| 1. |
|
|
| 2. |
- Johnsson, Lennart
(författare)
-
Band Matrix Systems Solvers on Ensemble Architectures
- 1986
-
Ingår i: Algorithms, Architectures and the Future of Scientific Computation. - : Texas Tech University Press. ; , s. 195-216
-
Bokkapitel (refereegranskat)abstract
- We present direct solvers for band matrix systems for processor ensembles configured a 2-dimensional meshes with end-around connections, binary trees, shuffle-exchange, perfect shuffle and boolean cube networks, and as clusters of processors with intracluster connections forming a torus or a boolean cube and intercluster connections forming binary trees, shuffle- exchange, perfect shuffle and boolean cube networks. The ensembles are assumed to be of the NIMD type, and each processor is equipped with substantial local storage. There is no shared storage, abd control is distributed.
|
|
| 3. |
|
|
| 4. |
- Johnsson, Lennart
(författare)
-
Data Parallel Supercomputing
- 1989
-
Ingår i: Use of Parallel Processors in Meteorology. - : Springer Verlag.
-
Bokkapitel (refereegranskat)abstract
- Supercomputers with a performence of a trillion floating-point operations per second, or more, can be produced in state-of-the-art MOS technologies. Such computers will have tens of thousands of processors interconnected by a network of bounded degree. Reducing the requried data motion trough a careful choice of data allocation and computational and routing algorithms is critical for performance. The management of thousands of processors can only be accomplished trough programming languages with suitable abstractions. We use Connection Machine as a model architecture for future supercomputers, and Fortran 8X as an example of a language with some of the abstractions suitable for programming thousands of processors. Some of the communication primitives suitable for expressing structured scientific computations are discussed, and their benefit with respect to performance illustrated. With thousands of processors engaged in the solution of a single scientific problem, several subtasks are often treaten concurrently in addition to the concurrent execution of each subtask. Some issues in constructing scientific libraries for such enviroments are discussed. Concurrent algorithms and performance data for matrix multiplication and the Fast Fourier Transformer are presented. The solution of the compressible Navier-Stokes equation in three spatial dimensions by an explicit finite difference method, and the solution of a paralbolic approximation of the Helmholtz equation by an implict method are two examples of applications for which data parallel implementations are described briefly. The Helmholtz equations models three dimensional acoustic waves in the ocean
|
|
| 5. |
|
|
| 6. |
- Johnsson, Lennart, et al.
(författare)
-
High Performance, Scalable Scientific Software Libraries
- 1994
-
Ingår i: <em>Portability and Performance in Parallel Processing</em>. - : John Wiley & Sons. ; , s. 159-208
-
Bokkapitel (refereegranskat)abstract
- Massively parallel processors introduces new demands on software systems with respect to performance, scalability, robustness and portability. The increased complexity of the memory systems and the increased range of problem sizes for which a given piece of software is used, poses serious challenges to software developers. The Connection Machine Scientific Software Library, CMSSL, uses several novel techniques to meet these challenges. The CMSSL contains routines for managing the data distribution and provides data distribution independent functionality. High performance is achieved through careful scheduling of operations and data motion, and through the automatic selection of algorithms at run--time. We discuss some of the techniques used, and provide evidence that CMSSL has reached the goals of performance and scalability for an important set of applications.
|
|
| 7. |
|
|
| 8. |
- Johnsson, Lennart
(författare)
-
Highly Parallel Banded Systems Solvers
- 1987
-
Ingår i: Parallel Computations and Their Impact on Mechanics. - : ASME Press. ; , s. 187-208
-
Bokkapitel (refereegranskat)abstract
- We present algorithms for the solution of banded systems of equations on parallel architectures, in particular ensemble architectures, i.e., architectures that have a large number of processing elements. Each processor has its own local storage. The band is considered dense. Concurrent elimination of a single variable yields a linear speed-up for ensembles configured as tori, or Boolean cubes if N>m, with a maximum ensemble size of m(m+R) (or 2m(m+R)) processors for a banded system of N equations, bandwith 2m + 1 and R right hand sides. The minimum attainable computational complexity is of order O(N). Concurrent elimination of multiple variables as well as concurrent elemination of each such variable yields a minimum complexity of O(m+ m log2N/m) for a total of (2m + R)N ensemble nodes. To attain this complexity the ensemble should be configured as clusters, each in the form of a torus of dimension m by 2m + R, or a Booelan cube of appropiate dimension. Furthermore, corresponding processors in different clusters are assumed to be interconnected to form a binary tree, shuffle-exchange, perfect shuffle, or Boolean cube network. The number of clusters should be of order 0(N/m) for minimum computational comlpexity
|
|
| 9. |
- Johnsson, Lennart
(författare)
-
Language and Compiler Issues in Scalable High Performance Libraries
- 1993
-
Ingår i: <em>Compilation Techniques for Novel Architectures</em>. - : Springer-Verlag New York.
-
Bokkapitel (refereegranskat)abstract
- Library functions for scalable architectures must be designed to correctly and efficiently support any distributed data structure that can be created with the supported languages and associated compiler directives. Libraries must be designed also to support concurrency in each function evaluation, as well as the concurrent application of the functions to disjoint array segments, known as multiple-instance computation. Control over the data distribution is often critical for locality of reference, and so is the control over the interprocessor data motion. Scalability, while preserving efficiency, implies that the data distribution, the data motion, and the scheduling is adapted to the object shapes, the machine configuration, and the size of the objects relative to the machine size. The Connection Machine Scientific Software Library is a scalable library for distributed data structures. The library is designed for languages with an array syntax. It is accessible from all supported languages (Lisp, C, CM-Fortran, and Paris (PARallel Instruction Set) in combination with Lisp, C, and Fortran 77). Single library calls can manage both concurrent application of a function to disjoint array segments, as well as concurrency in
|
|
| 10. |
- Johnsson, Lennart
(författare)
-
Massively Parallel Computing: Data distribution and communication
- 1993
-
Ingår i: <em>Parallel Architectures and their Efficient Use</em>. - : Springer-Verlag New York. ; , s. 68-92
-
Bokkapitel (refereegranskat)abstract
- We discuss some techniques for preserving locality of reference in index spaces when mapped to memory units in a distributed memory architecture. In particular, we discuss the use of multidimensional address spaces instead of linearized address spaces, partitioning of irregular grids, and placement of partitions among nodes. We also discuss a set of communication primitives we have found very useful on the Connection Machine systems in implementing scientific and engineering applications. We briefly review some of the techniques used to fully utilize the bandwidth of the binary cube network of the CM--2 and CM--200, and give some performance data from implementations of communication primitives.
|
|
