Incorporating prediction models in the SelfLet framework: a plugin approach

TL;DR

This paper explores integrating prediction models into the SelfLet autonomic framework to enable dynamic, plug-and-play predictive capabilities for managing complex, distributed pervasive systems.

Contribution

It extends the SelfLet framework to support dynamic integration of various prediction models, demonstrating feasibility in decentralized, cooperative environments.

Findings

Prediction models can be effectively integrated into SelfLet.

The extended framework supports dynamic plugging and unplugging of models.

A simple example illustrates the system's operation with a specific prediction model.

Abstract

A complex pervasive system is typically composed of many cooperating \emph{nodes}, running on machines with different capabilities, and pervasively distributed across the environment. These systems pose several new challenges such as the need for the nodes to manage autonomously and dynamically in order to adapt to changes detected in the environment. To address the above issue, a number of autonomic frameworks has been proposed. These usually offer either predefined self-management policies or programmatic mechanisms for creating new policies at design time. From a more theoretical perspective, some works propose the adoption of prediction models as a way to anticipate the evolution of the system and to make timely decisions. In this context, our aim is to experiment with the integration of prediction models within a specific autonomic framework in order to assess the feasibility of such integration in a setting where the characteristics of dynamicity, decentralization, and cooperation among nodes are important. We extend an existing infrastructure called \emph{SelfLets} in order to make it ready to host various prediction models that can be dynamically plugged and unplugged in the various component nodes, thus enabling a wide range of predictions to be performed. Also, we show in a simple example how the system works when adopting a specific prediction model from the literature.