Overview

The goal of the Delphoi is provide adaptive components to other modules of the Grid Application Toolkit (GAT). Common to these adaptive components is that they take (possibly dynamic) information about Grid resources (network status, CPU availability, memory footprint, etc.) as input to performance-prediction models and implement adaptive strategies that let applications use the given resources more efficiently. For grid-aware applications, this general mechanism is used for numerous purposes like data access, task scheduling, and dynamic application reconfiguration. Although use-case specific APIs and event models are necessary, the basic techniques are similar.

In the Delphoi, we generalize the basic techniques of gathering relevant monitoring data (e.g., provided by Mercury) and correlating this data with performance models to provide short-term performance predictions. These predictions can be used to adapt the behavior of applications, or as input to other services (e.g., GRMS).

This picture shows the main components of our information gathering software. A large number of components (in the core layer) are distributed over the testbed. These components, called Pythias, are responsable for gathering local information. The Pythias locally store this information for later use.

The Delphoi is a centralized component in the service layer. Using a web service interface, it allows the retrieval of historic data and short term predictions from the Pythias. It also provides detailed performance related data to several other services.

For more information see: