E N D
Wireless networks simulation Performance evaluation of a protocol for ad hoc networks is usually performed by simulating the wireless network. Simulation provides the researcher with a number of significant benefits, including repeatable scenarios, isolation of parameters and exploration of a variety of metrics. In particular, A Wireless Network simulation MUST model: Mobile Nodes’ characteristics such as: transmission range, limited buffer capacity, battery power limitations, signals radio propagation ect. Communication traffic model (what kind of data flow is there?) Mobility model (movements of the users i.e. devices) a few others • This presentation will focus on the importance of the MOBILITY PATTERN when simulating ad-hoc networks [1] ...
Trace vs. Synthetic mobility models (1) • Trace • Are those mobility models observed in real life scenarios. For instance, if a mobile phone carrier had the ability to trace the exact movements and behaviors of all its customers carrying their phone for a given period of time, a trace would be obtained. • Pros • Accurate information gathered especially for scenarios with many nodes • Cons • Privacy issues may prohibit their collection and distribution • Simulations can’t be run when new environments haven’t even been created yet. • Synthetic (used when tracing is not possible) • Models which attempt to realistically represent the behaviors of the MNs without use of traces.
Trace vs. Synthetic mobility models (2) • Tracing approach in Ad-Hoc vs. GSM • Requires an additional effort to be performed in advance. • Currently there exist traces for GSM but not for Ad-Hoc Networks. • While in GSM networks it is enough to trace the movements between a cell to another, in Ad-hoc networks is required to trace the mobility with a strict precision during the motion. Hence, a precise positioning system is needed on each device to be traced. As long as GPS is not commonly provided on these devices (Laptop, PDA) the tracing is rather tricky to obtain. • Synthetic • For the several reasons mentioned, the Synthetic approach is the only one which researcher can currently follow with ad-hoc. We are going to focus on these mobility models that are commonly used.
Entity MMs • A node’s movement does not influence in anyhow other nodes’ movements. Nodes move independently from each other. • Random Walk ( & its probabilistic version) • Random Waypoint • Random Direction • Gauss-Markov • City section mobility models Synthetic mobility models classification • Group MMs • represent MNs whose movements are dependent. Used when MNs collaborate together to accomplish a common goal. Typical situations do exist in military environments (soldiers move together)… • Column MM • Nomadic Community • Pursue • Reference Point Group MM
It wants represent the movement of the entities in an unpredictable way. In particular, a node moves from its current location to a new one by randomly choosing: • Direction between [0,2pi] • Speed between [MinSpeed, MaxSpeed] • Either duration of movement tm OR distance d • Direction and speed are both uniformly distributed • A node which “crashes” against the boundary keeps on moving on an opposite direction between [0,pi] depending on the incoming one. Random Walk (1) OBSERVATIONS: 1) Nodes start moving at t=0. Choosing a DURATION implies that all the nodes change directions at the same time and travel for different distances. In contrast, choosing a DISTANCE implies same distances but different duration. 2) The pattern is memory-less i.e. current speed and direction do not depend upon the previous ones. Therefore, there will be sharp and sudden turns. 3) Short tm or d lead the nodes to move around their current location. Unless it is necessary to study a semi-static network, they MUST be chosen large.
Random Walk (2) Example of a travelling pattern of a mobile node using the 2D Random Walk MM Surface size 300 x 600 m, tm = 60s
It is a variation of Random Walk. It introduces the concept of pause time. A node randomly chooses (Parameters uniformly distributed): • Pause Time (to wait before resuming the movement) [Pmin,Pmax] • Direction [0,2pi] • Speed [Minvel,Maxvel] • Destination point (x,y) to reach.. Random Waypoint (1) • OBSERVATIONS: • 1) the duration of the movement depends on the destination point chosen. • 2) nodes do not start roaming all together unless Pmin = Pmax = 0. • 3) the pattern becomes a Random Walk when (Pmin = Pmax = 0 ) AND ([Minvel,Maxvel] = [MinSpeed, MaxSpeed]) • 4) it is the most commonly used MM in ad-hoc network simulation studies (often times is modified). • It needs particular attention to choose the initial locations. Discard the initial part of the simulation OR save the node’s location OR do not place the nodes randomly. • The choice of Pauses and Speeds is relevant. Fast nodes and long pauses produces a more stable network than slow nodes and short pauses. • The most argued issue is that nodes are more likely to be in the central part of the topology rather than close to the bounds.
Random Waypoint (2) Example of a travelling pattern of a mobile node using the Random Waypoint MM • Clearly the motion is centrally happening • Nodes appear to converge, disperse and converge again • Nodes tend to have many neighbors when in the center and almost none when they disperse. Surface size 300 x 600 m
Designed to overcame the concentration of the nodes of which R. Waypoint suffers. Nodes start moving by choosing • Direction between [0,2pi] • Speed between [MinSpeed, MaxSpeed] • Nodes will travel till the bound is reached. On this position they’ll stand for a pause time before leaving to a New-Direction [0,pi]. Random Direction (1) • OBSERVATIONS: • 1) Nodes are forced to basically stay away from the center for the most of the time. In fact, they ALL pause somewhere on the perimeter. • Implications: • Average Hop count for Data-packets will be much higher than in R. Waypoint or R Walk. (nodes are on average far from each others) • Higher probability to have Network partition (especially with few MNs).
Random Direction (2) Example of a travelling pattern of a mobile node using the Random Direction MM Surface size 300 x 600 m
It creates movements which are dependent on node’s current speed and direction. The idea is to eliminate the sharp and sudden turns present in the R. Walk and R. Waypoint even by keeping a certain degree of randomness. At fixed intervals of time n new direction dn and speed snare chosen as: Gauss-Markov (1) • In addition… • dn and snkeep into account a parameter to tune the level of randomness in making the decisions • the desired statistical distributions SX and DX , for the random variables Speed and Direction MUST be chosen beforehand • some “tricks” are used to force the MNs to stay away from the edges
Gauss-Markov (2) Example • I describe the implementation used in [2]: WHERE: • is the tuning parameter chosen in [0,1] • , are constants. The mean values for SX and DX (both Gaussian) • sXn-1 and dXn-1 are the random variables from SX and DX • Observations • with the movements are totally random, with they are linear. • the trick to avoid the edges is to choose differently when near to the edge
Gauss - Markov (3) Example of a travelling pattern of a mobile node using the Gauss Markov MM • n=1 sec • sXn-1 and dXn-1 are chosen from a gaussian distribution with mean 0 and std.dev. 1 • is fixed at 10m/s • chosen according to the current position Surface size 300 x 600 m
Probabilistic Random WALK • The movements of the nodes are bases on a matrix of probabilities P. • x and y coordinates vary according to a state chart with 3 states: • State 0: the coordinate does not vary • State 1: the coordinate decreases i.e. Step Backward • State 2: the coordinate increases i.e. Step Forward The probabilities to switch from a state to another are: Each non zero probability is a transition in the state chart. A possible implementation is shown in Chiang [3] (speed const.)
Boundless simulation area • Sometimes can be interesting to discard the border of the simulation area. In fact, the perimeter of it does affect nodes when they reach it. • Nodes move on the torus internal surface • It is not any longer needed to describe the node’s behavior on the border. • Radio signal propagation must be modelled accordingly
City mobility model • in addition it could include • safe driving characteristics such as speed limit, minimum distance allowed between pairs of nodes. • High speed road along the border of the simulation area • A boundless topology to represent the whole city it represents a section of a city where an ad-hoc network operates [4]. It models factors as: A street network A set of buildings Destination points ( where nodes randomly start from and hear for ) • Pros • high realistic motion • Cons • tricky to fully describe
Column MM it is a group MM, Suitable for representing soldiers marching: Initially a reference point is chosen and assigned to each MNs. The peculiarity is to choose points on a line (culumn) Nodes are subsequently allowed to move around their reference point according to an Entity MM Reference points change: New_ref_point = Old_ref_point + advance_vector where: advance_vector = (x,y) When this happens, MNs move toward their ref. Point to start roaming around.
Nomadic community MM it is a group MM, Suitable for representing Nomadic Movements (a class of students visiting a museum ect.): Initially a reference point is chosen and SHARED between all the MNs. Nodes are subsequently allowed to move around it according to an Entity MM the reference point randomly changes causing the nodes to firstly reach it and then to roam around it Column vs. Nomadic community MM • Culumn • A ref. point per MNs • strict motion around (short trips, often changes in dir. & speed • Nomadic C. • A unique ref. Point • more space to roam (longer lasting movements)
A set of MNs want to catch a running away MN. NewPosition = OldPosition + Dist + RandomDist Pursue MM Dist is a vector (x,y) whose components are chosen in [MinDX, MaxDX] and [MinDY, MaxDY]. RandomDist is a vector (x,y) obtained via Entity MM.
A set of MNs want to move in group. The group has a logical center which moves. Each MN is assigned a moving reference point. Nodes randomly move around it. Two kinds of motion: i) Group Motion is represented by a vector GM ii) nodes Random Motion is represented by RM The group logical center is assigned a new position at regular intervals. Subsequently the RPs locations are updated accordingly. Finally MNs locations are computed based on GM and RM Reference Point Group MM (1) • GM can be EITHER predefined OR randomly chosen • RM is randomly chosen. The direction is uniformly distributed in [0,2pi], the length as well with a radius centered at the Reference point’s location.
Designed to represent avalanche rescue. (humans & dogs) The MNs movements are characterized by the group logical center’s motion. Many particular possible implementations. For instance: 1) Nomadic Community MM (no separation between RPs) 2) Column MM (by disposing the RPs in a column) 3) Pursue MM (no separation between RPs) Reference Point Group MM (2)
Simulation to compare the MMs • In [1] they have done a few simulation experiments to evaluate and compare: • Random Walk • Random Waypoint • Random Direction • RPGM ( Between-groups vs. Intra-group + Between-groups • Simulation Parameters: i) # nodes 50 ii) Routing Prot. DSR iii) Sim. Time 1000 sec iv) Random initial locations v) Sim. Results are shown by averaging 10 trials • Communication model: i) 20 UDP pairs S/R (CBR) ii)rate 1pck/s iii) pkt size 64 bytes each * RPGM Inter+Intra sends: -1pkt/2sec Inter (20 pairs distributed in 16 groups) -1pkt/5sec Intra.
Performance metrics • They evaluated DSR’s performance by measuring: • data packet delivery ratio Rcv/Snd • end-to-end delay • average hop count • protocol overhead • Their objective is to show how the protocol performance changes depending on the mobility model under study. In particular, they observe the significant metrics vs. average speed value.
The results obtained (1) • R. waypoint is the model which less stresses DSR. They observed: highest data packet delivery ratio, lowest end-to-end delay, lowest average hop count. Motivation: ‘cos nodes often travel through (or to) the center. The model is Idealistic rather than Realistic! • In contrast, Last ranked is R. walk. Lowest data packet delivery ratio, Highest end-to-end delay, highest average hop count were observed. Reason: nodes move to the boarder and stop there i.e. nodes are on average far apart. Network partition are more likely to occur. • DSR performance with other Entity MMs are ranked between the presented extremes.
The results obtained (2) • RPGM pure inter-group communication has roughly the same Average Hop Count than R-waypoint. Motivation: both MNs and RPs move according to R-waypoint MM. • In contrast, it has much higher end-to-end delay and lower data packet delivery ratio. Reason: Since only 16 groups exist, the network will be sparser than R-waypoint with 50 nodes. Furthermore, network partitions probably occur. • RPGM intra+inter comm. Has the lower average hop count and a high Delivery ratio. The half data sent intra group, are raising the performance!
The results obtained (3) Routing ovehead i) # control packets / Pck_received ii) # control byte transmitted / Pck_received it includes control bytes in both control and data packets. • They noticed that the average hop count is related with the routing overhead. In fact:- GRPM had the lowest average hop count, so that’s the explanation for the model lower routing overhead. - Likewise, R. walk and R. waypoint had a high hop count so they have high routing overhead as well.
Conclusions - Importance of the MM • The protocol performance vary significantly depending upon the MM being used. It definitely affects a lot! • The PP vary also in function of the MM’s parameters. Same MM with different input parameters leads to observe diverse findings. • GMM: Intra-group communication drastically increases the PP. Inter-group communication is more tricky to handle. • The MM should be chosen in a way such that more closely represents the situation being simulated. Difficulty to know what is a real-life case. No traces available for ad-hoc networks. • When the expected real-life scenario is unknown, performance must be evaluated under multiple MMs.
References [1] Tracy Camp, Jeff Boleng, Vanessa Davies. A survey of mobility for ad hoc network research. Dept. of Math. and Computer Sciences Colorado School of Mines, September 2002.[2] V. Tolety. Load reduction in ad hoc networks using mobile servers. Master’s thesis, Colorado School of Mines, 1999. [3] C. Chiang Wireless Network Multicasting., Phd. Thesis, University of California, Los Angeles 1998.[4] F. Bai, N. Sadagopan, A. Helmy, A framework to systematically the impact of mobility on performance of routing protocols for AdHoc networks. IEEE Infocom 2003.
Thank you! Any question? Write to: Stefano Marinoni marstefy@tcs.hut.fiOffice: T-B235 lab for theoretical CS