Results: Cold Switch

Figure: Network topology used for cold switch experiments. As shown by the dashed arrow, the mobile host moves from one location to another on the same subnet during handoff.

In this experiment, we compare the handoff performance of ROAM and MIPv6 during a cold switch when the MH is far away from the CH. Figure 15 shows the experimental setup. We use the NIST Net [34] network emulation package to emulate a round trip time (RTT) of 70 ms between the MH and CH. In the setup for ROAM, RTT between the MH and the $i3$server is approximately 3 ms. The NIST Net router delays packets between the $i3$server and the CH by 70 ms. We emulate the MIPv6 scenario by running the $i3$server on the same machine as the CH since binding updates are propagated to the CH in MIPv6. The NIST Net router delays packets between the MH and CH by 70 ms. During a cold switch, the first interface is shutdown around 35-40 ms before the second interface is brought up. During this disconnected interval, $t$, packets from the $i3$server to the MH are lost in both ROAM and MIPv6. However, the number of packets that are lost after cold switch completes is proportional to the delay between the MH and the indirection point.

Figure: TCP sequence trace showing a bulk transfer with a cold-switch for (a) ROAM, and (b) MIPv6.

Figure 16 plots the TCP sequence numbers seen at the CH (TCP sender) for the ROAM and MIPv6 scenarios during a cold switch. ROAM recovers from packet loss caused by the cold switch by entering fast retransmit when the MH receives duplicate acknowledgements generated by packets received after the lost packets. However, in MIPv6, the MH loses the entire window of data and the CH waits for a timeout and goes into slow start before retransmitting the lost packets.

If the disconnectivity time due to cold switch is $t$, and $t<RTT<2t$, then ROAM can recover by fast retransmit whereas MIPv6 has to recover by timeout. If RTT is greater than $2t$, then both ROAM and MIPv6 can recover through fast retransmit. However, ROAM will recover sooner because of its ability to choose a nearby $i3$server irrespective of the CH's location, thereby greatly reducing packet loss.