If I choose to implement multiple VLANs on a switch connected to a stack of Linux server (RHEL or SLES), do I need to configure VLAN interfaces on each server as well or do I only need to configure the VLANs on the switch?
For example, suppose I have a stack (call it 10) of Linux servers connected by two interfaces each to the same 48-port Gigabit switch. The first NIC on each server belongs to 192.168.10.0/24 and the second to 192.168.20.0/24. Traffic is not to be routed between these networks; both are connected primarily to provide higher overall station -to-station bandwidth. We can't use trunking since the stack may need the full bandwidth of both NICs between two individual nodes rather than between a server and many clients.
Up to this point, I would just connect the cables to the switch and rely on the switch to properly handle the presence of both networks. They do this quite well (Catalyst 4948's, Dell 5448's, SMC 8848's, and so forth) without setting up VLAN tagging. However, I have a minority of partners who are convinced that we require VLAN tagging to make this work properly. Despite having a couple hundred configurations in the field proving otherwise, I've agreed to look into the consequences of configuring the VLANs.
SO, if I do the following:
-setup VLAN 10 and VLAN 20 on the switch
-assign the access mode for the switch ports connected to NICs on the 192.168.10.0/24 network to VLAN 10
-assign the access mode for the switch ports connected to NICs on the 192.168.20.0/24 network to VLAN 20
Do I have to establish the vlan interface aliases with vconfig or ifconfig as well or is configuring the switch enough? If I don't configure the vlan interfaces in the OS, how is the VLAN tag added to the packets?