Pheromone based Mobile Agent Migration Mechanisms
1) Random Approach
4.6 Results and Discussions
4.6.1 Static Scenario
The parameters used for Random, Conscientious, EVAP, CLInG, G-B, PherCon and PherCon-C in the simulation of static scenarios were as follows:
Total Number of Nodes in the network=200, Total Number of Links=
1624. The values of the additional parameters used for PherCon and PherCon-C are given below:
d=10, Cmax=100, ΔC=20, Tmax=20, ΔT=2, δ =20.
Case (I) Single Mobile Agent and Four RRSs:
In this scenario, the network was populated with just one mobile agent and simulation commenced with the existence of four RRSs.
For each of the seven strategies viz. Random, Conscientious, EVAP, CLInG, G-B, PherCon and PherCon-C simulations were carried out five times.
In the G-B strategy this single mobile agent cloned to increase its population.
The graph in Figure 4-3 depicts the number of steps taken by each of the seven strategies while those in Figure 4-4 and Figure 4-5 show the energy in terms of intra-node computations and inter-node communications.
Fig. 4-3 Graph depicting Number of Step-Counts Versus Runs for Static
Scenario: Case (I) - Single Agent and Four RRSs obtained for five individual runs
Fig. 4-4 Graph depicting Number of Intra-Node Computations Versus Runs
for Static Scenario: Case (I) - Single Agent and Four RRSs obtained for five
individual runs
1538 4042 960 1100 2916
309358 208267 123191 188110 195417663496 5728376 2005644 818424 586032
65676 55572 67360 60624 75780
41908874722 41908874722 41908874722 41908874722 41908874722
31674 31652 32127 31601 32325
9294 8285 8418 10128 10135
1 10 100 1000 10000 100000 1000000 10000000 100000000 1E+09 1E+10 1E+11
1 2 3 4 5
Number of Intra-Node Computation (Y-axis in Logarithmic Scale)
Runs
Static: Case (I) - Single Agent and Four RRSs Intra-Node Computations Vs Runs
Random Conscientious EVAP CLInG G-B PherCon PherCon-C
1538
4042
960 1100
2916
303 237 171 224 231
394 2214 1191 486 348
39 33 40 36 45
10 10 10 10 10
21 21 24 20 25
19 19 21 20 26
1 10 100 1000 10000
1 2 3 4 5
Number of Step-Counts (Y-axis in Logarithmic Scale)
Runs
Static: Case (I) - Single Agent and Four RRSs Step-Counts Vs Runs
Random Conscientious EVAP CLInG G-B PherCon PherCon-C
Fig. 4-5 Graph depicting Number of Inter-Node Communications Versus Runs for
Static Scenario: Case (I) - Single Agent and Four RRSs obtained for five
individual runs
From the graph in Figure 4-3 it can be observed that the Random, Conscientious, EVAP and CLInG strategies consume far more number of steps than the rest. The G-B strategy consistently takes minimum number of steps followed closely by PherCon and PherCon-C. The maximum difference between the former and the latter two is just about 16 steps which is in contrast with the large differences in case of the other strategies.
The graphs in Figure 4-4 and 4-5 reflect the proportional energy consumed by each of these strategies in finding and servicing the four RRSs within the network. These graphs present a contrary performance to that reflected in the previous graph. As seen from them, the G-B strategy consumes most making it in no way energy efficient. The Conscientious, EVAP and CLInG seem to be energy efficient but do not fare as well as PherCon or PherCon-C in terms of time. The Random migration strategy is more energy efficient since both intra- node computations and inter-node communications are minimal but performs the
1538 4042 960 1100 2916
303 237 171 224 231
663466 5728376 2005644 818424 586032
131352 111144 134720 121248 151560
41908881256 41908881256 41908881256 41908881256 41908881256
8025 8025 8306 8025 85838025 5435 5435 9409 9582
1 10 100 1000 10000 100000 1000000 10000000 100000000 1E+09 1E+10 1E+11
1 2 3 4 5
Number of Inter-Node Computation (Y-axis in Logarithmic Scale)
Runs
Static: Case (I) - Single Agent and Four RRSs Inter-Node Communications Vs Runs
Random Conscientious EVAP CLInG G-B PherCon PherCon-C
worst in terms of the number of steps (time). When both time in terms of simulation steps and energy in terms of intra-node computations and inter-node communications is taken into consideration, it can be inferred that PherCon and PherCon-C perform the best.
Case (II) Multiple Mobile Agents and Four RRSs:
In this scenario, the network was populated with four RRSs and multiple mobile agents carrying the requested service. The number of mobile agents was varied from 2 to 12 to study the effect of an increase in agent population. Figure 4-6 through 4-8 depict the graphs of the results obtained. As can be seen, the graphs clearly indicate that an increase in number of agents hastens the service of the RRSs for all the strategies uniformly. A comparison of the performance of
each of the strategies however does not provide any extra information since the trend seems to be the same as discussed in Case-(I) both in terms of step-counts
Fig. 4-6 Graph depicting Number of Step-Counts Versus Runs for Static
Scenario: Case (II) - Multiple Mobile Agents and Four RRSs obtained for five
individual runs with 2, 3, 6, 9 and 12 Agents
315
833
108 132
109 211 31 40 100 29
218 148 47 18 12
20 7 11 9 3
8 5 4 4 3
10
7 5 5
4 10
6 5 5
4
1 10 100 1000
2 3 6 9 12
Number of Step-Counts (Y-axis in Logarithmic Scale)
Number of Agents
Static: Case (II) - Multiple Mobile Agents and Four RRSs Step-Counts Vs Number of Agents
Random Conscientious EVAP CLInG G-B PherCon PherCon-C
Fig. 4-7 Graph depicting Number of Intra-Node Computations Versus Runs for
Static Scenario: Case (II) - Multiple Mobile Agents and Four RRSs obtained for
five individual runs with 2, 3, 6, 9 and 12 Agents
Fig. 4-8 Graph depicting Number of Inter-Node Communications Versus Runs for
Static Scenario: Case (II) - Multiple Mobile Agents and Four RRSs obtained for
five individual runs with 2, 3, 6, 9 and 12 Agents
315 833 108 132 100
10272 58168 2967 7315 5163
367112 349232 79148 30312 2020833680 11788 18524 15156 5052
412622058
322001 56206 100043 9405
9874 3417 1697 3013 1709
4625 1110 779 791 541
1 10 100 1000 10000 100000 1000000 10000000 100000000 1E+09
2 3 6 9 12
Number of Intra-Node Computation (Y-axis in Logarithmic Scale)
Number of Agents
Static: Case (II) - Multiple Mobile Agents and Four RRSs Intra-Node Computations Vs Number of Agents
Random Conscientious EVAP CLInG G-B PherCon PherCon-C
315 833 108 132 100109 211 31 40 29
367112 349232 79148 30312 2020867360 23576 37048 30312 10104
412629273
325132 58922 103119 11553
6684 3624 2315 2902 1596
5435 2902 2315 2315 1596
1 10 100 1000 10000 100000 1000000 10000000 100000000 1E+09
2 3 6 9 12
Number of Inter-Node Computation (Y-axis in Logarithmic Scale)
Number of Agents
Static: Case (II) - Multiple Mobile Agents and Four RRSs Inter-Node Communications Vs Number of Agents
Random Conscientious EVAP CLInG G-B PherCon PherCon-C
and energy. The conclusion inferred earlier in Case-(I) that PherCon and PherCon-C perform better when both time and energy are taken into consideration, holds here too. However if energy is not a criterion, then the graphs in Figures 4-3 through 4-5 and Figures 4-6 through 4-8 may portray the G-B strategy to be the best. Comparisons between the G-B strategy and the proposed PherCon and PherCon-C together with their implementational issues given below, unravels several concerns which the graphs, presented herein, do not portray.
G-B versus PherCon and PherCon-C
From the graphs shown in Figure 4-3 and Figure 4-6, it can be seen that PherCon and PherCon-C take slightly more number of steps than the G-B strategy during simulation. However in the real world, the G-B strategy may take more time than PherCon or PherCon-C since the time required for one simulation step in G-B is greater than those of PherCon and PherCon-C. In the G-B strategy, cloning at each of the nodes could be assumed to be in parallel.
Cloning in G-B is carried out at the initial part of each step in the simulation.
Immediately after cloning, the actual times required for the agent and the clones to migrate to other nodes would be high and difficult to ascertain exactly. This is so because, in practice, during the time when an agent migrates from a certain node N1 to node N2, no other agent from any node in the network can migrate to either N1 or N2.This can cause agents to wait for their turn to migrate and contribute to the latency. If all the nodes contain agents ready for a migration (after cloning) then many migrations would be performed sequentially making the total time taken for all these agents to migrate to their respective destination nodes to be large. It may also be observed that if the G-B strategy were to be implemented in a real network, the actual number of intra-node computations and inter-node communications will increase exponentially for every subsequent time-step making the initial steps to consume far lesser time than the later ones.
This is so because of the explosion in the number of clones at subsequent nodes
having branches. The migration of these clones can cause a severe slow down as the execution progresses.
In a highly connected network topology, most inter-node communications would be sequential in nature, thus consuming more time than intra-node computations. Thus, though G-B takes far less number of steps in simulation, the total time required for their execution in the real world will be much more than the other strategies.
In PherCon and PherCon-C, the network is not cluttered with clones, which in turn allows for faster migration of agents between the nodes. The situation described for G-B will hardly ever occur in PherCon or PherCon-C due to the limited number of agents populating the network.
PherCon and PherCon-C, thus appears to be the best when a tradeoff between the time consumed and the energy expended in terms of intra-node computations and inter-node communications, is considered.
The problem of cluttering of the network has been further revisited, argued and a solution proposed in Chapter 6.
PherCon versus PherCon-C
Comparing PherCon and PherCon-C, it can be found that, with respect to step-counts, the latter is only slightly better than the former. Whenever an agent reaches a node where multiple pheromone requests for the same service from multiple RRSs are present, a marginal improvement due to PherCon-C is observed. If such a situation does not occur the agent behaves just as in PherCon migration strategy and performs accordingly with no improvement. With respect to intra-node computations and inter-node communication energy, PherCon-C performs similarly marginally better than PherCon.
Implementation Issues
Many an algorithm is portrayed to be the best using results obtained from simulations. Implementation issues play a major role in the viability of the use of the strategy or algorithm. Implementing G-B strategy accounts for a huge
number of clones to be generated that can hog the bandwidth of the network and subsequently degrade the performance in a real system.
In such systems, the agents populating the network could be heterogeneous and thus carrying different services as payloads. EVAP and CLInG seem to inherently support only homogeneous agents. In a heterogeneous agent scenario, the use of EVAP and CLInG would mean that each node in the network would have to maintain the idleness and propagated idleness values of each type of agent. More the number of distinct agents carrying different services more will be the overheads of maintaining these values in each node and propagating them to their immediate neighbours in every time step. An inherent assumption made in the simulation of EVAP and CLInG is that the exchange of the idleness values between all nodes and their respective neighbours is a single step parallel operation. As pointed out earlier this is not true in the real world.
The process of exchanging the values also calls for more bandwidth, thus affecting performance. Further the use of EVAP and CLInG would pose difficulties in scaling the system when agents carrying new services are introduced into the network on the run.