Abstract
Due to the complex coupling between physical and information spaces, establishing information associations for dynamic services has become a critical challenge for future Internet development. Traditional Internet theories struggle to meet these demands. This paper presents a Service Fusion Engine architecture designed for physical world information. By constructing an ontology-based environment, the system addresses the uncertainty and high hybridity of physical data. Leveraging physical information characteristics for search and user behavior prediction, the engine enables personalized service composition through a combination of ontology knowledge and inference rules.
Introduction: The Challenge of the Physical World
The evolution of the Internet is shifting from a human-computer dual model toward a "human-computer-thing" trinity, often referred to as the Information of Physical World (IOP). This transition introduces massive volumes of heterogeneous data that current architectures are ill-equipped to manage. Key challenges include:
- Heterogeneity: Diverse physical devices lack interoperability.
- Resource Intensity: High computational costs for protocol and data format conversion.
- Search Inefficiency: Standard tools fail to index real-time, dynamic physical information.
- Lack of Intelligence: Services often lack the context-awareness needed to predict user needs.
Proposed Architecture: The Service Fusion Engine
To address these issues, the authors propose a four-part architecture centered on a "Service Fusion Engine." This engine acts as a bridge, transforming raw sensor data into actionable, personalized services.
The architecture consists of several core components:
- Unified Characterization: Utilizing ontologies to standardize the description of heterogeneous physical nodes, allowing for reusable temporal and spatial data mapping.
- IOP Vertical Search Engine: A specialized engine that uses gateways and adapters to crawl dynamic information, building a multi-level distributed index.
- Ontology and Service Reasoning: The system's "brain" combines deterministic logic (Rules Lab) with Bayesian networks to manage uncertainty and predict user behavior.
Implementation and Tech Stack
The implementation follows a multi-view approach—Infrastructure, Software, and Data—to manage the complexities of big data in the IOP.
The tech stack leverages modern big data tools:
- Storage: Greenplum for structured data and HBase for NoSQL requirements.
- Processing: A combination of MapReduce for batch processing and Storm for real-time stream processing.
- Semantic Framework: The SIA-MAP ontology is used to label raw sensor data, creating "Virtual Sensors" and "Sensor Groups" within a resource pool.
Results and Prototype
The researchers validated their architecture through a prototype focused on smart domestic equipment monitoring. Key achievements include:
- Data Blending: Successful processing of mixed structured and NoSQL data using the Storm/MapReduce framework.
- Semantic Interoperability: Raw sensor feeds were effectively converted into virtualized services that applications could access via standard web servers.
- Automated Service Delivery: The platform demonstrated efficient "Pub-Sub" scheduling, matching real-time physical states to specific user demands.
Limitations and Future Work
While the prototype is promising, the paper lacks extensive quantitative metrics regarding latency, prediction accuracy, and large-scale scalability. Future research will focus on:
- Developing an open sensor cloud ecosystem for cross-industry collaboration.
- Refining spatiotemporal correlation algorithms for highly dynamic environments.
- Optimizing information release for resource-constrained embedded devices.
Summary updated