A Novel Agent Based Approach for Controlling Network Storms

TL;DR

This paper introduces an agent-based control strategy for preventing broadcast storms in Ethernet LANs by distinguishing normal from storm packet growth patterns using empirical models.

Contribution

It proposes a novel multi-agent approach that inhibits storm packet regeneration based on empirical pattern recognition, improving network stability.

Findings

Effective differentiation between normal and storm packet growth patterns.

Reduction in broadcast storm occurrences through agent-based control.

Enhanced network performance and reliability.

Abstract

One of the fundamental data transmission mechanisms in Ethernet LAN is broadcasting. Flooding is a direct broadcasting technique used in these networks. A significant drawback of this method is that it can lead to broadcast storms. This phenomenon is more common in multivendor switch environment. Broadcast storms usually results in dissension, collision and redundancy leading to degradation of the network performance. Most of the storms appear without much warning and it affects the efficiency of network even in situations when the network is expected to work most efficiently. There are several characteristic patterns by which storm can appear in a LAN, like rate monotonic repetition, transient appearances with different types of growth properties and decay profiles. In this paper we discuss the storm build up pattern in an industry and present various reasons for storm in LAN. We have identified a strategy for controlling network storms, using multiple static agents. These agents inhibit storm packet regeneration in the network using the knowledge of storm growth pattern. A model developed out of empirical studies is used to differentiate normal packet growth from storm packet growth and used in control mechanism of storms.