IP-Based Protocols for Mobile Internetworking

One-line summary: Mobility support by tunneling and encapsulation (IPIP) and a simple MSS-to-MSS protocol (MICP). M-IP requires no changes to higher-level protocols, scales reasonably well due to small amount of state at MSS (cached and timed out), and places the burden of mobility on the MH�s and MSS�s rather than the rest of the world.

Overview/Main Points

• MH maintains a permanent home address; its MSS is responsible for reaching it
• Routing cases:
  • MH-to-MH within a cell, normal ARP delivers pkts (MSS not involved)
  • MH to FH, normal gateway thru MSS
  • FH to MH, gateway thru MSS which then "discovers" what MSS is responsible for the MH (see below)
  • MH to MH across cells: receiving MSS proxy-ARPs for MH
• Handoff
  • MH detects beacons coming from MSS�s (multiple beacons is "still a research issue")
  • MH sends greeting and old MSS to new MSS, waits for ACK that includes new transient-IP addr
  • New MSS informs old MSS of handoff and starts forwarding before that message is ACK�d (in case of partition or old MSS down)
  • Beaconing done by MSS (not MH) to reduce traffic (no req/response needed), allow MH to verify it is still in range, and conserve power on MH
• MH periodically sends "still here" packet if no other traffic, to inform MSS it is still in this cell
• Redirect: MSS receives pkt for MH it no longer serves; redirect is sent to sender, and pkt is forwarded (to avoid having to rely on higher-level timeouts)
• Peer MSS discovery: multicast, link-level broadcast, or unicast; list can be static and distributed to new MSS�s as they join, or each MSS can propagate discovery requests to its immediate neighbors (modulo problems with connectedness of the graph, partitioning, etc)
• Some perf issues: handoff latency, MSS load, encaps overhead. It is argued that all are within reason during normal operation.
• Fault tolerance: exploit multiple beaconing?
• Scalability: "dogleg routing" is the worst problem. Argument: scalability should be compared to alternative of reconfiguring MH on every handoff, rather than to "MH-less" alternative.
• Tunneling: some discussion of earlier approach based on "loose source routing" (what is this?) and why it doesn�t work here, since dest addr of MH is a valid unicast address.

Relevance

Flaws