Abstract

The extensibility and evolution of network services and protocols had become a major research issue in recent years. The ‘programmable’ and ‘active’ network paradigms have been trying to solve the problems emanating from the immutable organization of network software layers by allowing arbitrary custom codes to be embedded inside network layers. In this work, we propose a new approach for building extensible network systems to support cross-layer optimization. The fundamental idea is to perform a simple, light-weight meta-engineering on the classical OSI protocols’ organization to make it interactive and transparent. The protocols become (interactive) since they can provide event notification to service subscribers, and they become (transparent) since they also allow controlled access to their state information. Actual protocol extensions (or modifications) can then be performed at the application space by what we call Transientware Modules. This organization provides the infrastructure needed for easy and practical extensions of the current network services and it becomes much easier to address other difficult issues like security and flexibility. We call this mechanism Interactive Transparent Networking (InTraN) and we call the extended kernel InTraN-enabled. We have realized a FreeBSD implementation of the extensible InTraN-enabled kernel. In this paper, we present a formal EFSM-based model for the proposed meta-engineering and illustrate the principles through a real example of TCP extension. Then, we demonstrate how it can be used to realize equivalents of other protocol modifications by showing the InTraN model of ‘Snoop’ [H. Balakrishnan, S. Seshan, R. Katz, Improving reliable transport and handoff performance in cellular wireless networks, ACM Wireless Networks 1 (1995)].

Keywords: Interactive Transparent Networking; TCP; EFSM; Active networks; Interactive TCP; Protocol meta-engineering

Article Outline

1. Introduction

2. Interactive Transparent Networking (InTraN)

2.1. Background

2.2. A framework for the InTraN paradigm

2.2.1. Components and architecture

2.2.2. SP–SM interfacing: subscription mechanism

2.2.3. TM–SM–PE interfacing: access mechanism

2.2.4. Protocol meta-engineering

2.2.5. Security model

3. Interactive-TCP (iTCP)

3.1. The SDL model

3.2. EFSM of iTCP

3.3. The TMs

3.4. Performance

4. Protocol modeling example

4.1. “Snoop”

4.2. ‘Snoop’ variants and InTraN

5. Performance issues

5.1. Overhead cost

5.1.1. CPU time analysis

5.1.2. Context switching analysis

5.2. Security and practice

6. Concluding remarks

Acknowledgements

Appendix A. Appendix

References

Vitae