This paper describes schemes for rapidly computing numerical values of definite integrals to very high accuracy (hundreds to thousands of digits) on highly parallel computer systems. Such schemes are of interest not only in computational physics and computational chemistry, but also in experimental mathematics, where high-precision numerical values of definite integrals can be used to numerically discover new identities. This paper presents performance results for 1-D and 2-D integral test suites on highly parallel computer systems. Results are also given for certain problems that derive from mathematical physics. One of these results confirms a conjecture to 20,000 digit accuracy. The performance rate for this calculation is 690 Gflop/s on 1024 CPUs of a state-of-the-art parallel system. Other results, which range in precision from 120 to 500 digits, and for 1-D, 2-D, 3-D and 4-D integrals, derive from Ising theory. The largest of these calculations required 28 hours on 256 CPUs. We believe that these are the first instances of evaluations of nontrivial 3-D and 4-D integrals to multi-hundred-digit accuracy.

Grids are semantically different from other distributed systems. Therefore, performance analysis, just like any other technique requires careful reconsideration. This paper analyses the fundamental differences between grids and other systems and points out the special requirements raised to performance analysis. The aim of this paper is to survey the special problems, the possible directions and the existing solutions. A monitoring system, that is able to support the posed requirements is introduced as an example.

. Applying fast scientific computing algorithms to large problems presents a difficult engineering problem. We describe a novel architecture for addressing this problem that uses a robust client-server model for interactive large-scale linear algebra computation. We discuss competing approaches and demonstrate the relative strengths of our approach. By way of example, we describe MITMatlab, a powerful transparent client interface to the linear algebra server. With MITMatlab, it is now straightforward to implement full-blown algorithms intended to work on very large problems while still using the powerful interactive and visualization tools that Matlab provides. We also examine the efficiency of our model by timing selected operations and comparing them to commonly used approaches. 1 Introduction We describe a novel architecture for a "linear algebra server" that operates on very large matrices. Matrices are created by the server and distributed across many machines or processors. Oper...

This paper presents parallel scalable algorithms for high dimensional surface fitting and predictive modelling which can be used in data mining applications. The presented algorithms are based on techniques like finite elements, thin plate splines, additive models and wavelets. They consist of two phases: First, data is read from secondary storage and a linear system is assembled. Secondly, the linear system is solved. The assembly can be done with almost no communication and the size of the linear system is independent of the data size. Thus the presented algorithms are both scalable with the data size and the number of processors. Key words: Data Mining, Thin Plate Splines, Additive Models, Wavelets, Parallel Linear System 1 Introduction In the last few years there has been an explosive growth in the amount of data being collected. The computerisation of business transactions and the use of bar codes in commercial outlets has provided businesses with enormous amounts of data. In s...

This paper describes MITMatlab, a system that enables users of supercomputers to transparently work on large data sets within Matlab. MITMatlab communicates with an external server that is responsible for storing and operating on the data. Through the use of Matlab's object oriented features, we can handle this data as though it were "in" Matlab. For example, we can type [u,s,v] = svds(a,5) in Matlab and get results regardless of whether the matrix a has fifty or fifty million non-zero elements. We present the structure and details of our implementation along with some examples showing MITMatlab in action. 1 Introduction This paper describes MITMatlab, a system that enables users of supercomputers to work in parallel transparently on large data sets within Matlab. MITMatlab is based on the Parallel Problems Server (PPServer)[8], a standalone linear algebra server that provides a mechanism for executing distributed memory algorithms on large data sets. This work is motivated by the des...

This paper discusses methods for expressing and tuning the performance of parallel programs, by using two programming models in the same program: distributed and shared memory. Case studies show how mixing these two approaches results in efficient machine use because the two models match the two levels of parallelism present in the architecture of current SMP clusters. This issue is important for anyone who uses these large machines for parallel programs as well as for those who study combinations of the two programming models.

Abstract. In this paper we present an all-optical network architecture and a systolic routing protocol for it. The sparse optical torus network consists of an n × n torus, where processors are deployed diagonally. The systolic routing protocol ensures that no electro-optical conversion is needed in the intermediate routing nodes and all the packets injected into the routing machinery will reach their targets without collisions. 1

Device Interface (ADI). The ADI provides a fairly high-level abstraction of a communication device that should be realized by the underlying low-level communication library like Nexus. The Nexus implementation of ADI establishes a full connection among the communication links used in the MPICH program and the ADI functions are realized as RSR message handlers. The implementation of ADI on different machines may apply different protocols for transferring data. 6.1.2 Resource management Globus has a hierarchical resource management concept which is built on three major components: . The Resource Specification Language (RSL) . A hierarchical broker architecture . Globus Resource Allocation Managers (GRAMs) The RSL is used to specify the resource requirements of a particular application. It contains expressions like: . "Run a distributed simulation with 100K entities" . "Perform a parameter study with 10K separate trials" . "Create a shared virtual space with participants X, Y and...

vii Chapter 1 Introduction 1 1.1 What is Seuss . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 1.2 Goals of this Thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 1.3 Structure of this Thesis . . . . . . . . . . . . . . . . . . . . . . . . . 4 Chapter 2 Seuss 5 2.1 Boxes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 2.2 Procedures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 2.2.1 Actions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 2.2.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 2.3 Clones . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 2.4 Program Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 2.5 Program Execution . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 Chapter 3 SeussCpp : Seuss for C++ 15 3.1 Program Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16 3.2 Module Definition . . . . ...

Abstract. This paper presents the Interval Categorizer Tessellationbased Model (ICTM) for the simultaneous categorization of geographic regions considering several characteristics (e.g., relief, vegetation, land use etc.). Interval techniques are used for the modelling of uncertain data and the control of discretization errors. HPC-ICTM is an implementation of the model for clusters. We analyze the performance of the HPC-ICTM and present results concerning its application to the relief/land-use categorization of the region surrounding the lagoon Lagoa Pequena (RS, Brazil), which is extremely important from an ecological point of view. 1