Back to ACN index

• The end-to-end argument is discussed.
• Packet vs circuit switching: packet switching is more efficient if used bandwidth varies, and in practice this is the case (computer networking applications can be quite bursty). Additionally, packet switching allows for connection preservation even when links fail, by routing packets along a different path when failure is detected -- in the circuit switching case, this may require connection teardown and setup.
• A number of data structures are discussed; mostly probabilistic ones.
  • The reason for probabilistic d.s.: resources are limited (both memory and compute), so we want very fast lookup using very little space (in the common case).
  • Bloom filters: used for testing set membership probabilistically.
  • CountMin sketches: used to give an upper bound on counts, e.g. a number of matching packets.