TL;DR
This paper introduces a novel frame-based system for monitoring large computer networks. It tackles the challenge of visualizing vast amounts of data by creating an "analogy" between the network's structure and a flexible graphics system, allowing for intelligent abstraction and presentation of information on a screen.
What Problem Does This Paper Address?
Monitoring large computer networks generates an overwhelming amount of data. Presenting this data effectively on a limited-resolution screen is a significant challenge. Traditional methods often struggle to provide both a high-level overview and the ability to drill down into specific details without overwhelming the user. This paper addresses the need for an intelligent system that can dynamically abstract and display network information, making it manageable and comprehensible for operators and administrators.
Methods
The core of the system is built using FRL (Frame Representation Language), a system that allows for representing knowledge in a frame-based structure. The authors developed two main frame models:
- Network Domain Model: This model captures information about the computer network, including its topology, operational status, and performance metrics. It uses FRL's demon mechanisms to automatically compute higher-level abstractions from raw data.
- Graphics Domain Model: This is a general-purpose graphics system that uses hierarchical regions and a constraint-based structure. It supports features like "intelligent pan and zoom," allowing users to navigate and explore graphical representations.
The key innovation is the analogy creation and maintenance mechanism that links these two domains. This analogy acts as a homomorphic correspondence, meaning that changes in the network domain are reflected in the graphics domain. This allows for the creation of various graphic descriptions (e.g., Cartesian graphs, tables, flow diagrams) that can be dynamically instantiated based on user needs, screen real estate, and desired level of detail. Heuristics are employed to control the selection and presentation of these descriptions.
Datasets
Not discussed in the paper. The paper describes a system designed to monitor a token-based ring network of approximately 50 Prime 750 systems.
Key Insights and Ideas
- Analogy as a Unifying Principle: The paper highlights the power of using an "analogy" to bridge the gap between a complex application domain (computer networks) and a flexible presentation domain (graphics). This allows for independent development of each system and their subsequent coupling.
- Intelligent Abstraction for Limited Displays: The system's ability to dynamically abstract and select appropriate graphic representations based on heuristics is crucial for managing information overload on limited screen space.
- Domain-Independent and Domain-Dependent Heuristics: The authors emphasize the importance of using a combination of general heuristics (e.g., "uniform level hypothesis") and domain-specific knowledge to guide the selection of viewable descriptions.
- Modularity and Reusability: The frame-based approach, coupled with the analogy mechanism, promotes modularity, allowing for easier extension and modification of both the network model and the graphics representations.
Key Results
The paper describes a functional prototype of the network monitor running on five machines. The graphics system is complete and has been tested on various small domains. Performance data is currently gathered and processed. The authors note that the ultimate usefulness of their heuristics will require testing with larger datasets.
Limitations and Future Work
- Scalability of Description Generation: The authors acknowledge that simply enumerating all possible graphic descriptions can be combinatorially explosive. They rely on heuristics to manage this, but further work is needed to fully ascertain their effectiveness with larger datasets.
- FRL Implementation Performance: The current FRL implementation is described as needing performance improvements for truly usable results, although it is acceptable for the current small model.
- Localization of Naming Scopes: The lack of clear naming scope localization in FRL can lead to name conflicts when combining frame systems, which the authors plan to address.
- Knowledge vs. Meta-Knowledge Distinction: Distinguishing between knowledge and meta-knowledge in FRL is identified as a troublesome aspect that needs further exploration, particularly for building systems that control the construction of other frames.
- Future Work: The authors express interest in applying their system to network control (not just monitoring) and integrating ideas from expert systems and pattern matching. They also aim to improve the FRL implementation and address the limitations mentioned above.
Summary updated