MPI Environment Research Articles

Nonlinear Behaviour in the MPI-Parallelised Model of the Rat Somatosensory Cortex

Mammalian brains consisting of up to 10 11 neurons belong to group of the most complex systems in the Universe. For years they have been one of the hardest objects of simulation. There are many different approaches to modelling of neurons, but one of the most biologically correct is Hodgkin-Huxley (HH) model. Simulations that require solving a large number of nonlinear differential equations (fundamental in HH model) are always time and power consuming. The structures discussed in this article simulate a part of the rat somatosensory cortex. We use a modular architecture of the network divided into layers and sub-regions. Because of a high degree of complexity effective parallelisation of algorithms is required. We propose method of parallelisation for the network and the results of simulations using GENESIS parallelised for MPI environment are presented. An occurrence of nonlinear behaviour is demonstrated. Most notably, in large biological neural networks consisting of the HH neurons, nonlinearity is shown to manifest itself in the Poincare sections generated for the varying value of neural membrane's potential.

High-Scalability Parallelization of a Molecular Modeling Application: Performance and Productivity Comparison Between OpenMP and MPI Implementations

Important components of molecular modeling applications are estimation and minimization of the internal energy of a molecule. For macromolecules such as proteins and amino acids, energy estimation is performed using empirical equations known as force fields. Over the past several decades, much effort has been directed towards improving the accuracy of these equations, and the resulting increased accuracy has come at the expense of greater computational complexity. For example, the interactions between a protein and surrounding water molecules have been modeled with improved accuracy using the generalized Born solvation model, which increases the computational complexity to O (n3). Fortunately, many force-field calculations are amenable to parallel execution. This paper describes the steps that were required to transform the Born calculation from a serial program into a parallel program suitable for parallel execution in both the OpenMP and MPI environments. Measurements of the parallel performance on a symmetric multiprocessor reveal that the Born calculation scales well for up to 144 processors. In some cases the OpenMP implementation scales better than the MPI implementation, but in other cases the MPI implementation scales better than the OpenMP implementation. However, in all cases the OpenMP implementation performs better than the MPI implementation, and requires less programming effort as well.

Blocking vs. non-blocking coordinated checkpointing for large-scale fault tolerant MPI Protocols

A long-term trend in high-performance computing is the increasing number of nodes in parallel computing platforms, which entails a higher failure probability. Fault tolerant programming environments should be used to guarantee the safe execution of critical applications. Research in fault tolerant MPIs has led to the development of several fault tolerant MPI environments. Different approaches are being proposed using a variety of fault tolerant message passing protocols based on coordinated checkpointing or message logging. The most popular approach is with coordinated checkpointing. In the literature, two different concepts of coordinated checkpointing have been proposed: blocking and non-blocking. However they have never been compared quantitatively, and their respective scalabilities remain unknown. The contribution of this paper is to provide the first comparison between these two approaches and a study of their scalabilities. We have implemented the two approaches within the MPICH environments and evaluate their performance using the NAS parallel benchmarks.

Second derivatives in generalized Born theory

Generalized Born solvation models offer a popular method of including electrostatic aspects of solvation free energies within an analytical model that depends only upon atomic coordinates, charges, and dielectric radii. Here, we describe how second derivatives with respect to Cartesian coordinates can be computed in an efficient manner that can be distributed over multiple processors. This approach makes possible a variety of new methods of analysis for these implicit solvation models. We illustrate three of these methods here: the use of Newton-Raphson optimization to obtain precise minima in solution; normal mode analysis to compute solvation effects on the mechanical properties of DNA; and the calculation of configurational entropies in the MM/GBSA model. An implementation of these ideas, using the Amber generalized Born model, is available in the nucleic acid builder (NAB) code, and we present examples for proteins with up to 45,000 atoms. The code has been implemented for parallel computers using both the OpenMP and MPI environments, and good parallel scaling is seen with as many as 144 OpenMP processing threads or MPI processing tasks.

Cumulvs: Interacting with High-Performance Scientific Simulations, for Visualization, Steering and Fault Tolerance

High-performance computer simulations are an increasingly popular alternative or complement to physical experiments or prototypes. However, as these simulations grow more massive and complex, it becomes challenging to monitor and control their execution. CUMULVS is a middleware infrastructure for visualizing and steering scientific simulations while they are running. Front-end “viewers” attach dynamically to simulation programs, to extract and collect intermediate data values, even if decomposed over many parallel tasks. These data can be graphically viewed or animated in a variety of commercial or custom visualization environments using a provided viewer library. In response to this visual feedback, scientists can “close the loop” and apply interactive control using computational steering of any user-defined algorithmic or model parameters. The data identification interfaces and gathering protocols can also be applied for parallel data exchange in support of coupled simulations, and for application-directed collection of key program data in checkpoints, for automated restart in response to software or hardware failures. CUMULVS was originally based on PVM, but interoperates well with simulations that use MPI or other parallel environments. Several alternate messaging systems are being integrated with CUMULVS to ease its applicability, e.g. to MPI. CUMULVS has recently been integrated with the Common Component Architecture (CCA) for visualization and parallel data redistribution (referred to as “MxN”), and also with Global Arrays. This paper serves as a comprehensive overview of the CUMULVS capabilities, their usage, and their development over several years.

A parallel Broyden approach to the Toeplitz inverse eigenproblem

AbstractIn this work we show a portable sequential and a portable parallel algorithm for solving the inverse eigenproblem for real symmetric Toeplitz matrices. Both algorithms are based on Broyden's method for solving nonlinear systems. We reduced the computational cost for some problem sizes, and furthermore we managed to reduce spatial cost considerably, compared in both cases with parallel algorithms proposed by other authors and by us, although sometimes quasi‐Newton methods (as Broyden) do not reach convergence in all the test cases. We have implemented the parallel algorithm using the parallel numerical linear algebra library SCALAPACK based on the MPI environment. Experimental results have been obtained using two different architectures: a shared memory multiprocessor, the SGI PowerChallenge, and a cluster of Pentium II PCs connected through a myrinet network. The algorithms obtained are scalable in all the cases. Copyright © 2004 John Wiley & Sons, Ltd.

Coordinated checkpoint versus message log for fault tolerant MPI

Large clusters, high availability clusters and grid deployments often suffer from network, node or operating system faults and thus require the use of fault tolerant programming models. MPI is one of the most widely adopted programming models for high performance computing. There are several approaches for fault tolerance in an MPI environment. The automatic and transparent ones are based on either coordinated or uncoordinated checkpoint associated with a message log strategy. There are many protocols and optimisations for these approaches and several implementations have been made. However, few results of comparison between them exist. Coordinated checkpoint has the advantage of a very low overhead as long as the execution stays fault free. In contrast, uncoordinated checkpoint must be complemented by a message log protocol which adds a significant penalty for all message transfers even for fault free executions. The drawbacks of coordinated checkpoint are the synchronisation cost before the checkpoint, the synchronised checkpoint cost and the restart cost after a fault. Message log does not suffer from these problems, as it processes checkpoint and restart independently. These differences suggest that the best approach depends on the fault frequency. This paper investigates this question from a fair experimental protocol: we implement and test two protocols (coordinated checkpoint and pessimistic message log) on the same system and we compare them on a cluster according to the frequency of faults that are generated artificially. The main conclusion is that uncoordinated checkpoint is relevant for a large scale cluster from one fault every hour for applications with large dataset.

Introduction to Parallel Programming with MPI

This article provides basic information for making parallel programs with MPI. The author have chosen MPI as a basic programming interface on PC clusters, because the parallel execution model it assumes is very close to clusters. The MPI functions introduced in this article are for sending or receiving messages, broadcasting or reducing data, and starting and ending MPI environments. This article also introduces how to compile and execute MPI programs.

Inexact Quasi-Newton methods for sparse systems of nonlinear equations

In this paper, we present the results obtained by solving consistent sparse systems of n nonlinear equations F( x)=0, by a Quasi-Newton method combined with a p block iterative row-projection linear solver of Cimmino type, 1≤ p≪ n. Under weak regularity conditions for F, it is proved that this Inexact Quasi-Newton method has a local, linear convergence in the energy norm induced by the preconditioned matrix HA, where A is an initial guess of the Jacobian matrix, and it may converge too superlinearly. The matrix H=[ A 1 +,…, A i +,…, A p +], where A i += A i T( A i A i T) −1 is the Moore–Penrose pseudo-inverse of the m i × n block A i , the preconditioner. A simple partitioning of the Jacobian matrix was used for solving a set of nonlinear test problems with sizes ranging from 1024 to 131 072 on the CRAY T3E under the MPI environment.

An Assessment of MPI Environments for Windows NT

In this paper we evaluate the MPI environments currently available for Windows NT on the Intel IA32 and Compaq/DEC Alpha architectures. We present benchmark results for low-level communication and for the NAS Parallel Benchmarks to allow comparison with other systems, but our primary interest is determining real application performance and robustness in production cluster environments. For this we use PAFEC-FE, a large FORTRAN code for finite-element analysis. We present results from three MPI implementations, two architectures, and three networking technologies (10 and 100 Mbit/s Ethernet and 1 Gbit/s Myrinet).

Parallelization of a finite element surface fitting algorithm for data mining

A major task in data mining is to develop automatic techniques to process and to detect patterns in very large data sets. An important data mining technique is multivariate regression, and an essential sub task is the estimation of interaction surfaces, i.e. the estimation of functions of two variables. Thin plate splines provide a very good method to determine an approximating surface. Obtaining standard thin plate splines requires the solution of a dense linear system of equations of order n , where n is the number of observations. Standard thin plate splines may not be practical, because the number of observations for data mining applications is often in the millions. We have developed a finite element approximation of a spline that can handle data sizes with millions of records. The resolution of the finite element method can be chosen independently from the number of observations. The observation data is read from secondary storage once, and does not need to be stored in memory. In this paper, we present a first parallel implementation of this method in an MPI environment.

Message-passing environments for metacomputing

In this paper, we present the three libraries PACX-MPI, PLUS, and PVMPI that provide message-passing between different high-performance computers in metacomputing environments. Each library supports the development and execution of distributed metacomputer applications. The PACX-MPI approach offers a transparent interface for the communication between two or more MPI environments. PVAMPI allows the user spawning parallel processes under the MPI environment. The PLUS protocol bridges the gap between vendor-specific (e.g., MPL, NX, and PARIX) and vendor-independent message-passing environments (e.g., PVM and MPI). Moreover, it offers the ability to create and control processes at application runtime.

Method of particles in visual clustering of multi-dimensional and large data sets

A method dedicated for visual clustering of N-dimensional data sets is presented. It is based on the classical feature extraction technique – the Sammon’s mapping. This technique empowered by a particle approach used in the Sammon’s criterion minimization makes the method more reliable, general and efficient. To show its reliability, the results of tests are presented, which were made to exemplify the algorithm ‘immunity’ from data errors. The general character of the method is emphasized and its role in multicriterial analysis discussed. Due to inherent parallelism of the methods, which are based on the particle approach, the visual clustering technique can be implemented easily in parallel environment. It is shown that parallel realization of the mapping algorithm enables the visualization of data sets consisting of more than 104 multi-dimensional data points. The method was tested in the PVM, MPI and data parallel environments on an HP/Convex SPP/1600. In this paper, the authors compare the parallel algorithm performance for these three interfaces. The approach to visual clustering, presented in the paper, can be used in visualization and analysis of large multi-dimensional data sets.