To evaluate our implementation, we have conducted three experiments. The first measures the effects of selective active scanning, the second measures the handoff behavior below the MAC layer, and the last measures the handoff behavior when a mobile station hands off between two APs located at different subnetworks.
5.1 Hardware and Software Configuration
In the experiments, we used two laptops labeled MS1 and MS2, and one desktop labeled PC1. Each laptop equipped with a 1.0 GHz Intel Pentium III with 374 MB of RAM running Mandrake Linux 10.0. Compaq HNW-100 PCMCIA wireless NICs were used by the laptops.
The desktop was an AMD Athlon XP 1700+ with 512 MB RAM running Mandrake Linux 10.0. The HostAP driver version 0.2.0 and KPhone UA version 4.1.0 were used for all laptops, but they had been modified to support the function of the fast handoff mechanism. Ethereal version 0.10.0a was used by MS2 to capture and analyze the IEEE 802.11-based management frames.
The experiments were conducted in the 802.11b wireless network deployed in the CSIE building of National Chiao Tung University (NCTU), on the first floor where APs were deployed on the channels 1, 6 and 11.
5.2 The Effects of Selective Active Scanning
In this experiment we measure the delay and the packet losses for an on-going VoIP conversation when selective active scanning is performed periodically.
5.2.1 Experiment Setup
We setup a call using G.711 codec with 20ms voice packetization interval between MS1 and PC1, and use MS2 for packet sniffing. Then we instruct MS1 to periodically perform selective active scanning on the sniffed channel every 200ms. MS1 recorded the information of RTP packets, including the time when a RTP packet arrived and the sequence number of the RTP packet.
5.2.2 Experiment Results
MS2 observed that the time between the probe request issued and the last probe response received ranges between 1 ms and 3 ms. This result doesn’t consider the channel switching delay and other factors caused the design of hardware, so that it cannot reveal the real influence on an on-going VoIP conversation.
According to the records in MS1, there is a maximum interval, 75ms, between two consecutive voice packets are received, but there is no packet loss in the course of selective active scanning. Since we expect the inter-packet arrival time to be 20 ms, our measurements indicate that the selective active scanning takes at most 55 ms.
5.3 The Effects of the Intra-subnet Handoff
In this experiment, we measure the delay and the packet losses for an on-going VoIP conversation when MS1 handoff between two access points on the same subnetwork.
5.3.1 Experiment Setup
We deploy a mobility server and a RTP relay agent to support the fast handoff mechanism described in Chpater 3. Then we set up a call using G.711 codec with 20ms voice packetization interval between MS1 and PC1 with the RTP relay agent lying in the middle of
RTP stream path to relay RTP packets for MS1 and PC1. MS1 walks along the U-shaped hallway on the first floor of the CSIE building, NCTU, so that it performs two phase handoff procedure. MS1 also recorded the arrival time of each RTP packet, the sequence number of this RTP packet, and the time when MS1 received the association event, which is used by driver to notify applications that a new association with an AP had been constructed.
5.3.2 Experiment Results
Table 1 and Table 2 present the results we obtained from the records in MS1. The handoff interval of the original 802.11 method is defined between the notification of association event from the driver and the time the last voice packet received prior to this notification. The handoff interval of our method is defined between the time when a handoff is issued and the time the first voice packet received after the handoff completes.
The results show that the handoff latency and the number of packet lost drop to about 20% and 14% of the values obtained with the original handoff because scanning target APs, which took more than 90% of the total handoff time, was done stealthily, and a MS directly associates with the target AP at the handoff.
Table 1: Handoff latency (ms)
Experiment 1 2 3 4 5 6 7 8 9 10 average
Original handoff 211 237 234 240 244 220 240 231 238 227 232
Fast handoff 41 47 40 65 46 47 55 44 37 47 47
Table 2: The number of packets lost during handoff
Experiment 1 2 3 4 5 6 7 8 9 10 average
Original handoff 9 10 11 10 11 8 9 11 10 9 9.8
Fast handoff 2 2 1 2 1 1 2 1 1 1 1.4
5.4 The Effects of the Inter-subnet Handoff
In this experiment, we measure the handoff delay and the packet losses of an on-going VoIP conversation during the handoff between two access points on different subnetworks.
5.4.1 Experiment Setup
Because the CSIE department of NCTU deploys all APs on the same subnetwork to form a large extended service set (ESS). We have deployed another ESS on a different subnetwork such that we can measure the handoff behavior across two subnetwork boundary.
We setup a call using G.711 codec with 20ms voice packetization interval between MS1 and PC1 with the RTP relay agent lying in the middle of RTP stream path to relay RTP packets for MS1 and PC1. MS1 walks around the boundary between the ESS of the CSIE department and our trial ESS, such that it can perform the network-layer and the application-layer handoff. MS1 recorded the same information as the experiment in Section 5.3.
5.4.2 Experiment Results
Table 3 presents the results of our method. Even though MS1 hands off between two APs located at different subnetworks, the handoff performance is still close to that of the intra-subnet handoff because we made lots of preparations, such as IP pre-assignment and RTP stream fork for the handoff.
Table 3: Delay (ms) and the number of packets lost during handoff
Experiment 1 2 3 4 5 6 7 8 9 10 average
Delay 58 120 40 81 43 59 40 85 58 52 64
Packet loss 2 5 1 3 1 1 1 5 2 1 2.2