We conduct analytical results to evaluate the performance of the 802.11e EDCA medium access mechanism in terms of saturation throughput and access delay. More specifically, we would like to gain a better understanding on how the be- haviour of the QoS prioritization of high priority ACs over lower priority effect the overall performance, which sometimes could lead into starvation for the lower priority ACs. We use the IEEE 802.11a as an example and its parameters can be found in [6, 11]. The data rate is 54 M b/s and the control rate is kept at 6 M b/s. The packet lengths are fixed at 1, 500 bytes for each STA and the number of STAs varies according to the scenario in consideration. For simplicity reasons the RTS/CTS frame exchange as a collision protection mechanism has not been utilized as the derived objective conclusions are analogous, in any case it can be easily applied.
For our results we defined three numerical experiments where each STA de- ploys one backoff entity of one and only AC to contend for the channel. Since we can’t vary the packet length or the time slot for individual STAs in order to increase the offered load of a specific priority class, we add more STAs of that AC. For the first scenario, there are ten (10) wireless STAs that comprise of one backoff entity per AC V O (STA1), two backoff entities for AC V I (STA2 and
STA3), three per AC BK (STA4, STA5 and STA6) and four per AC BE (STA7, STA8, STA9 and STA10). For the second scenario, there are again ten (10) STAs but a reverse approach is followed, instead of increasing the lower priorities, now we increase the higher priorities. So, the second scenario includes one backoff entity per AC BE (STA1), two backoff entities for AC BK (STA2 and STA3), three per AC V I (STA4, STA5 and STA6) and four per AC V O (STA7, STA8, STA9 and STA10). The third scenario has a totally different set-up. We employ thirty (30) STAs of BE traffic that constantly transmit throughout the simulation run and alongside ten backoff entities for AC V O (STA1 to STA0) are gradually imported. Finally, for all three (3) scenarios, we perform ten (10) separate sim- ulation steps where we add to the system progressively STA1 to STA10, one by one, and collect individual performance measurements.
Figure 4.3 shows the collected results for saturation throughput and expected mean access delay for each AC as non-QoS STAs increase. It can be observed that the total throughput (Figure 4.3a) for all participated ACs tends to a constant value around ∼ 0.6 of the normalised data rate. This again verifies the tendency of the poor channel efficiency, already high lighted in previous sections. As more STAs are included in the results, the saturation throughputs of each AC is affected analogously. Even though the majority of the wireless nodes belong to the lower priority classes, the saturation throughputs for the higher priorities yield graphs higher than the lower ACs, despite being less. For example, although the single STA for the AC V O backoff entity is effected from the increasing offered loads, it uses the channel more often. This behaviour is also derived from Figure 4.3b where the real-time applications satisfy their delay requirements, in contrast to heavy profiled traffic flows with no QoS demands which expect higher delay. More specific, the mean access delay for STA1 that is comprised with AC V O traffic flows remains around ∼ 0.5 ms throughout the simulation runs, while the subsequent AC, the AC V I, has a small incline every time a new STA is introduced to the system. Furthermore, the commencing mean delay for the AC BE is six (6) times more than the highest priority AC.
On the other hand, in Figure 4.4 we change the ratio of non-QoS and QoS STAs by introducing more of the latter. Again, for the saturation throughput results (Figure 4.4a), we observe an influential attitude towards the higher pri-
(a) Saturation Throughput vs. Stations (b) Access Delay vs. Stations Figure 4.3: EDCA performance measurements as non-QoS STAs increase
(a) Saturation Throughput vs. Stations (b) Access Delay vs. Stations
Figure 4.4: EDCA performance measurements as QoS STAs increasing
ority traffic since they ‘steal’ bandwidth from the lower priorities. Even though there are instances where the non-QoS flows show peaks of transmission, these are immediately dropped once higher ACs are introduced over the subsequent simulation steps. At the beginning of the simulation, the single STA of AC BE attains a normalized throughput of 0.4833. As newly arrived traffic (or STAs) enter the network the throughput for the BE application rapidly decreases with an exponentially rate and at the last simulation run achieved only 0.0283. The prioritization mechanism even effects the channel access delay as for the AC BE we observe a rapid increase while more QoS delay bounded flows enter the sys- tem. Even though the number of AC V O backoff entities prevail in this scenario,
they hold the least channel access delay (Figure 4.4b).
(a) Saturation Throughput vs. Stations (b) Access Delay vs. Stations
Figure 4.5: QoS STAs get more greedy
In order to demonstrate the magnitude of this unfairness over the lower ACs, for this third scenario we have introduced heavy-load traffic conditions of AC BE traffic. So, thirty (30) STAs constantly contend the channel and after are granted access permission, they transmit a payload size of 1, 500 bytes. In addition, at every simulation step, a QoS provisioned STA of AC V O traffic will be introduced in the system. Figure 4.5 shows both saturation throughput and channel access delay of this scenario. From the saturation throughput (Figure 4.5a), it can be observed that as the first AC V I backoff entity enters the system, the best effort flows maintain a higher portion of the channel’s bandwidth. Yet again, the delay results (Figure 4.5a) indicate that there is a huge burden over the low backoff entities as from the start the BE traffic flows count large values of delay which rapidly increases as the simulation runs progress. Also, for the third QoS provisioned STA that enters the network we observe a cross-over point for the ACs normalized throughputs, where the most prioritized traffic begins to absorb more bandwidth. Note that at that point, the number of STAs for the low priority AC are thirty (30) while the high priority AC STAs are only three (3). On the other hand the channel access delay for the delay bounded traffic remains in low values throughout the simulation. The EDCA through the prioritization process starves AC BE STAs in order to serve AC V O STAs that have delivery time boundaries.