This section details the evaluation of DoM-FP. We implemented DoM-FP through the NS-3 simulator [19].Which is a discrete event simulator for implementing network simulation routines. In the following, DoM-FP is implemented completely in NS-3, and then several adopt metrics and simulation experiments are carried out. Finally, we analyze the obtained results to verify the feasibility and effectiveness of DoM-FP.
3.1 Simulation setup
The grid-5d infrastructural network topology for simulation consists of 25 routers with distance 100m between each two. Among them, each router functions as both PoA (point of Access) and forwarder, that is, the router can realize user communication through wired links, and can also receive user Con-tent s through wireless links. Wired links have a capacity of C = 100Mb/s and 2ms delay. In the wireless link, we use IEEE 802.11n WiFi on 5GHz frequencies, with Minstrel rate adaptation [20] and log distance propagation model plus Rayleigh-fading model for wireless channel. In addition, we set the transmission range of PoA nodes as 50m, which indicates that once the server moves out of the communication region, it may no longer receive requests from users. Therefore, the time of signal loss may vary due to different speeds and location of the server. Finally, the user is connected to node 5, the server's initial location is at node 1, and the RV or map server is connected to node 13.
To evaluate the performance of DoM-FP, a dynamic scenario simulating server mobility is designed. The Random Walk model is adopted [21], in which the server moves a fixed distance at a constant speed and then randomly changes direction. The move steps of servers are independent and unrestricted, which will not be affected by routers or other protocols. When a server enters the communication range of a router, it will access the network. Moreover, we vary mobile servers’ speed from 5m/s to 30m/s and perform 200 simulation runs for each different speed. In each simulation, user continuously retrieves content from a mobile server for 100 seconds. We use a central global routing controller to automatically populate all routers FIBs. The specific simulation parameter values are shown in Table
Table 1
parameter | Value |
Wired link bandwidth/(Mbit·s-1) | 100 |
Wired link delay/ms | 2 |
Link layer protocol | IEEE 802.11n |
Wireless rate | DsssRate1Mbps |
Wireless signal transmission range/m | 50 |
Propagation loss model | Rayleigh-fading model |
Mobility model | Random Walk model |
Fixed moving distance/m | 200 |
Ways to router cache | “Nocache” |
Request sending rate/(packet·s-1) | 1 |
Server moving speed/(m·s-1) | 5 ~ 30 |
To analyze the pros and cons of the experimental performance, DoM-FP is compared with two mobility support strategies, the Trace-based scheme KITE [22] and the Rendezvous-based scheme (RB) [23]. In addition, the packet loss ratio (Failed Answer Ratio, FAR), average request delay (Average Request Delay, ARD) and server handover consumption (Handover Overhead, HO) indicators were measured to verify the mobility support capability of the solution.
3.2 Simulation Results
1) Analysis of packet loss rate
Figure 4 shows the simulation results of the average packet loss rate FAR, one of the performance indicators for delay-sensitive traffic. FAR is represented by the number of retransmitted Requests over the total number of Requests sent by all the users in the topology. The experienced average packet loss is a consequence of mobile speed (from 0m/s to 30m/s) of different solutions: 8.1% for DoM-FP, 17.6% for KITE, 31.9% for RB.
The reason is that DoM-FP utilizes the server's mobile information to dynamically establish an effective forwarding path. When the server disconnects, the user requests will be recorded by RV. When the server reconnects to the network, it will send the content requested by the user immediately. At the same time, DoM-FP sets an appropriate lifetime of the TrFIB table, so that the user requests will not be forwarded to the old location of the server due to expiration, thereby reducing the FAR. In KITE, the temporary forwarding path established by the mobile server is invalidated each time it disconnects, so the user's requests are forwarded to the server's old location and discarded due to timeout. New traces are established only after the server reconnects to the network. In RB, users request the server's new location information from the mapping server before requesting content. However, the mapping information is not real-time, which results in high FAR. In addition, the higher handover frequency of servers increases the hysteresis of the location information obtained by users. Therefore, as the speed of servers increases, the FAR gradually increases. In conclusion, the FAR of DoM-FP is lower than that of KITE and RB.
2) Analysis of Average Request Delay
Figure 5 shows the simulation results of the average request latency ARD for user requests, which indicates the efficiency of the solution to support the server's mobility. ARD specifies the duration time, in second, between users send Requests and the arrival of the corresponding contents. In the speed range of 0-30m/s, as the speed increases, the ARD of DoM-FP gradually de-creases, and finally stabilizes at about 230ms. The ARD of KITE gradually stabilized at 600ms, while the ARD of RB continued to increase.
In principle, the fast recovery mechanism of DoM-FP reduces its ARD. After receiving the CD returned by the RV, the NDN router will extract the user requests from the CD’s content field to create/update the PIT table. This mechanism enables the mobile server to return the Content requested by the user immediately, without waiting for the user to resend Requests, thereby reducing the ARD of the user requests. For KITE, it maintains a constant trans-mission delay. The user requests may be forwarded to the old location of the server, and the Request retransmitted by the user will be discarded due to the PIT entry, so its ARD is higher than DoM-FP. In RB, the server will handover frequently at high speed, which causes the location information obtained by the user expiring frequently. The user needs to re-request the latest location information of the server from the mapping server. It can be seen that RB is not suitable for frequent handover of the mobile server. To sum up, compared with KITE and RB, DoM-FP can support the communication between users and high-speed servers.
3) Analysis of Handover Overhead
The average handover overhead HO represents the extra traffic generated to support mobility over the total traffic. That is, with DoM-FP, the signal overhead is defined as the numbers of the packets needed to signal the reach-ability of mobile servers. As shown in Fig. 6, the HO of DoM-FP is lower than both that of KITE and RB, in which the HO of DoM-FP is 35.4, that of KITE is 98.7, and that of RB is 91.4.
In DoM-FP, the mobile server sends mobile information every time it dis-connects and connects to the network, and does not need to send additional information to maintain the forwarding path. In KITE, every time a new PoA is relocated, the server has to send mobility information to create a new for-warding path. In addition, the server needs to send update messages periodi-cally. Finally, for DoM-FP, KITE, the HO increases with the speed, which is also the result of the increase in the number of relocations. In RB, the server needs to update the locators of the PoA, which sends update information to the mapping server, and the user needs to obtain the location information of the server from the mapping server. Once the Request times out, users must re-query the mapping server, so the HO of RB is higher than that of DoM-FP. In summary, DoM-FP requires the least average additional consumption to sup-port server mobility compared to KITE and RB.