M
and the
V
k
’s are independent, the mean delay is
E[
W
] =
T
max
E[
M
]E[
V
] =
T
max
e
a
−
1
a
−
1
(2)If we deﬁne the normalized quantities
ˆ
Y
= (
Y
−
R
)
/R
and
ˆ
W
=
W/T
max
, then
a
= log(1 + E[ˆ
Y
]
R/D
)
and
E[ˆ
W
] =E[ˆ
Y
]
R/D
log(1 + E[ˆ
Y
]
R/D
)
−
1
(3)This is the trade-off between average dissemination delayand average dissemination coverage. As the latter increases,also the former increases. With VBN the trade-off is quasi-linear for a given ratio between the interval length
D
and thecoverage distance
R
. The analysis above holds for
R/D
≥
2
.In the following Sections we account for the effect of MAClayer, ﬁrst as for the time quantization implied by non nulldelivery times through the MAC interface, even for light load.Then, by means of simulation we assess the effect of differenttrafﬁc loads as the message ﬂow rate varies.IV. E
FFECT OF
MAC
DELAY ON TIMER DIMENSIONING
At light loads, MAC frame delivery times in the IEEE802.11p are in the order of 1 ms, depending on frame payloadand air bit rate. In case two vehicles are located so that theirrespective timers differ by an amount less than the MACdelivery delay, say
ϑ
, the inhibition rule will not work. Notethat this issue affects
any
timer-based dissemination protocol.In the following we analyze this effect on TBN, again withPoisson vehicle spatial distribution.Conditional on the number
k
of vehicle in a nominalrelay interval of length
D
, vehicle positions are i.i.d. uniformrandom variables over
(0
,D
)
. Since the location of each of the
k
vehicles is uniform over
(0
,D
)
, it will be in a segmentof length
Δ
with probability
q
= Δ
/D
. Let us divide theinterval into
n
equal size segments, namely
Δ =
D/n
, sothat
q
= 1
/n
. The timer
τ
is a map between the intervaland time, i.e.,
τ
:
i
∈ {
1
,...,n
} →
(0
,T
max
)
. Weconsider a generic nominal relay interval, say
(
−
D/
2
,D/
2]
,with nominal position
0
. Then, the timer is quantized to
n
values, the
i
-th of which is
τ
(
i
)
and it is associated to the set
(
−
i
D
2
n
,
−
(
i
−
1)
D
2
n
]
∪
((
i
−
1)
D
2
n
,i
D
2
n
]
,
i
= 1
,...,n
.We are then in a similar situation as with a random back-off mechanism where
k
competing nodes select at randomone out of
n
slots to attempt transmission. In our case, eachvehicle “selects” at random segment
i
and schedules therelaying of the packet with a timer value
τ
(
i
)
. Thanks to thePoisson distribution properties, the “selection” of each vehicleis independent of all other vehicles. We assume
τ
(
i
) =
iϑ
, sothat a node that sets the timer value
τ
(
i
)
has time enough topass the message on to the MAC layer and have it deliveredto the other nodes in the road interval
before
any other vehiclethat has selected a timer value
τ
(
j
)
,j > i
can release its copyof the message to the MAC layer. In other terms,
ϑ
is no lessthan the MAC “resolution” time.Let us denote
f
(
i
) =
P
(
node selects τ
(
i
))
and
G
(
i
) =
P
(
node selects τ
(
j
)
,j
≥
i
)
,
i
= 1
,...,n
. Let
q
k
be theprobability that there is no forwarding superposition, i.e., asingle node out of the
k
forwards the packets and all othersbecome aware of the forwarding action on time so as to canceltheir own scheduled forwarding. Then
q
k
=
n
i
=1
kf
(
i
)
G
(
i
+ 1)
k
−
1
(4)In our case
f
(
i
) = 1
/n
and
G
(
i
) = (
n
−
i
+ 1)
/n
,
i
= 1
,...,n
, so
q
k
= (
k/n
)
n
−
1
j
=0
(
j/n
)
k
−
1
. To recover theunconditional probability of no forwarding superposition,
q
,we weight
q
k
with the probability that there are
k
vehicle inthe considered interval of length
D
, conditional on there beingat least one vehicle in the interval, i.e.,
a
k
e
−
a
/k
!1
−
e
−
a
for
k
≥
1
,with
a
=
λD
. Therefore
q
=
∞
k
=1
e
−
a
1
−
e
−
a
a
k
k
!
q
k
=
n
i
=1
f
(
i
)
ae
−
a
1
−
e
−
a
∞
k
=0
[
aG
(
i
+ 1)]
k
k
!=
ae
−
a
1
−
e
−
an
i
=1
f
(
i
)
e
aG
(
i
+1)
(5)In our special case:
q
=
ae
−
a
1
−
e
−
an
i
=1
1
ne
a
(
n
−
i
)
/n
=
a/ne
a/n
−
1
(6)The probability of superposition
¯
q
= 1
−
q
tends to 0 as
n
increases since
q
∼
1
/
[1 +
a/
(2
n
) +
o
(1
/n
)]
as
n
→ ∞
.For
a
= 10
, we obtain
¯
q
≤
0
.
075
for
n
≥
65
. Note that themaximum value of the timer is
T
max
=
nϑ
. In the numericalexample above, it is
T
max
≈
97
.
5
ms
for
ϑ
= 1
.
5
ms
.Concurrent forwarding of two or more vehicles in the samenominal interval due to MAC induced timer quantization canhave a catastrophic effect on message dissemination if thereis an inhibition rule. In that case, vehicles in the
(
k
+ 1)
-th nominal interval receive two or more copies of the samemessage from the nodes in the
k
-th interval and are forced tocancel message forwarding, thus stopping dissemination. Thisimpairment can be worked around by means of TBN rules,thanks to the variables
k
A
and
k
S
(see Section II).The average value of the winning timer is given by
E[
τ
] =
ae
−
a
1
−
e
−
an
i
=1
iϑf
(
i
)
e
aG
(
i
+1)
=
ϑa/ne
a
−
1
n
−
1
j
=0
(
n
−
j
)
e
a/n
j
=
ϑa/ne
a
−
1(
e
a
−
1)
e
a/n
−
n
(
e
a/n
−
1)(
e
a/n
−
1)
2
∼
nϑ
1
a
−
1
e
a
−
1
the last relationship holding asymptotically as
n
→ ∞
. Hence,for large values of
n
, the expected delay induced by therelaying timer of the FL is of the order of
T
max
/a
. Thisanalysis provides equations to dimension the value of
T
max
,