Abstract
SMIng (next generation structure of management information), an information model for network management, is a prospective structure of management information. When deploying multi-agent systems to network management environments, we have established a lightweight self-contained knowledge model based on RDF (Resource Description Framework) and its extensions. We also present an implementation prototype to support agent communication and coordination by RDF-based languages.
Introduction
In the early 2000s, network management transitioned from static monitoring toward the use of intelligent agents—autonomous software entities capable of complex tasks. However, these agents were often hindered by a "language barrier." Traditional frameworks like SNMP (Simple Network Management Protocol) utilized a rigid Structure of Management Information (SMI) that, while efficient for reading variables, lacked the capacity to describe complex relationships or the semantic meaning of data. While agents could communicate via protocols like KQML, their interactions were limited to basic attribute-value pairs, lacking a shared formal knowledge base for reasoning or coordination.
Methodology: Bridging SMIng and RDF
To address these limitations, this research proposes a hybrid approach that upgrades traditional network information models into a semantic framework. The core of this methodology involves mapping the SMIng (Next Generation Structure of Management Information) model to RDF Schema (RDFS). SMIng serves as an ASN.1-independent "seed," allowing network definitions to be transformed into formal classes and properties.
By representing network data as RDF triples (subject-predicate-object), the system enables flexible descriptions of resources and their interrelationships. To empower agents with reasoning capabilities, the framework incorporates the Ontology Inference Layer (OIL) and FIPA-RDF. This allows agents to handle complex rules and perform SQL-like RDF queries for data selection and manipulation.
Architecture and Implementation
The implementation model follows a layered architecture that separates transport, agent communication, and semantic processing.
As shown in Figure 1, the system integrates several key components:
- Transport Layer: Utilizes standard protocols like TCP/IP, SNMP, and JMX (Java Management Extensions).
- Agent Toolkit: Employs JatLite for agent communication and MCT/RMI for mobile code transport.
- Semantic Layer: Features an RDF API and SMI Ontology to manage the Management Knowledge Base (MKB).
- Parsing and Processing: Uses XML tools like SAX and Xerces, alongside the Jena framework for RDF processing.
In this prototype, KQML acts as the "envelope" for messages, while the content is expressed in RDF-based languages, enabling agents to understand explicit semantics rather than just raw data.
Key Insights and Results
The transition from SMI to RDF represents a shift from simple data retrieval to knowledge management. Instead of merely querying the value of a variable, agents can now understand how a specific resource relates to broader network services. The use of RDF reification further allows for the inclusion of metadata, such as timestamps (sming:time), which is essential for managing dynamic network states.
The prototype successfully demonstrated that:
- Formalization is Feasible: Scalar variables and table structures from MIBs (Management Information Bases) can be effectively represented in nested RDFS forms.
- Enhanced Coordination: Agents can perform "added-value" functions, such as inference and coordination, which were previously unsupported by traditional frameworks.
- Tool Interoperability: Existing Semantic Web tools (e.g., Jena, SiRPAC) can be integrated with standard network management tools (e.g., JMX) to create a functional MKB.
Limitations and Future Work
While the prototype proves the concept, it remains an early-stage implementation. Future work must address the increasing complexity of real-world management knowledge models. Additionally, the performance overhead associated with parsing XML/RDF—compared to the lightweight Basic Encoding Rules (BER) used in SNMP—remains a challenge for high-speed network environments. Subsequent research should focus on optimizing these tools to coordinate agents more effectively across diverse and large-scale network infrastructures.
Summary updated