Showing papers presented at "Parallel and Distributed Processing Techniques and Applications in 2005"

Parallel SOR Iterative Algorithms and Performance Evaluation On a Linux Cluster

Abstract: : The successive over-relaxation (SOR) iterative method is an important solver for linear systems. In this paper, a parallel algorithm for the red-black SOR method with domain decomposition is investigated. The parallel SOR algorithm is designed by combining the traditional red-black SOR and row block domain decomposition technique, which reduces the communication cost and simplifies the parallel implementation. Two other iterative methods, Jacobi and Gauss-Seidel (G-S), ate also implemented in parallel for comparison. The three parallel iterative algorithms are implemented in C and MPI (Message Passing Interface) for solving the Dirichlet problem on a Linux cluster with eight dual processor 2.6ghz 32 bit Intel Xeons, totaling 16 processors. The performances of the three algorithms are evaluated in terms of speedup and efficiency.

Open Benchmarks for Load Balancing Heuristics in Parallel Adaptive Finite Element Computations.

Abstract: Load balancing plays an important role in large parallel numerical simulations. To address this problem, some general purpose libraries as well as a number of more specific approaches have been developed, and results on their performance and solution quality have been published. However, the evaluations in most of these publications are based on an application and problem instance combination that is very hard to come by if not impossible to obtain. This is a dilemma, because it means that results of newly developed approaches cannot be compared with published results what makes the latter less useful than they could be. The solution is of course to create a benchmark set that is open to the public. In this paper we describe a way to build a load balancing benchmark for general purpose libraries operating on graphs, and enumerate the required information, such that also existing applications might be modified to output the necessary data. Furthermore, we present a way to generate simple FEM related instances for testing purposes, which might form the basis of a future benchmark set. Applying the proposed methods, we run experiments with the two parallel state-of-the-art libraries Metis and Jostle and a prototype of the Party/DB library, and exemplarily compare their solutions according to several metrics and norms.

Barrier Synchronisation for occam-pi.

Abstract: This paper introduces a safe language binding for CSP multiway events (barriers) that has been built into occampi (an extension of the classical occam language with dynamic parallelism, mobile processes and mobile channels). Barriers provide a simple way for synchronising multiple processes and are the fundamental control mechanism underlying both CSP (Communicating Sequential Processes) and BSP (Bulk Synchronous Parallelism). The occampi barriers are more general than those of BSP (an occampi system can contain any number of barriers, with some processes ignoring them and some registered with many). On the other hand, they are also, currently, less general than those of CSP (occampi processes must commit to barrier synchronisation -- it cannot be used as part of a choice or ALT). Structured support for resignation, a higher-level CSP design pattern, is also built into occampi barriers. Applications are outlined for fine-grained modelling of dynamic systems, where the barriers are used for maintaining simulation time and synchronising safe access to shared data between millions of processes. Implementation details and early performance benchmarks (16 nanoseconds per process per barrier synchronisation on a 3.2 GHz. Pentium IV) are also presented, along with some likely directions for future research.

Distributed Models of Thread Level Speculation.

Abstract: This paper introduces a novel application of thread-level speculation to a distributed heterogeneous environment. We propose and evaluate two speculative models which attempt to reduce some of the method call overhead associated with distributed objects. Thread-level speculation exploits parallelism in code which is not provable free of data dependencies. Our evaluation of applying thread-level speculation to client-server applications resulted in substantial performance increases, on the order of 3 times for our initial model, and 21 times for the second.

Access Grid Technology in Classroom and Research Environments.

Abstract: Global communication is essential to industry, research and education. The Access Grid (AG) is a suite of hardware, software, and tools to facilitate communication and collaboration over the Internet. These resources are used at over 500 institutions worldwide to support group-to-group interactions across the Grid including collaborative research work sessions, tutorials, lectures, large-scale distributed meetings and training. This paper will provide an overview of the technology to encourage professionals to integrate benefits and tools of the Grid into their instruction and research. Furthermore, this paper will compare this new technology to more traditional videoconferencing and distributed collaborative working environments. Lastly, it will present issues and challenges that must be addressed to incorporate this momentous technology within the classroom and for collaboration throughout the world.

Strategies for Cache Invalidation of Location Dependent Data in Mobile Environment.

Abstract: Mobile computing as compared to traditional computing paradigms enables clients to have unrestricted mobility while maintaining network connections. Data management in this paradigm poses new challenging problems to the data base community. Location Dependent Information Services (LDIS) is an emergent application in this area where information provided to users depends on their current locations. Data caching at mobile clients plays a key role in data management due to its ability to improve system performance and overcome availability limitations. Spatial data cached in the mobile clients may become invalid because of the movement of the client. Cache Invalidation schemes aims to keep data consistency between the client’s cache and the server. To maintain consistency of LDD in cache, valid scope of that data item is identified and stored along with it in the client’s cache. In this paper, we study the selection procedure of finding best suitable candidate for valid scope (i.e., best suitable sub polygon of a given polygon) and propose a generalized algorithm which selects the best suitable candidate for valid scope. We compare its performance with the existing algorithms. More over, we also introduced new algorithm CEFAB which tries to improve the performance by considering the user movement pattern and speculation about its future access.

Self-Checking Fault Detection using Discrepancy Mirrors

Abstract: A method for robust detection of faults is developed based on pairwise parallel evaluation using Discrepancy Mirror logic. Discrepancy Mirrors provide coverage for the fault detector elements within the same mechanism used for the functional logic under test. The detector logic is self-testing and propagates functional outputs with adherence to a single fault-secure property so that erroneous outputs from any single fault are not propagated. Within the detector, bitwise equality comparisons are employed directly without additional data encoding/decoding schemes to determine the validity of the output. Fault handling is performed using the underlying data throughput so that additional test vectors are not required. The circuit was implemented for a Xilinx Virtex II Pro FPGA platform and fault-secure operation was verified using ModelSim-II for exhaustive stuck-at scenarios. Results indicate fault isolation in a pool of 100,000 resources using an expected value of 17.6 to 64.1 pairings when as little as one half of the inputs applied articulate the fault.

Distributed Service Discovery with Guarantees in Peer-to-Peer Networks using Distributed Hashtables.

Abstract: This paper proposes a protocol for decentralized service discovery with guarantees. We use a peer-to-peer network based on the distributed hashtable Chord that provides a structured overlay network in order to avoid flooding the whole network. Service descriptions are decomposed into portions which can be efficiently distributed and retrieved. We propose a way to evaluate our protocol by running simulations in comparison with a straightforward way of achieving the same goal in an unstructured, Gnutella-like network.

A game theoretical extended vickery auction mechanism for replicating data in large-scale distributed computing systems

Abstract: This paper addresses the problem of finegrained data replication on a set of Internet sites using the extended form of Vickrey auction called the N+1 st price auction. Specifically, we present an adaptive auction mechanism for replication of objects in a distributed system. The mechanism is adaptive in the sense that it changes the replica schema of the objects by continuously moving the schema towards an optimal one, while ensuring object concurrency control. The mechanism is experimentally evaluated against three well-known techniques from the literature: greedy, branch and bound, and genetic algorithms. The experimental results reveal that the proposed approach outperforms the three techniques in both the execution time and solution quality.

Web Services Agreement-based Resource Negotiation in UNICORE

Abstract: Service Level Agreements provide the foundation to negotia te for a distinct Quality of Service level between the provider and the consumer of a service. Since the Grid com munity is adopting concepts of Service-Oriented Architectures and Web Services are capturing their space wi thin the Grid landscape, resource management within Grids increasingly evolves towards the management of resou rces represented as services. To make allowance for this the Global Grid Forum develops the Web Services Agre ement specification to support standardised creation and negotiation of guarantees related to services. Th is paper illustrates the integration of a Web Services Agreement-based resource management framework into the UN ICORE Grid system, a development motivated by the system’s transition towards a service-oriented Grid an d the limitations of the current solution.

Scalable Analysis of Distributed Workflow Traces

Abstract: Bacterial response to nitric oxide (NO) is of major importance since NO is an obligatory intermediate of the nitrogen cycle. Transcriptional regulation of the dissimilatory nitric oxides metabolism in bacteria is Large-scale workflows are becoming increasingly important in both the scientific research and business domains. Science and commerce have both experienced an explosion in the sheer amount of data that must be analyzed. An important tool for analyzing these huge datasets is a compute cluster of hundreds or thousands of machines. However, debugging and tuning clusters requires specialized tools. Current cluster performance tools are more oriented towards tightly coupled parallel applications. We describe how the NetLogger Toolkit methodology is more appropriate for this class of cluster computing, and describe our new automatic work flow anomaly detection component. We also describe how this methodology is being used in the Nearby Supernova Factory (SN factory) project at Lawrence Berkeley National Laboratory.

ROME: Optimising Lookup and Load Balancing in DHT-Based P2P Networks.

Abstract: Distributed Hash Tables (DHTs) have been used in Peer-to-Peer networks to provide key lookups in typically O(log n) hops whilst requiring maintenance of only small amounts of routing state. We extend ROME, a layer which runs on top of the Chord DHT to provide control over network size through monitoring of node workload and propose the use of processes to reorganise nodes and add or remove them from a pool of available machines. We show this can reduce further the hop counts in networks where available node capacity exceeds workload, without the need to modify any processes of the underlying Chord protocol.

Millipede: A Multilevel Debugging Environment for Distributed Systems.

Abstract: The Millipede debugging environment is a realization of a multilevel debugging tool to quickly and easily correct the unique class of bugs which crop up in parallel and distributed systems. Through the use of the multilevel debugging approach, Millipede provides the abstractions, flexibility, and granularity to handle the unique and difficult challenges that arise in this field while avoiding several of the shortcomings of previous efforts.

Design and Analysis of Discrete Wavelet Transform(DWT) for Image Compression using VHDL.

Abstract: This paper presents the use of Electronic Design Automation (EDA) for implementing the 2-D Discrete Wavelet Transform (DWT) with JPEG2000 standard for image compression using Aldec Active HDL 3.5 tool. The design flow starts with specifications of the system behavior for implementation on silicon and it takes several steps before hardware implementation to meet the specifications. The software allows the analysis to ensure the designed chip satisfy the required specifications.

Cross Cutting Condensed Graphs.

Abstract: Condensed Graphs provide a graph based programming model that unifies availability-driven, coerciondriven and control-driven computing. In this paper we explore the issues of concurrency and synchronisation between condensed graph applications. A paradigm for condensed graph synchronisation is proposed and an architectural pattern is developed that forms the basis of a novel synchronisation mechanism.

A Performance Study of Parallel FFT in Clos and Mesh Networks.

Abstract: Though it is known that clos interconnection networks have many advantages over the other interconnection networks, it would be interesting to see the performance benefits that clos networks can bring to the real world applications. In this paper, we compare the performance of Fast Fourier Transforms (FFT) in clos networks with that of in mesh networks, a popular interconnection topology. We use both twodimensional (2-d) and three-dimensional (3-d) formulations of FFT for our studies. We show that the performance of FFT in clos networks is significantly better and we further show that 3-d formulation outperforms the 2-d formulation.

A Free Space Optical Bus Parallel Model Framework.

Abstract: Optical bus parallel computing models have gained research interest in recent years. Two types of models have emerged, the fiber-based models are theoretically relatively mature and used for algorithm design whereas free space models are more architecture-oriented but less theoretically mature. This paper proposes a framework model for describing attributes necessary for a viable free space bus architecture. Subsequently, the framework guides the design of a basic free space optical bus architecture that is based on a particular fiber-based model. In this way, the maturity of fiber-based models is combined with the advantages provided by free space

Fundamentals for Consistent Event Ordering in Distributed Shared Memory Systems

Abstract: A large number of tasks in distributed systems can be traced down to the fundamental problem of attaining a consistent global view on a distributed computation. This problem has been addressed by a number of studies which focus on systems with message passing as their only means of interprocess communication. In the paper at hand we extend this restricted system model by additionally accounting for an abstract memory to be shared by the processes. We specify necessary and sufficient conditions for constructing a consistent global view on such systems and present helpful definitions, which are meant to be a solid formal base for further studies.

A Peer-to-Peer Distributed Database System.

Abstract: This thesis implements a Peer-to-Peer (P2P) distributed database system (PDBS). In a PDBS, peers in a peer group have their own local databases, and information in these local databases can be shared among peers through user queries. As a fully decentralized P2P distributed database information sharing application, each node of a PDBS has both server and client functions. The peer server component includes the query service and schema service, and peer client component includes user interface and client controller. Both peer client component and peer server component access its local database component including user database and administration database. A PDBS supports peer users to create and send queries and update queries to other peers, and integrates the distributed query results. PDBS also provides facilities to establish virtual communication groups, as well as allowing peers to join and leave groups at any time. Peers are able to find other peers and query databases of the other peers. PDBS is implemented with the JXTA platform for Peer-to-Peer computing. PDBS is a complete application of JXTA. The query service and schema service of PDBS is implemented by the underling services of JXTA. PDBS schema service is used to create and search database schemas, and PDBS query service is to handle database queries.