When the "bonding" kernel module is loaded, it will always provide at least one bonding interface, i.e. bond0.
There is a max_bonds=N option for the bonding module. If you specify you want to create N bonds, the module would auto-create bonding interfaces from bond0 to bond(N-1), inclusive. So bond1 cannot exist before bond0 does, and bond2 cannot exist before bond1.
When you configure network interfaces with RedHat configuration tools (kickstart, system-config-network, etc.), those tools will automatically insert the proper option lines for the bonding module to /etc/modprobe.conf for you.
The same naming algorithm is used by the kernel to assign default names for real NICs: the first NIC detected will always be eth0, the next eth1, and so on. However, the "nameif" command can be used to change these names.
Most modern distributions include built-in scripting (either udev-based or included in the network configuration scripts) to save the MAC addresses of each detected NIC and keep the same ethX name assigned to it persistently.
In theory, you could set max_bonds=2 to allocate just bond0 and bond1, then re-name them to bond1 and bond2 before using them... but the nameif command requires you to identify the interface to be renamed by the MAC address, and an unconfigured bonding interface does not *have* a unique MAC address. Also, an interface must not be configured while it's being renamed. Making such a configuration work might be more trouble than it's worth.
If you have a 4-port Ethernet card, and you use only 3 ports of it, the remaining 4th port will also be visible in the "ifconfig -a" listing, and you cannot easily remove it. If some hardware monitoring application sees an unconfigured bond/NIC as an error, a better solution would be to find out how to configure that application to ignore the unconfigured NIC.
MK
MK
