Using realistic factors to simulate catastrophic congestion events in a network

Area of Science:

• Computer Science
• Network Engineering
• Network Science

Background:

• Internet growth necessitates research into communication network dynamics.
• Abstract graph models are common for studying network congestion and phase transitions.
• Previous models often oversimplify critical network factors like router speeds and buffer limits.

Purpose of the Study:

• To develop and analyze a more realistic computer network model.
• To investigate the impact of realistic network factors on catastrophic failures.
• To assess the mitigation of network congestion and phase transitions.

Main Methods:

• Modified a graph model of an Internet Service Provider (ISP) network with over 250,000 nodes.
• Incorporated realistic factors: router classes, variable router speeds, flows, Transmission Control Protocol (TCP), sources/receivers, and packet dropping.
• Gradually increased network load for various combinations of realistic factors to observe congestion spread.

Main Results:

• The addition of realistic network factors, specifically variable router speeds or TCP, prevented a phase transition to a fully congested state.
• Increased network load, even with realistic factors, ultimately led to performance decline and reduced node communication.
• Catastrophic failures were mitigated, but network operation still degraded under high load.

Conclusions:

• Realistic network factors are crucial for accurate modeling of computer communication networks.
• Models incorporating factors like variable router speeds and TCP can prevent abrupt, catastrophic network failures.
• Findings suggest caution when using abstract models for systems prone to phase transitions, like power grids or disease spread.