Abstract
This paper presents an architecture for an Emergent Software Service Platform designed to autonomously discover, compose, and execute software services at runtime. By conceptualizing service composition as a classical AI planning problem within a shared semantic domain, the platform enables complex IoT ecosystems to adapt to dynamic environments without manual design or centralized coordination. The feasibility of this approach is demonstrated through a smart mobility prototype.
Introduction
In the evolving landscape of the Internet of Things (IoT), software systems are increasingly characterized as "conglomerates" of heterogeneous services distributed across diverse hardware. Traditionally, these systems are designed using a top-down, hierarchical approach. However, as environments become more dynamic—such as in a smart city where services frequently appear and disappear—this manual design process becomes a significant bottleneck.
Current self-adaptive systems often rely on a central logical entity or a standardization body to define component interactions. This restriction limits "emergence," which the authors define as the system's ability to autonomously create new, higher-value functionality from available parts at runtime. The core challenge addressed in this research is building a platform capable of understanding user requirements, discovering appropriate services, and orchestrating them automatically without prior manual configuration.
The Emergent Platform Architecture
To facilitate emergent behavior, the authors propose a five-part architecture that treats service composition as a classical planning problem within a shared semantic domain.
The architecture consists of the following key components:
- Requirements Handler: This component elicits user needs either explicitly through a user interface or implicitly by monitoring sensor data. It translates these unformalized desires into a formalized goal format.
- Domain: A central ontology providing the foundational vocabulary used to describe both user requirements and service capabilities.
- Service Registry: A library where providers register Software Service Descriptions (the "what" and "how") linked to actual Service Instances.
- Self-adaptive Composition Mechanism: Acting as the system's "brain," this mechanism treats the user's requirement as a goal in a planning problem. It searches the registry for semantically matching services and determines the optimal sequence (the "plan") to fulfill the goal.
- Execution Engine: This component executes the composed sequence using a "meta-flow" concept, wiring individual service nodes together to transfer data and trigger real-world actions.
Implementation: Smart Mobility Case Study
The platform's viability was evaluated using a custom Smart Mobility scenario centered on a smart parking lot. The prototype included services such as parking spot reservation, electric vehicle (EV) charging, car washing, tire pressure measurement, and navigation. These services were described using OpenAPI specifications and implemented as REST endpoints.
Users interact with the system via a "Request Configurator" (as seen in Figure 3), where they can select desired services. The system then automatically generates a formalized composition result—a JSON-based plan—involving the necessary services to meet the request.
Key Insights and Results
The research yields several significant insights:
- Composition as Planning: Framing service composition as an AI planning problem allows the system to find a path from an initial state to a goal state without human intervention.
- Semantic Matching: By utilizing a shared "Domain" vocabulary, the platform avoids the need for hard-coded compatibility rules between every possible pair of services.
- Runtime Emergence: The prototype successfully demonstrated the ability to compose services that were not predefined at design time, proving the feasibility of automatic requirement elicitation, composition, and execution.
Limitations and Future Outlook
While the prototype is a successful proof of concept, several challenges remain for large-scale deployment. The current planning mechanism's performance must be tested as the number of services scales into the thousands. Additionally, the authors plan to incorporate Natural Language Processing (NLP) to allow for more intuitive user interaction via speech or text.
Robustness is another area for improvement; the current system lacks a "rollback" mechanism to handle service failures during execution. Finally, future iterations will need to incorporate real-time service availability checks during the planning phase to ensure that the generated plans are executable in the current environment.
Summary updated