Abstract
Argonne National Laboratory, with facilities in the states of Illinois and Idaho, is owned by the United States government, and operated by The University of Chicago under the provisions of a contract with the Department of Energy.
Introduction
In the late 1990s, the research and education (R&E) community faced a significant architectural challenge: the need to support high-reliability production services alongside high-risk experimental network research. Historically, these two domains were separated into parallel physical infrastructures. However, this approach was both economically inefficient and scientifically limited, as isolated "sandbox" environments failed to replicate the complex traffic dynamics of real-world networks. This paper introduces MORPHnet (Multi-Modal Organizational Research and Production Heterogeneous Network), a framework designed to integrate these disparate requirements into a single, virtualized infrastructure.
The MORPHnet Framework
The core objective of MORPHnet is to allow "crashable" experimental protocols to run concurrently with stable production traffic. This is achieved through the concept of Virtual Production Network Services (VPNS), which utilize virtualization across all layers of the networking stack—from hardware (WDM/SONET) and media (ATM virtual paths) to the network layer (IPv4 vs. IPv6).
The "Production Bar" Concept
A central innovation of the MORPHnet architecture is the "Production Bar." This is a logical boundary that defines the layer at which the network is considered stable for a specific user or application. For a physicist conducting remote supercomputing, the IP layer might be the production base. Conversely, for a network researcher, the IP layer itself may be the experimental subject, relying on a stable underlying ATM or physical layer. This relative definition allows the network to "morph" its service guarantees based on the specific needs of the traffic.
Native Mode Testing
Unlike traditional "tunneling" methods, which encapsulate new protocols within established ones, MORPHnet advocates for native concurrent support. By running experimental protocols directly on the hardware and switches, researchers can accurately measure how new technologies like IPv6 or RSVP (Resource Reservation Protocol) interact with physical hardware and production-level congestion.
Implementation and Trust Models
The paper outlines several deployment scenarios based on the level of trust between administrative domains:
- No Trust: Separate physical switches sharing a common fiber-optic backbone.
- Guarded Trust: Shared switching infrastructure using logical separation, such as ATM Permanent Virtual Circuits (PVCs), to isolate experimental traffic from production flows.
To facilitate this at the endpoint, the authors propose End-System Multihoming. By assigning multiple IP addresses to a single workstation and modifying the operating system kernel, traffic can be tagged as "experimental" or "production" at the process or window level. This allows applications like the Argonne BoilerMaker (collaborative engineering) to operate across both environments simultaneously.
Insights and Discussion
The authors argue that traditional Poisson modeling, used in telephony, is insufficient for modern data networks. Real-world traffic is self-similar and bursty, necessitating testing on live, congested infrastructure rather than isolated testbeds (the "Bonneville Salt Flats" model). Furthermore, the paper envisions a transition toward Active Networking, where applications can dynamically request Quality of Service (QoS) parameters or reconfigure network elements via software APIs.
Conclusion
The MORPHnet architecture, validated in part by the success of the I-WAY project, provides a roadmap for transitioning from legacy protocols to next-generation networking without sacrificing reliability. While 1997-era hardware limitations—specifically in dynamic ATM signaling and non-intrusive monitoring—posed significant hurdles, the framework established the principles of network virtualization and multi-modal operation that remain relevant in the era of Software-Defined Networking (SDN).
Summary updated