Naslov
Online Monitoring in Software DSM Systems

Autori
Zoraja, Ivan

Vrsta, podvrsta i kategorija rada
Ocjenski radovi, doktorska disertacija

Fakultet
Fakultät für Informatik

Mjesto
München

Datum
30.05

Godina
2000

Stranica
162

Mentor
Ludwig, Thomas

Neposredni voditelj
Arndt Bode, Vaidy Sunderam

Ključne riječi
Distributed Shared Memory; online monitoring; DSM

Sažetak
This thesis describes and evaluates issues that come up in the design and implementation of online monitoring systems which connect parallel tools to running applications based on software realisations of the DSM programming paradigm. The main motivation behind this thesis is the intention to develop monitoring systems which can provide parallel tools with an abstraction of shared memory via the transparent management of underlying resources that realise DSM mechanisms. In addition, those monitoring systems have to be able to support multiple tools that, via the monitoring systems, can perform consistent and yet efficient operations on the objects being monitored. As a first step, the shortages and shortcomings of the current DSM tools and monitoring, including the structure of the monitoring environment, the system functionality, and its software architecture are specified. The monitoring environment encompasses the objects being monitored and their actors. The monitoring functionality specifies operations which can be invoked on the target system, while the software architecture describes the software components of the system and their responsibilities and interaction patterns. Performance, portability, adaptability, and extensibility are recognised as important system feature that have to be considered when building software systems. The software architecture for online monitoring is based on a client-server architecture, an event-action model, and layering. Three logical layers, which can be implemented through the interaction of several components, have been suggested. The first layer manages parallelism, distribution, and consistency. The second layer provides the basic monitoring functionality based on the event-action paradigm, while the third one transparently handles DSM mechanisms dealing with complex dependencies that exits among DSM activities. For instance, it is able to destroy and rebuild shared resources including shared data and communication routes during process migration and checkpointing. During the course of this research, several important issues concernig the development of online monitoring systems have been understood, and an approach upon which to construct general DSM monitoring systems has been proposed. To validate the design and implementations choices, an experimental monitoring prototype, aimed at a selected DSM library, has been implemented. Also, its effectiveness in terms of functionality, performance, and other previously defined requirements has been evaluated. Particular emphasis is placed on the implementation of monitoring functionality for process migration and checkpointing, since in both cases, DSM monitoring systems have to implicitly deal with complex interaction patterns among DSM activities.

Izvorni jezik
Engleski

Znanstvena područja
Računarstvo

POVEZANOST RADA