Introduction:

With Red Hat Enterprise Linux version 6.1, Red Hat started using a feature they refer to as Consistent Network Device Naming.  This is provided as a solution to two problems:  First, Linux has a tendency to reassign names to network devices when new devices are added, based on either MAC address or the order in which udev finds the hardware.  Second, the actual device name (eth0, eth1, etc) lacks any descriptive quality.  This new system will provide a consistent name to a network device based on how it is connected to the system, and the name itself will describe that connection.

Applies To:

This feature is based on data obtained by the operating system from BIOS, specifically, the type 9 and type 41 fields of SMBIOS.  If the system does not support those fields or provide SMBIOS version 2.6 or higher, this feature will not work. All of our servers since we started purchasing Dell PowerEdge R610s should support this field.  By default, this feature is enabled on new (Generation 11 and up) Dell PowerEdge servers.  It is disabled on all other servers.

Naming Convention:

Built-in Network connections:  em[1,2,3,4]
PCI Card connections: p[slot]p[ethernet port] (For example, a 2 port card in slot 1 will provide NICs p1p1 and p1p2).
This behavior can be overridden in a few ways: udev rules can still be applied to override the default assignments, the package 'biosdevname' is a requirement for the feature, and entering "biosdevname=0" as an option on the boot command line should also disable the feature.

References:

Redhat.com: Consistent Network Device Naming
Dell.com: Consistent Network Device Naming in Linux (PDF)