both problems and solve the LP relaxations by the fast approximation algorithms for min-max resource-sharing problems in [K. Jansen, H. Zhang, Approximation algorithms for general packing problems and their application to the multicast congestion problem, Math. Programming, to appear, doi:10.1007/s10107-007-0106-8] (which is a generalization of the approximation algorithm proposed by Grigoriadis and Khachiyan [Coordination complexity of parallel price-directive decomposition, Math. Oper. Res. 2 (1996) 321-340]). For the global routing problem, we investigate the particular property of lattice graphs and propose a combinatorial technique to overcome the hardness due to the bend-dependent vertex cost. Finally, we develop asymptotic approximation algorithms for both problems with ratios depending on the best known approximation ratio for the minimum Steiner tree problem. They are the first known theoretical approximation bound results for the problems of minimizing the total costs (including both the edge and the bend costs) while spanning all given subsets of vertices.     

Processor interconnection networks from Cayley graphs

Cayley graphs of groups are presently being considered by the computer science community as models of architectures for large scale parallel processor computers. In the first section of this paper we discuss Cayley graphs and show how they may be used as a tool for the design and analysis of network architectures for these types of computers.

Observing that routing on a Cayley graph is equivalent to a certain factoring problem in the associated group, we have been able to use a known powerful factoring technique in computational group theory to produce a fast efficient routing algorithm on the associated Cayley graph. In the second section of this paper we present this work. This research can be regarded as a first attempt to find general purpose routing algorithms for interconnection networks.

Believing that average diameter of a network for a large scale MIMD machine is the predominant factor in determining network performance, we designed Cayley graphs to be used in a special study performed at the Supercomputing Research Center (SRC). The importance of the average diameter in determining network performance was supported by the fact that the graphs found by us had the smallest average diameter and outperformed all other graphs evaluated in the study. In fact, before being driven into saturation, one of our graphs sustained 9.4% more network traffic than the next best candidate, a butterfly architecture, and 74.3% better than the bench mark 2-d mesh. The last section of our paper is devoted to this work.

This paper is divided into three sections. In the first section we discuss Cayley graphs and show how they may be used as a tool for the design and analysis of network architectures for parallel computers. In the second section we present our research on the routing problem. This research can be regarded as a first attempt to find general purpose routing algorithms for interconnection networks. In the last section we present some evidence that average diameter of a network for a large scale MIMD machine is the predominant factor in determining network performance.     

On the Convergence to Nash Equilibrium in Problems of Distributed Computing

This paper studies two problems that arise in distributed computing. We deal with these problems from a game theoretical approach. We are interested in the convergence to the Nash equilibrium of algorithms based on the best reply strategy in a special case of linear costs. We present three specific types of algorithm that converge to the equilibrium. In our first model, composed of two processors, the convergence is established through monotonicity of the sequence of updates generated by each of the three algorithms. In the second model, made up of N processors, the convergence is due to the contraction of the algorithms.     

Bandwidth packing with priority classes

The bandwidth packing problem is defined as the selection and routing of messages from a given list of messages with prespecified requirements on demand for bandwidth. The messages have to be routed over a network with given topology so that the generated revenue is maximized. Messages to be routed are classified into two priority classes. An integer programming based formulation of this problem is proposed and a Lagrangean relaxation based methodology is described for solving this problem. A general purpose heuristic is then developed for generating feasible solutions of good quality. Several numerical experiments are conducted using a number of problem parameters such as number of messages, ratio of messages for lower and higher priority classes, capacity of links, and demand distribution of messages belonging to different classes and high quality solutions to the priority bandwidth packing problem are generated under the different situations.     

Asynchronous gradient algorithms for a class of convex separable network flow problems

We consider the single commodity strictly convex network flow problem. The dual of this problem is unconstrained, differentiable, and well suited for solution via distributed or parallel iterative methods. We present and prove convergence of gradient and asynchronous gradient algorithms for solving the dual problem. Computational results are given and analysed.     

On optimal management of resources in distributed networks

In a network of processors, a distributed operating system must handle the management of shared resources. In this paper, it is shown how to solve this problem in using the model previously introduced in [1]. This model (interconnection of N Markov chains each representing locally a distributed process) allows us to prove the good functioning properties for some distributed problems such as the mutual exclusion problem and the deadlock problem, We also prove that fairness is a basic notion for setting the model’s parameters and obtain an optimal working of the network.     

Optimal precision in the presence of uncertainty

We consider the problem of achieving coordinated actions in a real-time distributed system. In particular, we consider how closely (in terms of real time) processors can be guaranteed to perform a particular action, in a system where message transmission is guaranteed, but there is some uncertainty in message transmission time. We present an algorithm to achieve optimal precision in arbitrary networks. In networks where clocks run at the rate of real time, the optimal precision achievable in a network is exactly how tightly clocks can be guaranteed to be synchronized.     

Efficient sampling of random permutations

We show how to uniformly distribute data at random (not to be confounded with permutation routing) in two settings that are able to deal with massive data: coarse grained parallelism and external memory. In contrast to previously known work for parallel setups, our method is able to fulfill the three criteria of uniformity, work-optimality and balance among the processors simultaneously. To guarantee the uniformity we investigate the matrix of communication requests between the processors. We show that its distribution is a generalization of the multivariate hypergeometric distribution and we give algorithms to sample it efficiently in the two settings.