There's a couple other approaches that we thought about that didn't really work out.
They're listed here!

- Broadcast packets to the whole network, come up with some round robin scheme for Pis
  determined by Paxos (so Pis could go up and down on network) to figure out what to skip
  and what to accept, but Paxos is hard and couldn't think of a good way to determine
  what to drop with low overhead

- Capture packets with NetMap super early, but couldn't recompile any Pi kernel (see paper);
  the goal would be to avoid getting stuff in sk_buff so we didn't have the packet
  already copied in a kernel memory page and get it/map it in userspace instead.
  It's entirely possible that however the close-source Ethernet-to-USB driver on the Pi
  works would shoot this out of the water anyway

- GPU hashing packets, but GPU code is hard and we'd have to get packets to userspace
  to get online implementations to work, but can't really do that w/o copying packet
  as it was already in a sk_buff (see paper)

- Distributing the Pis across multiple, large subnets; i.e. put one in Gates,
  one on north campus, one on west campus, etc., so we could get a REALLY distributed
  honeypot that could capture a wide range of network traffic! We didn't really have
  the time to do this, but it'd be a really cool thing to try