The performance of the BA-TLS framework has been experimentally evaluated using LTE-Sim [34], an open source simulator for LTEnetworks. All simulations run for 120 secs. A summary of the main simulation parameters are presented
4.3 Experiments and Results
0 5 10 15 20 25 30 35 40 45
10 20 30 40 50 60 70 80 90 100
Avg Buffering %
#Video Flows BA-TLS-Optimal
BA-TLS-Genetic BA-TLS-PBRA TLS
Figure 4.4: Avg Buffering % Vs #Video Flows
in table 4.1. A series of experiments have been conducted in order to measure the performance achieved by the proposed strategy. The obtained performance results have also been compared against theThree Level Scheduling (TLS) frame- work. The specific performance metrics which have been considered for evaluation are: (i) Average buffering (%), (ii) Instantaneous playout buffer status and (iii) Average execution time.
Figure 4.4 depicts plots for the average buffering suffered by the different resource allocation strategies against varying number of video flows (or system load) during the entire simulation length. It may be noted that average buffering (%) for all the strategies increases as the number of flows/system load increase.
This is expected because the average number of RBs that may be allocated for a video flow reduces as the total number of video flows increases under a fixed resource block budget within a super-frame interval. Although, trends for all the methodologies in Figure 4.4 are similar, BA-TLS-Optimal is seen to encounter less buffering events compared to both the heuristic strategies, namely, BA-TLS- Genetic and BA-TLS-PBRA, while the basicTLS algorithm performs poorly in all cases. The reason for the poor performance of basic TLS originates from its ignorance of client side buffer status during the resource allocation process. On the other hand, the endeavour to maintain stable playout buffer sizes for each flow
0 5 10 15 20 25 30
0 20 40 60 80 100
Playout Buffer Size
Time (Secs) Playout Buffer Size (Secs)
(a) BA-TLS-Optimal
0 5 10 15 20 25 30 35
0 20 40 60 80 100
Playout Buffer Size
Time (Secs) Playout Buffer Size (Secs)
(b) BA-TLS-Genetic
0 5 10 15 20 25
0 20 40 60 80 100
Playout Buffer Size
Time (Secs) Playout Buffer Size (Secs)
(c) BA-TLS-PBRA
0 5 10 15 20 25 30 35
0 20 40 60 80 100
Playout Buffer Size
Time (Secs) Playout Buffer Size (Secs)
(d) TLS
Figure 4.5: Instantaneous playout buffer size achieved by BA-TLS-Optimal, BA-TLS- Optimal and TLS strategies
considerably reduces rebuffering events in the proposed buffer aware schemes. In Fig 4.4, BA-TLS-Genetic and BA-TLS-PBRA are seen to suffer slightly higher average buffering with respect toBA-TLS-Optimal, due to their inherent heuristic nature.
Figures 4.5(a) to 4.5(d) shows instantaneous buffer sizes achieved by BA- TLS-Optimal,BA-TLS-Genetic,BA-TLS-PBRAand TLS strategies respectively, for a single flow (namely, Star Wars) over the entire simulation duration. The scenario considers a cell with 100 active flows. It may be observed from the figures that the buffer-aware strategies, namelyBA-TLS-Optimal,BA-TLS-Genetic and BA-TLS-PBRA are able to maintain approximately stable buffer sizes for the flow during the entire simulation duration. Stable playout buffer sizes in BA- TLS is achieved by two principle mechanisms: (i) Providing a certain degree of robustness to each flow against varying channel conditions and (ii) Auto tunning the priority of the flow during resource allocation based on its instantaneous
4.3 Experiments and Results
Table 4.2: Comparative results for average run time (in millisecs)
# FlowsBA-TLS-Optimal BA-TLS-Genetic BA-TLS-PBRA
10 144.2 4 0.016
20 297.7 6.7 0.024
30 447.3 9.3 0.025
40 600.8 12.4 0.029
50 744.8 15.3 0.031
60 902.1 18.6 0.035
70 1061 21.5 0.038
80 1210.2 24.3 0.04
90 1360.2 27.4 0.041
100 1523.4 29.6 0.049
playout buffer size and received CQI feedback (i.e. assigning relatively higher reward values to flows having comparatively lower playout buffer sizes and/or CQIs).
Table 4.2 shows the average execution time (in millisecs) taken by BA-TLS- Optimal,BA-TLS-Genetic andBA-TLS-PBRAas the number of video flows vary from 10 to 100. It may be observed from the table that the average execution time required for BA-TLS-Optimal is comparatively much higher than the BA- TLS-Genetic and BA-TLS-PBRA strategies. This happens because BA-TLS- Optimal calculates partial solutions for all possible bounds on number of flows (∀i∈[0, N]), robustness levels (∀l∈[0, Li]) and RBs (∀β ∈[0, B]). On the other hand, the stochastic strategy BA-TLS-Genetic is observed to generate good and acceptable solutions much quicker as compared to the optimal strategy. This is expected because the BA-TLS-Genetic tries to locate a globally optimal solution after examining a limited number of candidates in the solution space. Therefore, theBA-TLS-Genetic strategy may be seen to achieve good speed-ups (∼40 to 50 times) with respect to the BA-TLS-Optimal strategy. On the other hand, it may be observed that BA-TLS-PBRA perform far better than BA-TLS-Optimal and BA-TLS-Genetic in terms of computational overhead. This is because the com-
Table 4.3: Comparative results for speed-ups achieved by the proposed heuristics
# FlowsOptimal Vs Genetic Optimal Vs PBRA Genetic Vs PBRA
10 36 9013 250
20 44 12404 279
30 48 17892 372
40 48 20717 428
50 49 24026 494
60 49 25774 531
70 49 27921 566
80 50 30255 608
90 50 33176 668
100 51 31090 604
putational complexity of BA-TLS-PBRA only depends on the available number of flows and the number of available robustness level. Therefore,BA-TLS-PBRA is able to achieve drastic speed-ups with respect to the other two strategies.
Comparative results for achieved speed-ups are shown in Table 4.3.