Heterogeneous Network-Based Concurrent Computing Systems

Grid computing originated as paradigm for high performance computing and massive parallel processing. Currently, Grid systems have become an important paradigm for efficiently solving large scale complex problems from many fields. On the other hand, P2P paradigm originated from file sharing but each time more is being used for the development of large scale distributed platforms. Grid and P2P systems have thus followed different trajectories pushed by different motivations, needs and research communities. Yet, both paradigms are evolving in a way that each time more they are sharing common characteristics and are mutually benefiting from their best features. Among these characteristics, we could distinguish the cooperative model for solving complex problems by exploiting the large computing capacity offered by the nodes of the system altogether. As such, Grid and P2P systems have achieved notable success, in particular, for e-Science applications, a family of complex applications arising in science and engineering that need considerable computation power.

Recently users from both high-performance scientific community and general-purpose applications have shown keen interest in parallel processing due to its higher performance, lower cost, and sustained productivity [148]. To solve a computationally intensive problem efficiently on a cluster of existing computers, distributed computing involves a significantly lower cost factor [156]. Although it is difficult for a distributed computing user to achieve the computational capacity of large massively parallel processors (MPP), it is possible to solve large-size problems by combining a variety of distributed computing resources, connected by high-speed networks. This approach has advantages in terms of flexibility, scalability, and low cost. The advantage of using a cluster of workstations as the computational platform is that a cluster of a large number of workstations is easily available. A disadvantage is that there may be many users running unrelated tasks on the workstations so that the available computing resource for each task fluctuates in an unpredictable manner. Furthermore, communication between workstations is relatively slow. Although the performance of generalized cross-validation (GCV)–based threshold selection scheme is excellent, it is costly from CPU time viewpoint when implemented sequentially. In contrast to the traditional parallel approaches, which rely on specialized parallel machines, present work explores the potential of distributed systems for parallelism. The master/slave model is adopted for control of machines. This chapter is organized as follows. Section 4.1 presents background material and review of work in related area. Section 4.2 presents parallel algorithm of DWT. Basics of PVM programming has been discussed in Sect. 4.3. KeywordsPVMParallel algorithmsSpeedup

The JPVM library is a software system for explicit message-passing based distributed memory MIMD parallel programming in Java. The library supports an interface similar to the C and Fortran interface provided by the Parallel Virtual Machine (PVM) system, but with syntax and semantics modifications afforded by Java and better matched to Java programming styles. The similarity between JPVM and the widely used PVM system supports a quick learning curve for experienced PVM programmers, thus making the JPVM system an accessible, low-investment target for migrating parallel applications to the Java platform. At the same time, JPVM offers novel features not found in standard PVM such as thread safety, multiple communication end-points per task, and default-case direct message routing. JPVM is implemented entirely in Java, and is thus highly portable among platforms supporting some version of the Java Virtual Machine. This feature opens up the possibility of utilizing resources commonly excluded from network parallel computing systems such as Macintosh and Windows-NT based systems. Initial applications performance results achieved with a prototype JPVM system indicate that the Java-implemented approach can offer good performance at appropriately coarse granularities.

Grid computing emerged as a paradigm for high‐performance computing and massive parallel processing. Currently, Grid systems have become an important paradigm for efficiently solving large‐scale complex problems from many fields. On the other hand, P2P paradigm originated from file‐sharing, but each time more is being used for the development of large‐scale distributed platforms. Grid and P2P systems have thus followed different trajectories pushed by different motivations, needs and research communities. In fact, both paradigms are evolving in a way that each time they are sharing more common characteristics and are mutually benefiting from their best features. Among these characteristics, we could distinguish the cooperative model for solving complex problems by exploiting their large computing capacity. As such, Grid and P2P systems have achieved notable success, in particular, for e‐Science applications, a family of complex applications arising in science and engineering that need considerable computation power. Despite important advances in the design and use of Grid and P2P systems, they remain still difficult to implement and apply to real‐life problems. The main difficulties reside in the lack of easy‐to‐use middleware for Grid and P2P, in the complexities of setting up and in the tedious task of deploying real‐world Grid/P2P platforms as well as in experimental studies which are often complex and not easy to repeat. In this paper we survey and analyze the advances in communication libraries and middleware for both Grid and P2P systems as well as their limitations when used in real Grid and P2P infrastructures. We also bring examples of real‐life applications of massive data processing that can be efficiently handled through Grid and P2P approaches. Copyright © 2010 John Wiley & Sons, Ltd.

The Parallel Virtual Machine (PVM) system is a software framework that enables heterogeneous concurrent computing in heterogeneous environments. Since its conception in 1989, it has evolved into a very popular and widely used platform for cluster and network computing. The programming model supported by PVM, straightforward formulation of the API, robust portable implementations, and timeliness have all played an important role in PVM's impact on high performance computing. We discuss the overall design of PVM and describe its ongoing evolution, recently introduced features, and future directions.KeywordsVirtual MachineMessage PassingRemote Procedure CallCommunication VolumeParallel Virtual MachineThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

Recent advances in the power of parallel computers have made them attractive for solving large computational problems. Scalable parallel programs are particularly well suited to Massively Parallel Processing (MPP) machines since the number of computations can be increased to match the available number of processors. Performance tuning can be particularly difficult for these applications since it must often be performed with a smaller problem size than that targeted for eventual execution. This research develops a performance prediction methodology that addresses this problem through symbolic analysis of program source code. Algebraic manipulations can then be performed on the resulting analytical model to determine performance for scaled up applications on different hardware architectures. >

The linear quadtree is a useful data structure for representing an image for the sake of the storage saving and further image manipulations. In this paper we propose a linear quadtree coding scheme and implement this algorithm on the parallel virtual machine (PVM). Our goal is to demonstrate the applicability of using the PVM in combining the computing power of computers in a network to solve this kind of image processing problems. The processors in the PVM are organized as a master-slave paradigm and various numbers of processors are applied for different PVM’s to compare their performances. Experimental results show that the speedup of solving this image encoding problem in parallel is quite satisfactory. With such a PVM environment which is easily accessible in the public domain, high performance computing is truly possible without additional hardware cost.

The Evolving Tree Transformation System (ETTS) has been developed as a model of a reconfigurable learning machine. In this paper we present an examination of the opportunities for parallel processing in the Evolving Tree Transformation System (ETTS) as well as an implementation which takes advantage of the inherent parallelism. The implementation uses the distributed software system Parallel Virtual Machine (PVM) to handle process control and inter-process communication. Experiments conducted on the implementation are presented which reveal a strong correlation between performance, the size and nature of the input data and the size of the parallel machine. Using a sixteen processor virtual parallel machine, we are able to achieve a speedup of 13.5 using idealized data and a speedup of 7.6 using real (natural) data. Performance models for the parallel implementations are also presented.

6508 Background: Electronic decision support is increasingly prevalent in clinical practice. Traditional tools map guidelines into an interactive platform. An alternative method builds on experience-based learning. Methods: Memorial Sloan-Kettering (MSK), IBM and WellPoint teamed to develop IBM Watson – a cognitive computing system leveraging natural language processing (NLP), machine learning (ML) and massive parallel processing – to help inform clinical decision making. We made a prototype for lung cancers using manufactured and anonymized patient cases. We configured this tool to read medical language and extract specific attributes from each case to identify appropriate treatment options benchmarked against MSK expertise, anonymized patient cases and published evidence. Treatment options reflect consensus guidelines and MSK best practices where guidelines are not granular enough to match treatments to unique patients. Analysis and building accuracy is ongoing and iterative. Results: 420 manufactured and 525 anonymized patient cases trained the initial models. Early results show accuracy improvement in NLP and ML in identifying treatment options (Table). All treatment plans were guideline adherent. A proportion of cases showed the need to incorporate tailored treatment plans reflecting MSK’s practice beyond guidelines – e.g. 11% of cases required addressing a site of critical metastasis before initiating guideline supported treatment. Conclusions: IBM Watson is extracting information from free text medical records that supports building ML models to assist in selecting treatments for persons with lung cancers. This tool can select treatment options from consensus guidelines, and, through ML, it will identify personalized treatment plans. Training is ongoing to improve individualized decision making and optimize the web-based tool that connects with IBM Watson. [Table: see text]

Optical interconnects in massively parallel processing (MPP) systems will require a variety of optical components regardless of the specific implementation of the optics in such systems. These components span a range of complexity from conceptually simple structures such as connectors to quite complex structures such as VCSELs. While numerous optical components have been developed for fiber-based systems, in general, many of these components do not adequately address the specific needs of MPP systems. Specifically the components must provide specific performance requirements while also lending themselves to automated manufacturing techniques and being highly compatible with current electronic manufacturing techniques. In this work, we present a novel structure for the demultiplexing (and multiplexing) of guided-wave optical signals. This device has the unique feature of being suitable for single-mode as well as multimode systems. In addition, its structure is highly manufacturable due to its planar technology that does not require adding highly reflecting surfaces. The device works using the principal of total internal reflection off a grating surface. In this paper we present a theoretical analysis of the structure and evaluate several performance issues.

We propose a new scheme of subband adaptive digital filter(ADF) which allows massive parallel processing. We discuss the power spectral density of the down-sampled processes and propose a new cost function which is suitable for subband ADF. The subband scheme of adaptive system identification using the proposed cost function is considered. It is shown that we can carry out precise system identification based on the frequency domain sampling theorem, in which the proposed cost function is applied to the subband ADF using polyphase filter banks. The cost function is defined in each bin; as a result, it enables the perfect parallel processing of ADFs. >

It is often necessary to get steady-state waveforms of a power electronic circuit for its design or analysis. Direct steady-state analysis techniques have been researched for efficient simulation. The shooting method is one of these and it is often utilized. It is known that its multiple type form is numerically more stable than the usual single type one. Furthermore it is suitable for parallel processing and it is possible to reduce its computation time. Today parallel computing systems, or PC cluster systems which are based on PVM (parallel virtual machine) or MPI (message passing interface), are easily available. This paper proposes an efficient steady-state simulation technique based on the multiple shooting method for a power electronic circuit by parallel processing. First the principle is investigated and made clear how it works. Then it is extended to be applicable to the envelope following analysis for much more efficient simulation by combining the principles. Finally the developed method can utilize both advantages of the two principles. It takes a long CPU time to get steady-state solution even by the single shooting method. By applying the single shooting envelope following technique, it is reduced to 1/10 for an example case. By applying the proposed multiple shooting EF technique, it is reduced about to 1/30 with 10 CPUs. This shows effectiveness of the method.

MATLAB Distributed Computing Server (MDCS) and Parallel Virtual Machine (PVM) software are two types of distributed computing environments. MDCS is recently used in selecting the best network training algorithm and assessing the effect of parallelization. Since it is practical and user friendly, it gives PVM the opportunity to become the communication paradigm of choice. The PVM provides a powerful set of process control and dynamic resource management features. In the distributed parallel computing (DPC), however, both solutions have different strengths and limitations. Based on these concerns, this paper compares the numerical analysis for mathematical modeling of large sparse 2D and 3D of second order parabolic partial differential equations (PDE) on MDCS and PVM based on parallel performance indicators (PPI). The PDE geometry is discretized into a sparse grid structure using the FDM method. In using a method with the highest accuracy, Parallel Alternating Group Explicit (PAGE) scheme was chosen. The parallel strategies focus on the PAGE's convergence speed and various domain de-composition techniques, as well as a block iterative scheme and load balance using fine granular techniques. Furthermore, comparison of distributed computing environments also relies on multiple processors running on Unix-like operating systems with Fedora installed to support large-scale simulations. The analysis and validation of PPI for both communication software is also investigated in this paper. For a sequential algorithm, accuracy, estimation of error, and stability are employed as indicators. As a conclusion, based on the PPI and numerical analysis, the tables and graphs show that compared to MDCS, PVM is a better environment for solving multidimensional parabolic PDE modeling.KeywordsDistributed computingParabolic PDEsMDCSPVM

Parallel implementation of 2D LGA and MD simulations

The author presents the architecture of a new family of parallel operating systems based on an alternative design approach when compared to multiprogramming, namely, generic construction of parallel virtual machines. This construction is made according to needs of models for parallel programming which shall be supported as subsystems by the parallel operating system kernel. The author presents the PAROS architecture and its generic kernel ParX which has been designed for message passing parallel computers. Several parallel virtual machines are also described, such as for instance a diffusion virtual machine, and a message passing-shared data virtual machine. >