From: Rick Jones <redacted>
Date: Wed, 06 Jun 2007 16:48:59 -0700

  > RX queues - yes, I can see;  TX queues, it doesnt make sense to put

quoted

different rings on different CPUs.

To what extent might that preclude some cachelines bouncing hither and 
yon between the CPUs?

I think per-TX-queue locking takes locality as another advantage.
You only touch the TX descriptors for queue N, rather than a single
globally shared one.

Same goes for RX.