Networking Named Content

Jacobson et al

content based addressing networking routing internet infrastructure

@conference{jacobson:conext-2009,
  title={Networking Named Content},
  author={Jacobson, V. and Smetters, D.K. and Thornton, J.D. and
          Plass, M.F. and Briggs, N.H. and Braynard, R.L.},
  booktitle={Proceedings of the 5th International Conference on
             Emerging Networking Experiments and Technologies},
  year={2009},
  organization={ACM}
}

Modern networking dominated by content distribution

• Original use and design focused on connecting particular endpoints
• Use now focused on retrieving content, design hasn't changed
• "People value the Internet for what it contains, but communication is still in terms of where.

By more directly operating over content, can improve availability, security, and location-dependence.

Original IP design has many desirable qualities

• Simplicity
• Weak demands placed on layer 2
  • Stateless, unreliable, unordered, best-effort

CCN addresses content directly

• Interest packets matched with Data packets/chunks via prefix matching on hierachical names
• Content may not exist yet, Interest is thus subscribing to it, or it may be generated based on that interest
• Names may be context dependent, e.g., /ThisRoom/projector
  • Doesn't discuss how that might be controlled; presumably strategy rules
• Names include simple directives for relative addressing
  • E.g., leftmost right descendent of a given prefix
  • Used to refer to sequences of chunks, i.e., when the next name is unknown
• Working on directives to explore namespace at higher level

Packets arrive on an interface, and a longest-match look up is performed on it

• Three forwarding structures
  • Forwarding Information Base (FIB)
    • Stores list of outgoing interfaces, rather than a single next hop as in IP
  • Content Store (buffer memory)
    • Data is self-identifying and self-authenticating, so it's potentially of repeat use
    • Therefore stored as long as possible; LRU or LFU replacement rather than MRU
  • Pending Interest Table (PIT)
    • Track Interests forwarded upstream, creating reverse return trail
    • Erased as soon as they are used to forward a Data packet
    • Eventually time out
• When an Interest is received, content Store match is preferred over PIT match, which is preferred over FIB match
  • Thus, if there's already a Data packet stored matching an Interest, it's forwarded
  • If there's already a exact-match Pending Interest, the source interface will be added to the list of subscribers for the PIT entry and the timers updated
  • Otherwise, if there's a prefix-matching FIB entry, then the Interest needs to be sent upstream to that data
    • The source interface is removed from the interface list of the FIB
    • Interest is forwarded to the remaining interfaces in the FIB list
    • Interest is added to the PIT
• When Data packets arrive, they're matched similarly
  • If any Content Store entries exist it's a duplicate, so it is tossed
  • A PIT match consumes the Interest and is forwarded to all listed interfaces, excluding the source interface, and the Data is added to the Content Store
  • A FIB match means there are no interests, so it's discarded
• Interests not answered in some amount of time should be retransmitted

Data can't loop, but Interests can

• Random nonce is added so duplicates can be discarded

Data must return via reverse of same path as Interests

• Different from IP, where paths may differ

Strategy and security layers to IP stack

• Strategy layer controls forwarding of Interests
  • Can utilize multiple links to improve adaptability, performance
• Security layers allows data packets to be self authenticating
  • This is required in order to reuse data packets, prevent trivial spoofing & capture
  • Effectively secures the whole routing infrastructure

Effectively implicitly implements several things

• Network wide, self optimizing caching
• DTN
• Multicast
• Content based addressing
• Multi-homing and mobile nodes
• Load balancing
• Per-hop flow control, by at worst pairing one Interest to one Data
• Fairly intuitive mechanisms for growing trust networks
• Control/DoS prevention
  • Since returning Data packets are visible, node can determine if it's being bombed by valid Interests or not
  • Faking Interests are basically the only way to DoS the system

Questions

• Effective/efficient forwarding strategies?
• Use of broadcast link? I.e., how to determine facing?

Of note:

• Bunch of useful related work linked here