OpenMP is a popular programming environment for high-level shared-memory parallel processing in C, C++ and Fortran on multiple platforms. Using OpenMP, application developers can focus on the code they want to paralle...
详细信息
ISBN:
(纸本)9780889867048
OpenMP is a popular programming environment for high-level shared-memory parallel processing in C, C++ and Fortran on multiple platforms. Using OpenMP, application developers can focus on the code they want to parallelize for better performance without having to do all the work in code parallelization. While hiding much of the complexity in parallel processing, it imposes a great challenge for the developers to make sure the run-time behavior of the code is as expected. In this paper, we explore potential risks with using OpenMP through reduction, a frequently performed operation in parallel processing.
This paper addresses the workflow rescheduling problem in the Grid where resource performance is fluctuating and hard to predict. Unlike existing rescheduling approaches targeting the same problem, the proposed a mobi...
详细信息
ISBN:
(纸本)9780889867048
This paper addresses the workflow rescheduling problem in the Grid where resource performance is fluctuating and hard to predict. Unlike existing rescheduling approaches targeting the same problem, the proposed a mobile agent based distributed approach. Once a workflow is submitted to a central workflow scheduler, an initial schedule for tasks in that workflow will be created first. Then, at the workflow run-time, mobile agents which carry those tasks will reschedule themselves by a rescheduling algorithm when resource performance changes.
Grid computing inevitably evolves to an infrastructure that satisfies various kinds of requirements, thus demanding more complex resource management. The scheduler introduced here is designed to serve a local resource...
详细信息
ISBN:
(纸本)088986568X
Grid computing inevitably evolves to an infrastructure that satisfies various kinds of requirements, thus demanding more complex resource management. The scheduler introduced here is designed to serve a local resource domain in complex resource management. It uses MCT (Minimum Completion Time) for the immediate mode scheduling and an improved GA (Genetic Algorithm) for the batch mode scheduling to obtain the highest possible throughput and utilization. We created a simulation and compared our scheduler with three other benchmark algorithms. The experimental results indicate that our scheduler can achieve higher throughput and utilization than the benchmark algorithms subject to varying task flow, except when the arrival rate of tasks is very low. The dynamic scheduling cycle and batch size make the scheduler more adaptive for task flow varying, and decrease the number of tasks missing deadlines when resources fluctuate. The improved genetic algorithm is the key to obtain better results in a busy task flow.
In the last few years one of the most interesting questions in IT world is that of convergence between data and voice networks. This procedure cuts down overall expenses of implementation, usage and support of a telep...
详细信息
In the last few years one of the most interesting questions in IT world is that of convergence between data and voice networks. This procedure cuts down overall expenses of implementation, usage and support of a telephone network, as the expenses of other telephone. Beside the procedures that are prerequisite for the realization of this concept (e.g. ensuring the quality of the transferred voice), it is necessary to ensure the security of that traffic from various threats «inherited» from data traffic. This paper is about preventive protection of voice traffic transferred through a converged data/voice network.
Efficient communication in distributed systems is essential for optimal system performance. As parallel computation develops, many different kinds of messages need to be sent between machines. These messages may have ...
详细信息
ISBN:
(纸本)088986392X
Efficient communication in distributed systems is essential for optimal system performance. As parallel computation develops, many different kinds of messages need to be sent between machines. These messages may have different requirements for network latency and bandwidth. In this paper, we revisit combining two physical interfaces into one virtual network (channel bonding). In contrast with most existing implementations, we communicate via heterogenous interfaces, and explore a number of more sophisticated scheduling algorithms than the round-robin technique used in the standard Beowulf channel-bonding driver. We explore the relative tradeoffs of interface scheduling overhead versus the additional bandwidth possible with multiple interfaces. Using both Gigabit Ethernet and Fast Ethernet networks, we evaluate system performance using netPerf. We show significant performance gains for small messages, with incremental improvements for larger packets.
A stabilizing system guarantees that, regardless of the current configuration, the system reaches a legal configuration in a bounded number of steps and the system configuration remains legal thereafter. Whereas, a st...
详细信息
ISBN:
(纸本)9780889867741
A stabilizing system guarantees that, regardless of the current configuration, the system reaches a legal configuration in a bounded number of steps and the system configuration remains legal thereafter. Whereas, a stabilizing system that maintains no explicit variables in the processes of the system is referred to as an inherently stabilizing system, and hence all system states are legal by construction. Due to this attribute, inherently stabilizing systems are immune to transient faults and do not experience any delay due arbitrary system initialization. We view a fault that perturbs the system configuration but not the program as transient fault. Due to these features, inherently stabilizing distributed protocols for peer-to-peer, sensor and mobile networks are desirable. Hypercube, star networks and their variations that provide an increased degree of scalability have been initially design for parallelnetworks. However, their scalability and the presence of multiple disjoint paths in these topologies make them viable alternatives to existing peer-to-peer and sensor networks topologies. In this paper, we proposed an inherently stabilizing algorithm for delivering messages over all node-disjoint paths from a process to another in star networks. The proposed algorithm has numerous applications including VLSI layout, reliable networks routing, secure message transmission, and network survivability. The proposed routing algorithm is optimal with respect to its state space and lengths of the node-disjoint paths.
In this paper we present and evaluate the performance of two different strategies for the deployment of parallel multifrontal and multiple frontal sparse linear solvers in the context of a parallel finite element code...
详细信息
ISBN:
(纸本)0889865701
In this paper we present and evaluate the performance of two different strategies for the deployment of parallel multifrontal and multiple frontal sparse linear solvers in the context of a parallel finite element code. Direct sparse linear solvers are based on sophisticated reorganisation of the standard Gaussian elimination algorithm with the aim of exploring matrix sparsity and reducing the amount of fill-in. Such codes can be successfully applied to very large linear systems, and are especially effective when a sparse linear system needs to be solved for multiple right-hand sides. Unfortunately, many important applications, such as finite element solutions of non-linear, transient problems, require repeated factorisation of the coefficient matrix. In such cases the only way of achieving good performance is parallelisation of both the computation of the finite element matrices and the linear system solution phase. We have developed two different designs for deployment of parallel multifrontal and multiple frontal sparse linear solvers in this context, each deploying three different strategies for the assembly of the global data. These designs are suitable for parallel and heterogeneous architectures. Experiments confirm high efficiency, low communication cost, and reduced initial memory requirements of our deployment designs, compared to a standard deployment strategy.
In this paper, we present the improving capability of accuracy and the parallel efficiency of self-organizing neural groves (SONGs) for classification on a MIMD parallel computer. Self-generating neural networks (SGNN...
详细信息
In this paper, we present the improving capability of accuracy and the parallel efficiency of self-organizing neural groves (SONGs) for classification on a MIMD parallel computer. Self-generating neural networks (SGNNs) are originally proposed on adopting to classification or clustering by automatically constructing self-generating neural tree (SGNT) from given training data. The SONG is composed of plural SGNTs each of which is independently generated by shuffling the order of the given training data, and the output of the SONG is voted all outputs of the SGNTs. We allocate each of SGNTs to each of processors in the MIMD parallel computer. Experimental results show that the more the number of processors increases, the more the classification accuracy increases for all problems.
This paper proposes two distributed algorithms for the heuristic solution of the Steiner Tree Problem in Network (SPN). The problem has a practical application in the construction of a minimum cost distribution tree f...
详细信息
ISBN:
(纸本)088986392X
This paper proposes two distributed algorithms for the heuristic solution of the Steiner Tree Problem in Network (SPN). The problem has a practical application in the construction of a minimum cost distribution tree for multicast transmission. Multicast transmission represents a necessary lower network service for the wide diffusion of new multimedia network applications. Currently, given the lack of efficient distributed methods, the existing protocols build the multicast distribution tree using some selected central node. The proposed distributed algorithms allow the construction of effective distribution trees using a coordination protocol among the network nodes. The algorithms have been implemented and tested both in simulation and on experimental active networks, and their performance values are presented.
A distributed object systems is said to be K-tolerant if every object is available after the simultaneous failure of up to K nodes. The problem is that a K-tolerant system, after failures, is no longer K-tolerant;that...
详细信息
ISBN:
(纸本)088986568X
A distributed object systems is said to be K-tolerant if every object is available after the simultaneous failure of up to K nodes. The problem is that a K-tolerant system, after failures, is no longer K-tolerant;that is, subsequent failures may compromise the availability of the objects. A continuously K-tolerant system is one which starting from a K-tolerant configuration, after the failure of up to K nodes, reconfigures itself so to remain K-tolerant. The existing protocols for maintaining continuous K-tolerance do so without regard to the resulting structure of the available data. For example if the distributed set of objects was sorted, this ordering would be most likely lost after restructuring. Analogously, a balanced distribution of the objects among the nodes might also be not achieved in the new distribution after reorganization. In this paper, we present a mechanism for maintaining continuous K-tolerance while keeping the load balanced and the objects sorted. The proposed solution uses minimum amount of replication and has a cost comparable to the one of the known unstructured solutions.
暂无评论