Computer network architecture and software engineering

Yan Wang, Jiazhi Zeng, Ming Liu, Yue Wu, Haigang GongBROADNETS 2009

Abstract

Traditional computer network architectures, such as the OSI and TCP/IP models, rely heavily on layered structures that often introduce functional redundancy and high overhead. This paper proposes a paradigm shift by applying formal Software Engineering (SE) methodologies and Shannon’s Communication Model to network design. By treating network architecture as a large-scale software problem, the researchers introduce "Shannon Architecture" and "SUNA" (Service Unit Based Network Architecture). These modular frameworks aim to streamline communication by replacing rigid layers with functional modules, thereby enhancing scalability and efficiency.

Introduction: The Limitations of Layered Architectures

The modern Internet is increasingly burdened by demands for enhanced security, Quality of Service (QoS), and mobility—features that the original TCP/IP stack was not primarily designed to handle. The authors argue that the industry's reliance on layered architectures is a fundamental flaw. While layering was intended to simplify complex systems, it has resulted in several critical issues:

  • Functional Redundancy: Tasks like error checking and addressing are frequently repeated across multiple layers, wasting computational resources.
  • High Overhead: The accumulation of headers from various layers can consume significant bandwidth, sometimes reaching up to 58 bytes per packet.
  • Poor Scalability: Fixed layer counts make it difficult to integrate new features without modifying the entire stack.
  • Complexity: Interfaces between layers are often non-standard and difficult to comprehend.

Quoting Andrew Tanenbaum, the authors describe layering as a "bandit" that forces users to accept inefficient standards to maintain the stack's integrity.

Methodology: From Communication to Software Engineering

The researchers utilize Structured Analysis (SA), a classic software engineering technique, to re-evaluate network design. Their approach begins by mapping communication terminology to SE concepts, effectively creating a "Rosetta Stone" for network architects.

A core component of their methodology is the expansion of Claude Shannon’s 1948 Communication Model. Shannon's original model describes a point-to-point communication system consisting of an information source, transmitter, channel, receiver, and destination.

Shannon’s Communication Model serves as the foundational framework for the proposed architecture.

The researchers expand this model from a single pair to a multi-user network environment, creating a General Network Model. This model incorporates multiple end systems connected via a channel consisting of various switches and media, while also accounting for noise sources.

The General Network Model expands Shannon's model to a multi-user environment with multiple end systems and a complex channel.

Proposed Architectures: Shannon Architecture and SUNA

Instead of traditional layers, the paper proposes breaking the network into functional modules. Two primary architectures are introduced:

  1. Shannon Architecture: A modularized version of the general network model that treats communication functions as distinct software modules.
  2. SUNA (Service Unit Based Network Architecture): A more granular approach where the "Transmitter" and "Receiver" are decomposed into specific "Service Units." These units handle specialized tasks such as encryption, compression, and digital signatures only when required, reducing unnecessary processing.

Furthermore, the authors propose replacing traditional routing tables with "Source Trees." This compact representation of network paths can potentially store path information using roughly half the memory required by standard tables.

Key Insights and Results

The central insight of this research is that a computer network is essentially a large-scale software system. By applying rigorous software requirement analysis, the "spaghetti" of legacy protocols can be untangled. The authors distinguish between "Functional Modules" (which perform a single, well-defined task) and "Contingency Modules" (where functions are grouped arbitrarily by layer). Shannon Architecture prioritizes Functional Modules to create cleaner interfaces.

To validate these concepts, the authors developed a prototype of SUNA in a Linux environment. The prototype successfully supported standard Internet applications, including FTP, E-mail, and Web traffic. The analysis suggests that this modular approach can significantly reduce header overhead by only employing the necessary service units for a given transmission.

Conclusion and Future Work

This research provides a formal SE-based definition of network architecture, demonstrating how modularization can eliminate redundancy and improve efficiency. While the paper focuses heavily on the requirement analysis phase, it sets the stage for future development.

Future work includes completing the remaining phases of the Software Engineering life cycle—design, implementation, and testing—to create specialized operating systems for routers and end systems. Additionally, further investigation is needed to determine how these modular architectures can natively support guaranteed QoS for real-time data streams, moving beyond the "best effort" limitations of current networks.

Summary updated

BibTeX
@inproceedings{wang2009computer,
  author = {Yan Wang and Jiazhi Zeng and Ming Liu and Yue Wu and Haigang Gong},
  title = {{Computer network architecture and software engineering}},
  booktitle = {BROADNETS},
  year = {2009},
}
Figure