Do we want ifenslave?

[Rob Landley]

There's a feature request for ifenslave (with patch, even, although it's only 
lightly busyboxified and would need a lot of clean-up work).  My question is: 
do we want this?

First, ifenslave isn't hard to build standalone.  It's a standalone C file in 
the kernel sources, and shouldn't be that big.

Secondly, what it does is kind of a niche.  When I thumped on this in 2000, it 
only worked with a very small subset of the switches out there, and was only 
really interesting at the 100baseT level.  Linking two interfaces together 
does indeed give you twice the speed: but gigabit ethernet gives you 10 times 
as much without taking up two ports.  (Maybe these days you can link together 
two gigabit interfaces, except that 10gigE is out now for people who care.  
The only reason we were looking at bonding was because a second interface was 
there in the boxes anyway, going unused.  Buying twice as many switches 
turned out to make it cost prohibitive anyway.)

If it's started providing failover or something since last I checked, that's 
one thing.  But I'm unaware of it if so.  Unplugging a cable used to mean you 
lost every second packet, which wasn't much fun...

Then again, somebody does seem to want it.  I'd email the guy who submitted 
the bug, but this is the bug system we're talking about.  Having a way to 
contact the submitter would be far too useful to include.

Rob