Algorithm Engineering: 3rd International Workshop, Wae'99 London, Uk, July 19-21, 1999 Proceedings

ISBN-13: 9783540664277 / Angielski / Miękka / 1999 / 368 str.

Jeffrey Scott Vitter; Christos D. Zaroliagis; J. S. Vitter

Opis

This work considers practical parallel list-ranking algorithms. The model for which programs are written is a single-program multiple-data (SPMD) ri- ingmodel." Thismodel isdesignated asa programmer smodelfora ne-grained computation framework called Explicit Multi-Threading (XMT), which was - troduced in VDBN98]; the XMT framework covers the spectrum from al- rithms through architecture to implementation; it is meant to provide a pl- form for faster single-task completion time by way of instruction-level par- lelism (ILP). The performance of XMT programs is evaluated as follow: the performance of a matching optimized XMT assembly code is measured within an XMT execution model. (We use in the current paper the so-called Spawn- MT programmingmodel - the easier to implement amongthe two programming modelspresented in VDBN98]). The XMT approach deviatesfromthe standard PRAM approach by incorporating reduced synchrony and departing from the lock-step structure in its so-called asynchronous mode. Our envisioned platform uses an extension to a standard serial instruction set. This extension e ciently implements PRAM-style algorithms using explicit multi-threaded ILP, which allows considerably more n e-grained parallelism than the previously studied parallel computing implementation platforms/models. The list ranking problem was the rst problem considered as we examined and re ned many of the concepts in the XMT framework. The problem arises in parallel algorithmson lists, trees and graphs and is considered a fundamental problemin the theory of parallelalgorithms. Experimental results are presented."

Spis treści

Invited Lectures.- Selecting Problems for Algorithm Evaluation.- BSP Algorithms — “Write Once, Run Anywhere”.- Ten Years of LEDA: Some Thoughts.- Contributed Papers.- Computing the K Shortest Paths: A New Algorithm and an Experimental Comparison.- Efficient Implementation of Lazy Suffix Trees.- Experiments with List Ranking for Explicit Multi-Threaded (XMT) Instruction Parallelism.- Finding Minimum Congestion Spanning Trees.- Evaluation of an Algorithm for the Transversal Hypergraph Problem.- Construction Heuristics and Domination Analysis for the Asymmetric TSP.- Counting in Mobile Networks: Theory and Experimentation.- Dijkstra’s Algorithm On-Line: An Empirical Case Study from Public Railroad Transport.- Implementation and Experimental Evaluation of Graph Connectivity Algorithms Using LEDA.- On-Line Zone Construction in Arrangements of Lines in the Plane.- The Design and Implementation of Planar Maps in CGAL.- An Easy to Use Implementation of Linear Perturbations within Cupgal.- Analysing Cache Effects in Distribution Sorting.- Fast Regular Expression Search.- An Experimental Evaluation of Hybrid Data Structures for Searching.- LEDA-SM: Extending LEDA to Secondary Memory.- A Priority Queue Transform.- Implementation Issues and Experimental Study of a Wavelength Routing Algorithm for Irregular All-Optical Networks.- Estimating Large Distances in Phylogenetic Reconstruction.- The Performance of Concurrent Red-Black Tree Algorithms.- Performance Engineering Case Study: Heap Construction.- A Fast and Simple Local Search for Graph Coloring.- BALL: Biochemical Algorithms Library.- An Experimental Study of Priority Queues in External Memory.

O autorze

Jeffrey Scott Vitter is the Gilbert, Louis, and Edward Lehrman Professor of Computer Science at Duke University. With a PhD from Stanford University, he is a National Science Foundation Presidential Young Investigator, a Fulbright Scholar, a Guggenheim Fellow, an IBM Faculty Development Awardee, and a Fellow of both the ACM and IEEE. He also holds several patents in the fields of external sorting, prediction, and approximate data structures.