L17: Names, bindings, and network architecture

Quick review of DHTs:
  - Distributed overlay network of symmetric nodes (peers)
  - Lookup primitive -> put/get interface
    - Just like a regular hash table, hence the name
    - scalable resolution of arbitrary identifiers
    - DOA uses basic put/get, while i3 uses basic lookup to route
  - Nodes maintain routing tables with entries to other nodes
  - Nodes in routing table chosen based on some geometry/metric
  - Nodes forward messages on behalf of other nodes

(These security problems mostly taken from Castro et al, OSDI 2002)
Routing security:
  - Securely assigning node identifiers
    - Violates uniform random distribution assumption
    - Can partition network
    - Control who is in a particular node's routing table
      (assuming away proximity routing)
    - can target specific data for corruption, deletion or denial of service
  - Routing table maintenance
    - Attackers can supply bad info, giving a higher fraction of bad nodes 
      in table
    - routing updates in protocols that don't constrain IDs of neighbors
      are particularly vulnerable
  - Message forwarding
    - Intermediate nodes can drop messages


(Some of this intro taken from Hari's "Layered Names" SIGCOMM 04 talk)

DHTs facilitate many applications that couldn't be done scalably beforehand

Flat names:
  - IP and DNS names both tied closely to an underlying structure
    (motivated by "host-centric" apps such as web, ssh, etc)
  - However, referent can have mutable properties that might change the name
    --> e.g., web link breakage
  - This constrains movement
  - Undergrad at Berkeley to grad at MIT, had to change webpage

  - Instead we want flat names for data and hosts
  - Sequence of "random" bits
    - 160 bits (a few orders of magnitude from the number of atoms in the 
      universe, between 4^78 and 6^79)
  - No semantic meaning
    - no hierarchy
    - does not impose restriction on the thing that it names
  - These can be resolved in a scalable fashion using DHTs
    - DHTs make flat names plausible, where in early Internet days maybe 
      they weren't
  - Can be used to name anything given a large enough namespace
  - Flexible migration

  - Downsides:
    - Not user-friendly (maybe AOL keyword search interface?)
    - No locality
    - No control over names

DOA:
  - Uses flat names and a DHT resolution service to form a new kind
    of Internet architecture where delegation (to NATs, firewalls, etc) is a 
    primitive.

i3: 
  - A forwarding infrastructure for routing packets to services using flat 
    identifiers