In this section, we evaluate the performance of ABACS in terms of computational delay and transmission overhead. To the best of our knowledge, there is no similar security scheme for emergency services. Therefore, we compared ABACS with the ECDSA scheme, which is adopted by the current IEEE1609.2 standard [9] as a security scheme for VANETs.
3.5.1 Computational Delay
As described in Section 3.3.4.1, the total computational delay (denoted as TV−total) for an EV in ABACS is 2d × Tpair+ 4 × Tmul ms. Here, we focus on the computational delay for TTA, because TTA is designed to handle all rescue-related messages sent from a number of EVs. Table 3.2 shows the computational delay of the dominant cryptographic operations, including point multiplication Tmul and bilinear pairing Tpair, for TTA in ABACS and ECDSA schemes in communications with a single EV or multiple EV s. Since ECDSA does not provide message con dentiality, we assume that the Elliptic Curve ElGamal encryption is adopted in ECDSA. Thus, it costs 2 Tmul for an encryption operation and 1 Tmul for a decryption operation. According to [5, 27], the time required to perform ECDSA signature
Table 3.2: Comparisons of computational delay for TTA (ms)
Communication with a single emergency vehicle Communication with n emergency vehicles
ABACS ECDSA ABACS ECDSA
RQMEncryption (1 + i)Tmul 3Tmul (1 + i)Tmul 3nTmul
RRM Decryption 1Tmul 5Tmul nTmul 5nTmul
M AMEncryption (a + d − 1)Tmul 3Tmul (a + d − 1)Tmul 3aTmul
Total (a + d + i + 1)Tmul 11Tmul (a + d + i + n)Tmul (3a + 8n)Tmul
i: Total number of selected attributes (i ≥ d), d: Minimal number of overlapped attributes; a: Number of the assigned EV s.
and certi cate veri cation is 4 Tmul, and the time required to sign an ECDSA message is 1 Tmul. Therefore, an encryption in ECDSA costs 3 Tmul (EC ElGamal encryption + ECDSA signing) and a decryption in ECDSA costs 5 Tmul (ECDSA veri cation + EC ElGamal decryption). An RQM encryption in ABACS requires 1 Tmul for ¯E and i Tmul
for Ei, where i is the total number of selected attributes(i ≥ d). For ease of evaluation, we set i = d + 2 in the following performance evaluation. A decryption in ABACS requires 1 Tmul for EC ElGamal decryption. The computational delay for an MAM encryption is (a + d − 1)Tmul, in which aTmul is for the ELP s attributes of the assigned EV s and (d − 1)Tmul is for the dummy attributes.
Referring to [22], the computational delay for Tmul and Tpair is 0.78 ms and 2.82 ms, respectively. Figure 3.5.1 (a) shows the relationship between the computational delay and the number of queried emergency vehicles (n), if the number of assigned EV s involved in a rescue is 5 (a = 5). It is observed that ABACS can greatly reduce the computational delays for dierent values of d. Moreover, Figure 3.5.1 (b) shows the ratio of the computational delay of ABACS to that of ECDSA. In a general rescue mission with only a few assigned EV s, ABACS is more than 80% faster than ECDSA when the number of queried EV s is greater than 40. Moreover, we investigate the computational delay for an disaster event requiring dierent numbers of assigned EV s. As shown in Figure 3.5.2 (a), when 100 EV s are queried, i.e., n = 100, ABACS achieves smaller computational delays than ECDSA for dierent numbers of assigned EV s. In fact, ABACS generates only an MAM for all the assigned EV s, whereas ECDSA has to produce distinct MAMs to individual EV s. This also explains why the computational delays in ABACS moderately increases by at most
0 50 100 150 200 250 300
Number of queried emergency vehicles (#)
Computational delay for TTA (ms)
a=5
ABACS (d=4) ABACS (d=10) ECDSA
(a) Computation delay vs. number of queried emergency vehicles
Number of queried emergency vehicles (#)
Computational delay ratio
a=5
ABACS(d=4)/ECDSA ABACS(d=10)/ECDSA
(b) Computational delay ratio vs. number of queried emergency vehicles
Figure 3.5.1: Computation delay evaluation in regular emergency events
0 20 40 60 80 100
Number of assigned emergency vehicles (#)
Computational delay for TTA (ms)
n=100
ABACS(d=4) ABACS(d=10) ECDSA
(a) Computational delay vs. number of assigned emer-gency vehicles
Number of assigned emergency vehicles (#)
Computational delay ratio
n=100
ABACS(d=4)/ECDSA ABACS(d=10)/ECDSA
(b) Computational delay ratio vs. number of assigned emergency vehicles
Figure 3.5.2: Computational delay evaluation in disaster events
other hand, the computational delay in ECDSA also increases as more EV s are assigned;
however, all are greater than 626.34 ms due to the computations in RQM and RRM. The computational delay ratios, illustrated in 3.5.2 (b), show that the computational delay of ABACS is only at most only 19% and 20.1% of that of ECDSA when d = 4 and d = 10, respectively.
0 50 100 150 200 250 300
Number of queried emergency vehicles (#)
Transmission overhead (KBytes)
ABACS(d=4) ABACS(d=10) ECDSA
(a) Transmission overhead vs. number of queried emer-gency vehicles
Number of queried emergency vehicles (#)
Transmission overhead ratio
ABACS(d=4)/ECDSA ABACS(d=10)/ECDSA
(b) Transmission overhead ratio vs. number of queried emergency vehicles
Figure 3.5.3: Transmission overhead evaluation
3.5.2 Transmission Overhead
In this section, we compare the transmission overhead of the two schemes. The transmission overhead mostly arises from the communications between TTA and the RSUs. The following evaluation focuses on the transmission overhead introduced by the signature, certi cate, and encryption/decryption parameters, while the message itself is not considered. According to [9], the format of a signed message contains a 56-byte ECDSA signature and a 125-byte certi cate. In ABACS, the transmitted parameters of RQM include 4*i 125-bytes8 for the identity and 20*i bytes for the decryption parameters, where i is the total number of selected attributes(i ≥ d). As in Section 3.5.1, we set i = d + 2 in the following performance evaluation. As for RRM, the parameters consist of 4 bytes for the identity, and 20 bytes for the decryption parameters. With regard to MAM, the parameters consist of 4 * (a + d- 1) for the identity, 20 bytes for the encrypted credentials, and 20 * (a + d - 1) for the decryption parameters. According to DSRC [1], the bandwidth of a wireless data channel in VANETs is 10 MHz, corresponding to a channel data rate within the range of 3-27 Mb/s [28]. A typical data rate of 6 Mb/s is usually assumed for VANETs. Under the assumption of d = 10 and i = 12, the length of RQM will be 4*12 + 20 + 20*12 = 308 bytes. According to [5], there can be 180 vehicles within the communication range of an RSU. In a extreme
8We assume each attribute is of 4 bytes.
Table 3.3: Comparisons of transmission overhead (bytes)
Communication with a single emergency vehicle
ABACS ECDSA
RQM 4i + 20i 181
RRM 4 + 20 181
M AM 4(a + d - 1) + 20 + 20(a + d - 1) 181
Total 24(a + d + i ) + 20 543
Communication with n emergency vehicles
ABACS ECDSA
RQM (4i + 20i)NT RSU 181n×NT RSU
RRM (4 + 20)n 181n
M AM (24a + 24d - 4)NARSU 181a × NARSU
Total (4i + 20i)NT RSU + (24a + 24d - 4)NARSU + 24n 181(n × NT RSU + a × NARSU + n ) NT RSU : Total number of RSUs; NARSU : Number of RSUs where the assigned EV s are visiting.
case that all 180 vehicles are EVs, the demanded throughput for RQM is at most 0.42 Mb/s (180×1×308×8
1024×1024 Mb/s). Similarly, the throughput for RRM and MAM is 0.05 Mb/s and 0.45 Mb/s, respectively. Therefore, the maximal demanded throughput of ABACS is much smaller than 6 Mb/s.
Suppose that NT RSU is the total number of RSUs and NARSU is the number of RSUs where the assigned EV s are visiting. Because the RQM is disseminated by broadcasting over NT RSU RSUs, the transmission overhead of the RQM delivery can be estimated by (4i + 20i)NT RSU. The transmission overhead of MAM is (24a + 24d - 4)NARSU, because the MAM is only multicast to NARSU RSUs. Table 3.3 summarizes the transmission overhead in ABACS and ECDSA schemes. From Figure 3.5.3 (a), it can be seen that the transmission overhead of ECDSA is signi cantly higher than that of ABACS for d = 4 and d = 10.
Because of the use of broadcasting and multicasting, the transmission overhead incurred by ABACS moderately increases as the number of queried EV s increases. Figure 3.5.3 (b) shows the ratio of the transmission overhead of ABACS to that of ECDSA is shown. It can be seen that the more the number of EV s are queried, the lower is the transmission overhead ratio that can be achieved. More precisely, when the number of queried EV s is greater than 61, the transmission overhead of ABACS for d = 4 and d = 10 is only 1.4%
and 2.6% that of ECDSA, respectively.
Table 3.4: Simulation Parameters
City simulation area 1000m × 1000m
RSU Communication range 400 m
Simulation time 100 s
Wireless Protocol 802.11a
Wireless channel bandwidth 6 Mbs Wired channel bandwidth 100 Mbs Packet size for ECDSA message 181 bytes Packet size for RQM message (d = 4 or 10) 164 or 308 bytes Packet size for RRM message (d = 4 or 10) 40 bytes Packet size for MAM message (d = 4 or 10) 184 or 328 bytes
3.5.3 Simulation
In addition to the theoretical analysis of computational delay in Section 3.5.1, we further evaluate the average processing delay and message loss ratio via a simulation on ns-2 [29]. In the simulation, we consider an area of 1×1 km2 in urban areas. The simulation parameters are given in Table 3.4. We also adopt the TraNS [30] tool in the simulation for a better mobility model for vehicles. It is assumed that the maximum vehicle speed is 70 km/h.
Predictive transmission is also implemented in the simulation. The Medium Access Control (MAC) protocol follows the IEEE 802.11a standard, which is the basis of DSRC [27, 30].
The average processing delay (denoted as avgD) is de ned as avgD = N1
where Nr is the number of emergency event reports, and NEV is the number of EVs.
TSendi,RQM,j is the time when the application layer of TTA sends the rescue query message (RQM) of the i-th emergency event report to the j-th EV. TRecvi,M AM,j is the time when the application layer of the j−th EV receives the mission assignment message (MAM) of the i-th emergency event report sent from TTA.
Figure 3.5.4a shows the average processing delay versus the number of queried EVs in regular emergency events. Note that the short waiting period (ξ) is not included in the average processing delay. As in Section 3.5.1, we assume the number of assigned EVs is 5 (a = 5). The simulation result shows that the average processing delay of ABACS (d
= 4) is close to that of ABACS (d = 10), and ECDSA consumes more processing delay
20 40 60 80 100 120
Number of queried emergency vehicles (#)
Average processing delay (s)
ABACS (d=4) ABACS (d=10) ECDSA
(a) Average processing delay in a regular emergency event
0 20 40 60 80 100 120
Number of assigned emergency vehicles (#)
Average processing delay (s)
ABACS (d=4) ABACS (d=10) ECDSA
(b) Average processing delay in disaster events
Figure 3.5.4: Average processing delay
than the others. It is also seen that the more EVs are queried, the more advantages of ABACS can be achieved. This result is basically the same as the analysis shown in Figure 3.5.1a. The simulation result for disaster emergency events is shown in Figure 3.5.4b. In general, there are only slight variations of processing delay in ABACS with respect to the number of assigned EVs. However, the processing delay of ECDSA increases as more EV s are involved. This result also corresponds with the analysis shown in Figure 3.5.2a.
The average loss ratio, denoted as avgLR, is de ned as avgLR = N1
Recv +MRecvj,RRM,i+MRecvi,M AM,j MSendi,RQM,j+MSendj,RRM,i+MSendi,M AM,j),
where Nris the number of emergency event reports. MSendi,RQM,j is the number of RQMs sent to the j-th EV for the i-th emergency event report9, MSendj,RRM,iis the number of RRMs sent by the j-th EV for the i-th emergency event report, and MSendi,M AM,j is the number of M AMs sent to the j-th EV for the i-th emergency event report. MRecvi,RQM,j represents the number of RQMs received by the j-th EV for the i-th emergency event report, MRecvj,RRM,i represents the number of RRMs received by TTA for the i-th emergency event report, and MRecvi,M AM,j represents the number of MAMs received by the j-th EV for the i-th emergency event report.
Figure 3.5.5a shows the relationship between the average loss ratio and the number of
9Note that messages sent via broadcasting should be counted multiple times as many as the number of receivers.
20 40 60 80 100 120
Number of queried emergency vehicles (#)
Average message loss ratio (%)
a=5
ABACS (d=4) ABACS (d=10) ECDSA
(a) Average message loss ratio in regular emergency events
0 20 40 60 80 100 120
Number of assigned emergency vehicles (#)
Average message loss ratio (%)
n=120
ABACS (d=4) ABACS (d=10) ECDSA
(b) Average message loss ratio in disaster events
Figure 3.5.5: Average loss ratio
queried EVs in regular emergency events. The loss ratio of ECDSA is up to about 40%
when the number of queried EVs is more than 50, while ABACS attains the same loss ratio when the number of queried EVs is more than 100. Furthermore, we also investigate the loss ratio in disaster events, shown in Figure 3.5.5b. It can be seen that the loss ratio of ECDSA rapidly increases as the number of assigned EVs grows. The reason is that in ECDSA there needs a dedicated MAM message for each assigned EV . Each MAM message is encrypted using the public key of the EV , and is sent separately over the VANET. On the other hand, the loss ratio of ABACS only gradually rises no more than 40% as the number of assigned EVs increases, because in ABACS only one encrypted MAM message is required. It can be observed that the average loss ratio of the two ABACS-based schemes is only slightly aected by the number of assigned EV s. Some studies [17, 18] in the MAC layer can be used to further improve the packet loss problem.