workflow and repository orginization for tracking router configs

I need to track router/firewall configs for several hundred to a few thousand devices.

For each device the config is a single file.

Each device is updated independantly of the others.

The configs diff well not only against prior versions but against each other.

The different devices are accessable (in many cases via serial ports) from different boxes (in different datacenters in many cases)

I need to data to be replicated across many locations

I want to be able to easily/quickly track changes to a particular device, and compare different devices against each other.

I am thinking that a DVCS like git may be a good fit for this situation. I can do commits on the boxes that can directly access the devices, and then do fetches/pulls between boxes to make sure that every box has the full history of every device.

The question is, what is the best way to orginize this data.

One approach is to have a seperate branch for each device.

   This has the advantage that each branch can have a new config committed 
independantly of the others, making pulls/fetches trivial

Another approach is to have one branch with a different file for each device.

   This is easy for people to understand, but it seems to me that it would 
make pulling from different boxes that updated different files an issue, 
as well as adding confustion over which commit is related to which device.

Any thoughts or suggestions on what people think would work well? (both from a usability and from an efficiancy point of view)

David Lang