5.2 Simulation results
5.2.1 Results from each evaluation case
5.2.1.1 With LB vs. without LB
Load balancing with handovers has the potential to improve system wide Resource Utilization by distributing the load to the less congested BSs in the system. Fig- ures 5.2 a-b present overall results from the first evaluation case consisting of two simulation runs, one where the basic load balancing scheme was used and another where no load balancing was conducted.
20 40 60 80 100 120 140
35 40 45 50 55 60 65 70 75
Time/seconds
Resource utilization/%
(a) Resource utilization for BS2
Without LB With LB
20 40 60 80 100 120 140
0 0.2 0.4 0.6 0.8 1
Time/seconds
Load balance index
(b) Load balance index of the system
Without LB With LB
Figure 5.2: Resource Utilization of BS 2 (a) and the load balance index (b) with and without load balancing.
When load balancing was not used no directed handovers were conducted (no Re- source Utilization decreases in BS 2) and as a result admission control had to block 19 new non-BE (VoIP based) calls in the congested BS 2 whereas with the basic load balancing algorithm the non-BE load was distributed to the other BSs and no new calls had to be blocked in BS 2. Also we can observe that the load balancing algorithm was able to keep the load balancing index close to the target value 1 throughout the simulation whereas without load balancing the index resulted in a
value of 0.9 even when a large portion of the non-BE load was blocked.
To get a better understanding of what actually happens in the system, the Up- link and Downlink Resource Utilizations for all three Base Stations are depicted for the simulation run without load balancing in Figures 5.3 a-f and for the simulation run with load balancing in Figures 5.4 a-f.
The Figures 5.3 a-f and 5.4 a-f present four Resource Utilization curves: one for subframe headers corresponding to the DL- and UL-MAPs in the DL subframe and the contention and bandwidth slots reserved in the UL subframe, one for non-BE data corresponding to the VoIP based data traffic (including corresponding MAC headers and management messages), one for BE data corresponding to the FTP and HTTP data in the DL and acknowledgments in the UL and one for the total Resource Utilization10.
20 40 60 80 100 120
0 10 20 30 40 50 60 70 80 90 100
(a) UL Resource utilization for BS1 without LB
Time/seconds
Resource utilization/%
20 40 60 80 100 120
0 10 20 30 40 50 60 70 80 90 100
(b) UL Resource utilization for BS2 without LB
Time/seconds
Resource utilization/%
20 40 60 80 100 120
0 10 20 30 40 50 60 70 80 90 100
(c) UL Resource utilization for BS3 without LB
Time/seconds
Resource utilization/%
20 40 60 80 100 120
0 10 20 30 40 50 60 70 80 90 100
Time/seconds
Resource utilization/%
(d) DL Resource utilization for BS1 without LB
Overhead Non-BE data BE data Total
20 40 60 80 100 120
0 10 20 30 40 50 60 70 80 90 100
(e) DL Resource utilization for BS2 without LB
Time/seconds
Resource utilization/%
20 40 60 80 100 120
0 10 20 30 40 50 60 70 80 90 100
(f) DL Resource utilization for BS3 without LB
Time/seconds
Resource utilization/%
Figure 5.3: UL (a-c) and DL (d-f) Resource Utilizations when no load balancing used.
As can be seen the UL subframe is clearly the bottle neck for non-BE traffic and will determine the final Resource Utilization value as defined in equation (3.3), used to trigger load balancing. We can see that as more MSs are dropped to BS 2, the non-BE data Resource Utilization gradually increases until admission control starts
10Note that the Resource Utilization reported in the Spare Capacity Report will be a sum of header and non-BE data Resource Utilization.
to block new VoIP based flows. This happens because a limit in the uplink Resource Reservation has been reached after which the QoS of the existing VoIP calls would degrade. The 19 non-BE VoIP based calls blocked result to a 19 % blocking rate in BS 2 during the simulation, as 100 VoIP based MSs were dropped to BS 2. Despite of the call blocks we can see that at the end of the simulation the system is still quite unbalanced with about a 30 % difference in the uplink Resource Utilization.
In addition what is very interesting here is that even though some bandwidth is left for the acknowledgments of BE traffic in BS 2, the slight decrease in the UL BE throughput results in quite a large drop in the downlink BE throughput.
20 40 60 80 100 120
0 10 20 30 40 50 60 70 80 90 100
(a) UL Resource utilization for BS1 with LB
Time/seconds
Resource utilization/%
20 40 60 80 100 120
0 10 20 30 40 50 60 70 80 90 100
(b) UL Resource utilization for BS2 with LB
Time/seconds
Resource utilization/%
20 40 60 80 100 120
0 10 20 30 40 50 60 70 80 90 100
(c) UL Resource utilization for BS3 with LB
Time/seconds
Resource utilization/%
20 40 60 80 100 120
0 10 20 30 40 50 60 70 80 90 100
Time/seconds
Resource utilization/%
(d) DL Resource utilization for BS1 with LB
Overhead Non-BE data BE data Total
20 40 60 80 100 120
0 10 20 30 40 50 60 70 80 90 100
(e) DL Resource utilization for BS2 with LB
Time/seconds
Resource utilization/%
20 40 60 80 100 120
0 10 20 30 40 50 60 70 80 90 100
(f) DL Resource utilization for BS3 with LB
Time/seconds
Resource utilization/%
Figure 5.4: UL (a-c) and DL (d-f) Resource Utilizations when load balancing is used.
When the same simulation is run with load balancing we can see that as the non- BE Resource Utilization increases the load is distributed to the less congested Base Stations, BS 1 and BS 3 and hence no calls need to be blocked. A 10 % hysteresis value is used, so each time the Resource Utilization of BS 2 surpasses the average Resource Utilization in the system with 10 percent of the average, the BS initiates directed handovers for the non-BE MSs in the overlapping areas. In this simulation run load balancing is initiated three times at around 16, 76 and 95 seconds. The directed handovers can be seen as three steep drops in the Resource Utilization of BS 2 and corresponding increases in BS 1 and BS 3 (clearly in the UL and to some degree also in the DL).
We can see that at the end of the simulation run the system is quite well bal- anced. The slight unbalance is due to the hysteresis margin. What can also be observed is that since the BE connections in BS 2 have enough bandwidth to send the acknowledgments the BE DL Resource Utilization doesn’t decrease as in the simulation run without load balancing, but is here reduced only slightly as non-BE data Resource Utilization increases. We can recognize this difference also in the total system DL FTP throughput presented in Figure 5.5.
20 40 60 80 100 120
26 27 28 29 30 31 32 33 34 35
Time/seconds
Throughput/Mbps
Average DL FTP throughput in the system
FTP w/o LB FTP w/ LB
Figure 5.5: System wide downlink FTP throughput with and without load balanc- ing.
From the Figure 5.5 we can see that after the UL gets too congested in the sim- ulation run without load balancing, a clear decrease in FTP downlink throughput can be seen in the whole system when compared to the simulation run where load balancing is used.
This simulation case has shown that load balancing can in fact have a rather large impact on the system wide Resource Utilization in case of congestion and is there- fore a valid method. In the next part we will evaluate the behavior of the system as a function of the hysteresis margin.