What problem does the architecture of the input/output queuing and shared buffers solve on Catalyst 5000 switches?
It avoids head-of-the-line blocking.