How to Avoid a Split-Brain Scenario with Cisco Switches by Enabling Multicast Communication

Updated

Red Hat Enterprise Linux 5+ and Red Hat Enterprise Linux 6+ use multicast for cluster communication by default, and multicast can be enabled in Red Hat Enterprise Linux 4+. Certain Cisco switches, such as the Catalyst line, come with IGMP snooping enabled by default and will not forward multicast traffic without an IGMP querier to determine the intended recipients. In this case, nodes will not receive traffic being sent by other cluster members during bootup, which might result in a non-quorate cluster or a split-brain scenario.

One simple solution is to use broadcast in the cluster. Broadcast mode can be enabled in Red Hat Enterprise Linux 4 from system-config-cluster or Conga. However, because multicast has several advantages over broadcast and Red Hat Enterprise Linux 5 must use multicast, broadcast is not recommended.

This document details methods for enabling multicast communication between Red Hat Enterprise Linux cluster nodes and Cisco Catalyst switches to prevent a split-brain scenario.

Article Type