Annual Review of Scalable Computing

The following sections are included:

• EDITORIAL BOARD

• Preface

• Contents

The following sections are included:

• Introduction

• CCS Architecture

• Resource and Service Description

• Site Management

• Related Work

• Summary

• Acknowledgments

• Bibliography

This paper presents a motivation and issues concerning the development and adaptation of on-line tools to the OCM, an OMIS Compliant Monitoring system. It covers the general structure of an OCM-based tool environment and other enhancements needed both at low-level parts of the monitoring environment and at the user interface level in order to achieve full tool support for parallel applications. A high demand for tools support for parallel programming, which comprise several classes of tools, e.g. debuggers, performance analyzers, visualizers, etc., reveals a strong need in co-operation of tools which thus should constitute a well designed tool environment. This should be based on the monitoring of parallel applications developed using different programming models and offer a wide range of tools capable of cooperating with each other. This paper presents the recent development of the OCM-based environment of on-line tools and outlines perspectives for further research.

This paper presents a Java framework for high performance computing (HPC) on networks of workstations (NOWs). We introduce a new lottery-based work stealing algorithm for efficient scheduling of large-scale multithreaded computations on NOWs. In the proposed algorithm, idle workstations actively search out work to do rather than wait for work to be assigned. In the lottery game, each workstation is equipped with a set of tickets and the number of tickets is proportional to the age of the oldest thread in the ready pool of the workstation. A winning ticket is drawn at random and the workstation with the winning ticket becomes the victim from which the idle workstation steals work. The proposed selection procedure serves for two purposes. First, we try to lower communication costs by stealing large amounts of work, with the logic behind being that old-aged computations are likely to spawn more work than relatively young computations. Second, we would like to bias the search to obtain favourable results while at the same time avoid system bottleneck. Our approach has been implemented and tested on NOWs under the Solaris OS. Several examples have been used to demonstrate the potential performance gain.

Advances in wireless networking technology and portable computing devices have led to the emergence of the mobile computing paradigm. As a result, the traditional notion of timely and reliable access to global information sources in a distributed system or multidatabase system is rapidly changing. Users have become much more demanding in that they desire and sometimes even require access to information anytime, anywhere. The amount and the diversity of information that is accessible to a user are also growing at an exponential rate. Compounding the access to information is the wide variety of technologies with differing memory, network, power, and display requirements. Within the scope of distributed databases and multidatabases, the issue of concurrency control as a means to provide timely and reliable access to the information sources has been studied in detail. In an MDAS environment, concurrent execution of transaction is more problematic due to the power and resource limitations of computing devices and lower communication rate of the wireless communication medium. This article is intended to address the issue of concurrency control within a multidatabase and an MDAS environment. Similarities and differences of multidatabases and MDAS environment are discussed. Transaction processing and concurrency control issues are analyzed. Finally, a taxonomy for concurrency control algorithms for both multidatabases and MDAS environments is introduced.

The following sections are included:

• Introduction

• Concurrency

• Data Parallelism

• Memory Consistency

• Tuple Locks

• Using Tuple Locks in Parallel Algorithms

• Tuples in Objects

• Towards Architecture Inclusive Parallel Programming

• Acknowledgement