C. Siva Ram Murthy

Task allocation algorithms for maximizing reliability of distributed computing systems

We consider the problem of finding an optimal and suboptimal task allocation (i.e., to which processor should each module of a task or program be assigned) in distributed computing systems with the goal of maximizing the system reliability (i.e., the probability that the system can run the entire task successfully). The problem of finding an optimal task allocation is known to be NP-hard in the strong sense. We present an algorithm for this problem, which uses the idea of branch and bound with underestimates for reducing the computations in finding an optimal task allocation. The algorithm reorders the list of modules to allow a subset of modules that do not communicate with one another to be assigned last, for further reduction in the computations of optimal task allocation for maximizing reliability. We also present a heuristic algorithm which obtains suboptimal task allocations in a reasonable amount of computational time. We study the performance of the algorithms over a wide range of parameters such as the number of modules, the number of processors, the ratio of average execution cost to average communication cost, and the connectivity of modules. We demonstrate the effectiveness of our algorithms by comparing them with recent competing task allocation algorithms for maximizing reliability available in the literature.

Scheduling precedence constrained task graphs with non-negligible intertask communication onto multiprocessors

The multiprocessor scheduling problem is the problem of scheduling the tasks of a precedence constrained task graph (representing a parallel program) onto the processors of a multiprocessor in a way that minimizes the completion time. Since this problem is known to be NP-hard in the strong sense in all but a few very restricted eases, heuristic algorithms are being developed which obtain near optimal schedules in a reasonable amount of computation time. We present an efficient heuristic algorithm for scheduling precedence constrained task graphs with nonnegligible intertask communication onto multiprocessors taking contention in the communication channels into consideration. Our algorithm for obtaining satisfactory suboptimal schedules is based on the classical list scheduling strategy. It simultaneously exploits the schedule-holes generated in the processors and in the communication channels during the scheduling process in order to produce better schedules. We demonstrate the effectiveness of our algorithm by comparing with two competing heuristic algorithms available in the literature. >

Preferred link based delay-constrained least-cost routing in wide area networks

Multimedia applications involving digital audio and/or digital video transmissions require strict QoS constraints (end-to-end delay bound, bandwidth availability, packet loss rate, etc.) to be met by the network. To guarantee the real-time delivery of packets satisfying these constraints, a real-time channel (D. Ferrari and D.C. Verma, A scheme for real-time channel establishment in wide-area networks. IEEE JSAC, 8(3), 368-379, 1990) needs to be established before the transmission of packets of a connection can begin. The establishment of such channels requires the development of efficient route selection algorithms that are designed to take into account the QoS constraints. The general problem of determining a least-cost delay-constrained route in a given communication network has been proved to be NP-hard (M.R. Garey and D.S. Johnson, Computers and Intractability: a guide to the theory of NP-completeness, W.H. Freeman, 1979). In this paper, we describe a preferred link approach to distributed delay-constrained least-cost routing in order to establish real-time channels. The approach attempts to combine the benefits of probing and backtracking based algorithms (better adaptiveness and wider search) with the advantages of distance-vector type algorithms (lower setup time). The scheme is flexible in that a variety of heuristics can be employed to order the neighbouring links of any given node. Three heuristics are proposed and their performance is studied through simulation experiments. The simulation results indicate that the proposed heuristics provide better performance than other preferred neighbour methods, in terms of increased call acceptance rate and lower average route cost. The heuristics are also shown to adapt much better to dynamic variations in network and link characteristics.

Algorithms for delay-constrained low-cost multicast tree construction

With the proliferation of multimedia group applications, the construction of multicast trees satisfying quality of service (QoS) requirements is becoming a problem of prime importance. Multicast groups are usually classified as sparse or pervasive groups depending on the physical distribution of group members. They are also classified based on the temporal characteristics of group membership into static and dynamic groups. In this paper, we propose two algorithms for constructing multicast trees for multimedia group communication in which the members are sparse and static. The proposed algorithms use a constrained distributed unicast routing algorithm for generating low-cost, bandwidth and delay constrained multicast trees. These algorithms have lower message complexity and call setup time due to their nature of iteratively adding paths, rather than edges, to partially constructed trees. We study the performance (in terms of call acceptance rate, call setup time and multicast tree cost) of these algorithms through simulation by comparing them with that of a recently proposed algorithm (V. Kompella, J.C. Pasquale, G.C. Polyzos, Two distributed algorithms for the constrained Steiner tree problem, in: Proc. Comp. Comm. Networking, San Diego, CA, June 1993) for the same problem. The simulation results indicate that the proposed algorithms provide larger call acceptance rates, lower setup times and comparable tree costs.

Algorithms for reliability-oriented module allocation in distributed computing systems

Abstract We consider the problem of finding an allocation of program modules onto processors of a distributed computing system such that the reliability of successfully executing these modules is maximized. The distributed system consists of a number of processors interconnected by means of communication links. Certain constraints such as storage and load limits may be present at each processor. At any point of time, each component of the distributed system (processor or communication link) can exist in either of two states — operational or failed. The probability of a component being operational is given. A program module can be executed on any one of a set of processors. For execution, it requires access to certain data files. If a particular file it requires is not available locally, it has to access the file remotely, and for the remote access to be possible, at least one path (sequence of links and processors) from the processor at which the program module is executing, to one of the processors where the required file is available, must be operational. To improve reliability, there may be multiple copies of certain files, dispersed at various processors. Our aim is to allocate the program modules to processors in a manner that maximizes the probability of it being able to successfully access all the files it requires for execution, and the allocation should not violate any of the constraints. This problem is known to be NP-hard. We use a state space search technique — the A ∗ algorithm to obtain an optimal allocation. We also present a heuristic algorithm which obtains sub-optimal allocations in a reasonable amount of computation time. Through simulations over a wide range of parameters, we demonstrate the effectiveness of our approach.

New Algorithms for Resource Reclaiming from Precedence Constrained Tasks in Multiprocessor Real-Time Systems

The scheduling of tasks in multiprocessor real-time systems has attracted many researchers in the recent past. Tasks in these systems have deadlines to be met, and most of the real-time scheduling algorithms use worst case computation times to schedule these tasks. Many resources will be left unused if the tasks are dispatched purely based on the schedule produced by these scheduling algorithms, since most of the tasks will take less time to execute than their respective worst case computation times. Resource reclaiming refers to the problem of reclaiming the resources left unused by a real-time task when it takes less time to execute than its worst case computation time. In this paper, we propose two algorithms to reclaim these resources from real-time tasks that are constrained by precedence relations and resource requirements, in shared memory multiprocessor systems. We introduce a notion called a restriction vector for each task which captures its resource and precedence constraints with other tasks. This will help not only in the efficient implementation of the algorithms, but also in obtaining an improvement in performance over the reclaiming algorithms proposed in earlier work 2]. We compare our resource reclaiming algorithms with the earlier algorithms and, by experimental studies, show that they reclaim more resources, thereby increasing the guarantee ratio (the ratio of the number of tasks guaranteed to meet their deadlines to the number of tasks that have arrived), which is the basic requirement of any resource reclaiming algorithm. From our simulation studies, we demonstrate that complex reclaiming algorithms with high reclaiming overheads do not lead to an improvement in the guarantee ratio.

Imposed Route Reuse in Ad Hoc Network Routing Protocols Using Structured Peer-to-Peer Overlay Routing

The use of wireless in local loop (WiLL) has generated considerable interest due to the advantages it offers such as ease and low cost of deployment and maintenance. With an increase in the number of subscribers in the network, it becomes expedient to employ spectrum reusability techniques such as the use of multihop relaying in order to improve the capacity of the wireless systems. Throughput enhanced wireless in local loop (TWiLL) is one such architecture that employs multihop relaying and shortcut relaying to reuse bandwidth in WiLL systems. Compared to other multihop wireless network architectures, TWiLL architecture assumes significance due to its potential use in fixed wireless broadband services such as LMDS (local multipoint distribution service) and MMDS (multichannel multipoint distribution system). Analysis of the call acceptance ratio (CAR) in multihop wireless architectures including TWiLL is nontrivial as the Erlang B formula no longer holds. In this paper, we build multidimensional Markov chains to analyze the performance of multihop wireless systems such as TWiLL that has multiple types of channels. We also compare the results of our analysis with results from simulations. We observe that multihop relaying and shortcut relaying lead to a significant increase in the CAR of WiLL systems. Also, the free space propagation model that is normally used to model the radio channel is a very unrealistic model and does not consider reflection, diffraction, scattering, and multipath propagation that hinder transmissions in WiLL systems. In this paper, we studied the effect of several realistic radio channel propagation models on the performance of the TWiLL system through analysis and simulations

Efficient mapping of backpropagation algorithm onto a network of workstations

In this paper, we present an efficient technique for mapping a backpropagation (BP) learning algorithm for multilayered neural networks onto a network of workstations (NOWs). We adopt a vertical partitioning scheme, where each layer in the neural network is divided into p disjoint partitions, and map each partition onto an independent workstation in a network of p workstations. We present a fully distributed version of the BP algorithm and also its speedup analysis. We compare the performance of our algorithm with a recent work involving the vertical partitioning approach for mapping the BP algorithm onto a distributed memory multiprocessor. Our results on SUN 3/50 NOWs show that we are able to achieve better speedups by using only two communication sets and also by avoiding some redundancy in the weights computation for one training cycle of the algorithm.

An Improved Algorithm for Module Allocation in Distributed Computing Systems

We consider the problem of finding an optimal and sub-optimal allocation of program modules onto processors of a distributed computing system. A module causes two types of cost to be incurred at the processor to which it is allocated?an execution cost for processing the module, and a communication cost if the module communicates with other modules which are not allocated to the same processor. The distributed computing system is heterogeneous, that is, both costs vary from processor to processor. Certain constraints, such as storage and load constraints, may be present at each processor. Our aim is to allocate the modules to the processors in an optimal manner, that is, the sum of execution and communication costs over all processors should be the minimum possible, without violating any of the constraints. It is an NP-hard problem and we use a state space search technique?theA* algorithm to obtain an optimal allocation. We propose a method to reduce the number of nodes generated in the search tree. The distributed computing system model that we have considered here is the same as that considered by Chernet al. (Inform. Process. Lett.32(2) (July 1989), 61?71). Through simulations over a wide range of parameters, we have compared our method with that of in Chernet al. We also present a heuristic algorithm which obtains sub-optimal allocations in a reasonable amount of computation time.

A genetic algorithm for the knowledge base partitioning problem

Allocation of production rules among several partitions in a knowledge base can shorten the compilation and execution time of expert systems applications and facilitate their verification, validation and maintenance. The problem of allocation of rules or groups of rules (areas) among partitions of limited size such that the sum of inter-partition connections is minimized is termed as the knowledge base partitioning problem. This problem is known to be NP-complete. In this paper, we first present a genetic algorithm for solving the knowledge base partitioning problem and then compare its performance with that of a recent heuristic clustering algorithm for the same problem available in the literature. The results of our study clearly demonstrate that our genetic algorithm produces better quality solutions than those obtained by the clustering algorithm.