Mobility
- end-point moves (person, virtual machine, or WiFi to/from cellular), may
     be planned or unplanned
- happens all the time, e.g., move laptop from one WiFi hotspot to another,
     and applications keep working
- but, the key question is whether existing sessions continue seamlessly
- does this matter? simplifies apps and avoid (re)finding the remote end-point

Many proposals
- dozens of different proposed solutions
- different layers (e.g., Ethernet vs. IP vs. TCP)
- different kinds of networks (LAN vs. cellular)
- what is the design space?
- what stays the same as the device moves, and what changes?
- how are session end-points identified?
- two main patterns (dynamic routing and session location), and all solutions
      fall into one of these patterns or a composition of the two

Dynamic-routing mobility
- example #1: Ethernet MAC learning
      * name is a MAC address
      * learned entries expire
      * MAC learned at new location
      * costs: forwarding table size (though MACs are flat anyway), update cost
      * (can have directory approaches like SEATTLE and VL2)
- example #2: injecting into IP routing protocol
      * name is an IP address or prefix
      * intradomain (e.g., OSPF) or interdomain (e.g., BGP)
      * breaks hierarchical name space
- example #3: mobile IP (home agent, foreign agent)
- example #4: multicast MSM-IP (group address, mobile host joins/leaves)
- example #5: cellular network (within a single provider)      
- figure 6
- advantages
      * it simply works (except for mobile IP and MSM-IP which require host support)
      * no need for special directories
      * naturally handles simultaneous mobility of both end-points
- disadvantages
      * breaks hierarchical name space
      * so, larger forwarding tables and more frequent updates to network
      * doesn't naturally handle multihoming

Session-location mobility
- example #1: tell the remote end-point your new address, and update directory
- identifier (stays the same) and locator(s) (changes with attachment)
- simutaneous handoff? use rendezvous server or directory service
- figure 7 and 11
- advantages
       * keeps hierarchical name space
       * no need for changes to routing inside the network
       * can naturally handle multihoming
- disadvantages
       * mobile host must update all (possibly many) correspondents
       * complex protocol, need to get it right
       * need a scalable directory
       * requires changes to endpoints
       * or middleboxes to interact with legacy end-points
       * security issues because end-points can initiate updates