7. Chapter VII

7.5 Simulations

while the other 25 are static and placed at different intersections. The third and fourth network types consist of 75 user nodes and 25 mesh points. The fourth ad hoc mesh network consists of 9 mesh access points, with 16 mesh relay points. A mesh relay point only forwards mesh traffic, and operates on a different channel than is used by the user nodes.

Every node in the network will use 802.11g at the physical layer. The MAC protocol and routing protocol, will differ between the different scenarios as ex-plained below.

While this simulation area might be smaller than what is normally used in ad hoc network simulation studies, it should be noted that the urban simulation area consists of 16 city blocks, where each block is a square of 100m, intersected by 25m wide streets. This area will produce a topology with an average hop-count of about 6-7 hops, which is significantly longer than what is normally used in ad hoc simulation studies. The suburban environment consist of the same city topology, but here a block consists of many smaller and lower houses. Mesh Points as used in the mesh topologies are placed above these houses, while user nodes are placed below. This means that for the suburban environment it might be possible for two user nodes to find a connecting signal path through a city block. This is not possible in the urban environment, as the walls around the city block completely blocks the signal path, although the signal might be reflected or refracted around corners and thus eventually reach an other user node.

The power consumption of each individual node is measured during each simu-lation. Different power values are used depending on whether the node is transmit-ting, receiving, sensing or if it is idle. The power values of these different modes are modeled after the Cisco Aironet 802.11 a/b/g chip [28].

In the first network traffic type we are simulating are a varying number of bidirectional 56 kbps CBR traffic sources, modeled after the G.711 [29] codec.

A technique that can be used to predict user satisfaction of a conversational speech quality is the ITU-T E-model. The E-Model is standardized by the ITU as G.107 [30], and is a tool that can estimate the end-to-end voice quality, taking the IP telephony parameters and impairments into account. This method com-bines individual impairments (loss, delay, echo, codec type, noise, etc.) due to both the signal’s properties and the network characteristics into a single R-rating.

This method can be used as a good quality of service measure for VoIP calls that consider a user’s opinion about the service. The Mean Opinion Score (MOS) is a method recommended by the ITU and the IEEE 802.20 group [31] to measure speech quality. In this method, the users rate the call quality in a range varying from 1 (bad) to 5 (excellent). We will apply this rating to both the G.711, and the enhanced G.711 [32], codecs.

In these cases we are simulating a constant random load of 5, 10, 15, 20 and 25 source destination traffic pairs. By constantly random we mean, that the traffic load on the network is constant, but that the source destination pairs are constantly changing. Each CBR flow is one minute long, and when one flow ends, a new flow will instantly start between a new source destination pair somewhere else in

7.5. Simulations 141

the network. These simulations are running for a total of 60 minutes. These traffic scenarios are designed to roughly model voice communications between two voice capable devices, when the network is loaded by various amount of traffic.

For each simulation scenario and network type, user nodes will move accord-ing to the pedestrian mobility model defined in Qualnet 4.0. Here, user nodes move along the streets to a randomly chosen street corner somewhere on the city map.

When the nodes arrive at their destination, they will randomly chose a new destina-tion somewhere on the map. Every time they arrive at a street corner, they run a 50

% chance of having to wait for a red light before proceeding across the street. Each pedestrian user node will move at a constant speed uniformly choosen between 1.5 and 2.5 m/s.

Radio signals will be affected by multi path fading according to the Ricean fading model, with a k-factor of either 0 or 1, depending on whether a direct line of sight between each pair of hops are available or not. The multi-paths and atten-uations are calculated by the IRT tool.

7.5.2 Pure Ad Hoc Network Simulations

Here all user nodes are mobile as described above. The first simulation setup con-sists of user nodes running standard 802.11g DCF, with AODV as the routing pro-tocol, on a single channel. The second setup we are considering is the same as the first, but DIVR is used as the routing protocol, and our DIVM MAC protocol.

7.5.3 Supported Ad Hoc Networks Simulations

These simulations are exactly the same as those described in section 7.5.2. The only difference here is that 25 nodes will be static, and strategically placed in in-tersections.

7.5.4 Single Hop Mesh Network Simulations

In these setups, all user nodes will run the registration application defined in sec-tion 7.3.2. Each user node will have access to an MP within a single hop.

Within the mesh, we will use AODV or our DIVR routing, with all mesh nodes having two different interfaces running either IEEE 802.11g DCF and/or our DIVM MAC protocol. Every mesh node has at least one interface using 802.11g DCF, which is used for communication with user nodes.

Mesh nodes, use the simple address mapped multi channel scheme [24]. No coordination whatsoever is done by a 802.11 source before it chooses a channel, it depends purely on the address of target node. The channel is chosen as chan-nel=(address) mod (number of channels).

Contrary to the two ad hoc scenarios, all user nodes are equipped with a single 802.11g interface. Since the user nodes connect directly to the MPs, no dynamic routing protocol needs to be running on these interfaces.

7.5.5 Mesh Ad Hoc Network Simulations

This setup considers a more sparsely deployed mesh network where user nodes may need to connect to the MP over several hops. Mesh access points are config-ured in the same way, and with the same protocol types, as in the single hop case.

Mesh relay points lacks the 802.11g DCF interface used to communicate with user nodes.

User nodes on the other hand, still have a single wireless interface, but are now also running the AODV routing protocol. Otherwise, they are configured as in the previous case.

7.5.6 Simulation Results

5 10 15 20 25

Delay (ms) 0

0.2 0.4 0.6 0.8 1

f(x), F(x)

Ad hoc Ad hoc Supported ad hoc Supported ad hoc Mesh Mesh Ad hoc mesh Ad hoc mesh

Delay Urban

5 flows AODV

(a) AODV Urban.

5 10 15 20 25

Delay (ms) 0

0.2 0.4 0.6 0.8 1

f(x), F(x)

Ad hoc Ad hoc Supported ad hoc Supported ad hoc Mesh Mesh Ad hoc mesh Ad hoc mesh

Delay Urban

5 flows DIVR

(b) DIVR Urban.

5 10 15 20 25

Delay (ms) 0

0.2 0.4 0.6 0.8 1

f(x), F(x)

Ad hoc Ad hoc Supported ad hoc Supported ad hoc Mesh Mesh Ad hoc mesh Ad hoc mesh

Delay Suburban

5 flows AODV

(c) AODV Suburban.

5 10 15 20 25

Delay (ms) 0

0.2 0.4 0.6 0.8 1

f(x), F(x)

Ad hoc Ad hoc Supported ad hoc Supported ad hoc Mesh Mesh Ad hoc mesh Ad hoc mesh

Delay Suburban

5 flows DIVR

(d) DIVR Suburban.

Fig. 7.7:Delay distributions for 5 flows

7.5.7 Discussion

First let us consider figure 7.7 that illustrates the delay distributions when we have 5 bidirectional 64 kbps UDP flows. The most obvious difference we can observe

7.5. Simulations 143

10 20 30 40 50

Delay (ms) 0

0.2 0.4 0.6 0.8 1

f(x), F(x)

Ad hoc Ad hoc Supported ad hoc Supported ad hoc Mesh Mesh Ad hoc mesh Ad hoc mesh

Delay Urban

20 flows AODV

(a) AODV Urban.

10 20 30 40 50

Delay (ms) 0

0.2 0.4 0.6 0.8 1

f(x), F(x)

Ad hoc Ad hoc Supported ad hoc Supported ad hoc Mesh Mesh Ad hoc mesh Ad hoc mesh

Delay Urban

20 flows DIVR

(b) DIVR Urban.

10 20 30 40 50

Delay (ms) 0

0.2 0.4 0.6 0.8 1

f(x), F(x)

Ad hoc Ad hoc Supported ad hoc Supported ad hoc Mesh Mesh Ad hoc mesh Ad hoc mesh

Delay Suburban

20 flows AODV

(c) AODV Suburban.

10 20 30 40 50

Delay (ms) 0

0.2 0.4 0.6 0.8 1

f(x), F(x)

Ad hoc Ad hoc Supported ad hoc Supported ad hoc Mesh Mesh Ad hoc mesh Ad hoc mesh

Delay Suburban

20 flows DIVR

(d) DIVR Suburban.

Fig. 7.8: Delay distributions for 20 flows

here is the significantly lower delays for the DIVR ad hoc scenarios. In fact, here the cumulative probability starts approaching one for packet delays at around 6-7 ms. If we now also look at the 2nd and 4th tables in table 7.1 and table 7.2, we can see that the average delay for DIVR is around 2 ms independent of the amount of traffic, and whether the environment is urban or suburban. So, the conclusion here is that for the ad hoc scenarios, the delay DIVR is fairly independent of the type of environment and the amount of traffic.

If we consider figure 7.7(c) we can see that the delay for AODV (suburban) especially for the pure ad hoc case is significantly higher than for any of the sce-narios and environments for 5 bidirectional flows. A look at table 7.2 reveals that the average delay for this case is 16.7 ms. For the AODV urban ad hoc case, the delay is (see table 7.1) 6.2 ms. With a higher traffic load, see figure 7.8 , table 7.2 and 7.1, the effect is much more severe with a delay of 195ms and 640ms for 25 AODV urban and suburban flows.

What we see here is the effect of the environment itself, how the height of the buildings affects the signal path and the performance. In a city environment, build-ing walls completely block the signal from one parallel street to another, while in

Tab. 7.1:URBAN ad hoc scenarios. Delivery ratio with standard deviation. Delay in ms (d), Jitter in ms (J), Probability of delay less than 50ms (P), Mean Opinion Score for G.711 (MOS), MOS for enhanced G.711 (MOS2), Battery lifetime in hours for 1200mAh (B)

the simulated suburban environment the signal is not completely blocked but is still affected by multi-path fading. The main difference this has on the MAC layer is how carrier sensing are affected and hidden terminals are created. In the subur-ban environment, carrier sensing is possible across a block, but not in the ursubur-ban environment. In the suburban environment, RTS and CTS packets will protect a 802.11 transmission from parallel transmitters, which increases the time it takes for a packet to access the channel. But since some links now experience non line of sight multi-path propagation fewer, packets will also be delivered on average.

For the supported ad hoc scenarios, see table 7.1 and 7.2. When a relay mesh node is placed at each intersection, the delay increases for low AODV traffic in the urban environment, while it decreases for the suburban environment. An important factor here is the connectivity and medium contention on the routes. In the urban environment, the best path across a few blocks will always pass through a relay

7.5. Simulations 145

Tab. 7.2:SUBURBAN Ad hoc scenarios. Delivery ratio with standard deviation. Delay in ms (d), Jitter in ms (J), Probability of delay less than 50ms (P), Mean Opinion Score for G.711 (MOS), MOS for enhanced G.711 (MOS2), Battery lifetime in hours for 1200mAh (B)

point, which increases the contention for those nodes and therefore the delay. In the suburban environment they increase the connectivity of the network, but all routes doesn’t necessarily pass through them, resulting in a lower delay. With very high traffic, it is more complicated. Now links, or routes, may be reported as broken due to collisions, when in fact they are not. When the route is then repaired and resetup, RREQ packets are rebroadcasted by neighboring nodes. With a higher connectivity, more packets will be rebroadcasted, increasing the probability for collisions, and the delay. In the urban environment, the rebroadcasting collision effect is not high enough to decrease performance and overcome the positive effect of the higher connectivity. The delay is thus lower for the urban environment than for the suburban environment.

If we look at the results for mesh scenarios, table 7.4 and 7.3, the biggest differ-ence between the urban and suburban environments are the much higher suburban

Tab. 7.3:URBAN Mesh scenarios. Delivery ratio with standard deviation. Delay in ms (d), Jitter in ms (J), Probability of delay less than 50ms (P), Mean Opinion Score for G.711 (MOS), MOS for enhanced G.711 (MOS2), Battery lifetime in hours for 1200mAh (B)

delivery ratios. For low traffic rates, both AODV and DIVR manages to sustain high delivery ratios in the suburban environment, but AODV more or less main-tains these ratios for a higher number of flows than DIVR.

For the urban environments, the delivery ratio isn’t very high, the maximum delivery ratio for any urban mesh scenario is 75%. In fact, if we compare table 7.3 and 7.1 we see that for low traffic rates, the ad hoc scenarios actually perform better, which is not true for suburban (tables 7.4 and 7.2). We can now also take a look at the different Mean Opinion Scores. To at least have some form of acceptable VOIP experience in an urban environment, the enhanced G.711 codec should be used. We can also make the interesting conclusion, that for urban environments, it is actually better to use the ad hoc technologies, with the supported ad hoc network performing slightly better. If we also look at the battery lifetimes, we see that the lifetimes are significantly longer for the ad hoc scenarios. This is an interesting, and

7.5. Simulations 147

Tab. 7.4:SUBURBAN Mesh scenarios. Delivery ratio with standard deviation. Delay in ms (d), Jitter in ms (J), Probability of delay less than 50ms (P), Mean Opinion Score for G.711 (MOS), MOS for enhanced G.711 (MOS2), Battery lifetime in hours for 1200mAh (B)

somewhat unexpected result. Even though the mesh network operate on separate channels than the user nodes, we don’t really gain anything by using their extra interfaces in a harsh urban environment. With a different mesh configuration, and by using more interfaces in each mesh point, or a different physical layer with a higher capacity, this will probably change. But for this configuration, with the same physical layer on both user nodes and mesh nodes, we can’t see any significant gain for urban environments. We leave it for future work to study the needed capacity, and the various dependent factors, for a mesh network to outperform an ad hoc network in an urban environment.

Continuing looking at battery lifetimes, for the ad hoc scenarios we see that DIVR achieve its longest for the suburban environment. For AODV however, the longest lifetimes are achieved for the urban environment.

So, when looking at all the tables, we can see that the best VOIP MOS

perfor-mances are for suburban mesh scenarios. If we limit the number of flows to 20, we can still achieve very high VOIP performance by using an ad hoc mesh topology instead of using a single hop mesh. We can do this by maintaining the same battery lifetime, and with a cheaper infrastructure.

The MOS VOIP performance in the urban scenarios are always a bit lower than for suburban. With the consideration of the higher lifetime, in combination with comparable MOS performance, it seems better to use a supported ad hoc network for urban environments.

In conclusion we can say that what type of protocol that is optimal for a certain situation, depends on the environment, the type of network and the amount of traffic.

I dokument Wireless Multi Hop Access Networks and Protocols Nilsson Plymoth, Anders (sidor 148-157)