3.1 Introduction
The draft of IEEE 802.11e introduced a new medium access mechanism to guarantee a high level of QoS for the real time services, the working group E has proposed the Hybrid Coordination Function (HCF) as an enhancement to the legacy medium access functions: the Distributed Coordination Function (DCF) and the Point Coordination function. The DCF can’t guarantee QoS, while PCF can guarantee a low level of QoS but it still has some problems as described earlier in chapter 1. The HCF is a coordination function that combines aspects of the distributed coordination function and point coordination function, it consists of two medium access functions that work together: Enhanced Distributed Channel Access (EDCA) and HCF-Controlled Channel Access (HCCA). EDCA is an enhanced version of DCF, it is a contention based mechanism, while HCCA is an enhanced version of PCF that is contention free based and uses the polling mechanism to grant QoS stations the access to the medium for a limited amount of time calculated at the Hybrid Coordinator (HC), this time is called transmission opportunity (TXOP). IEEE 802.11e draft specifies neither a scheduling mechanism nor a polling mechanism for HCCA, and only an example that satisfies the minimum real time traffic requirements has been proposed, the scheduler has been left for designers. Although many scheduling mechanisms have been proposed, a lot of work still needs to be done to achieve an efficient scheduler.
In this chapter we will focus on some scheduling problems in general and on the Round-Robin (RR) polling mechanism in particular, we will discuss its weaknesses, and show its performance through simulations.
RR polling mechanism have some problems that we are going to discuss in this chapter, those problems are:
-
Fairness polling problem.-
Polling overhead problem.-
High delays with the increase of network population.3.2 Fairness polling
In RR polling mechanism, HC polls stations sequentially in the order they are placed in the polling list. When the current polling round is over, the AP memorizes the place and starts polling it in the next round.
In a voice conversation users normally tend to stop their conversation, listen to their counter part and restart their conversation. The effect is known as the talkspurt-silence alternation, and the behavior is independent of the codec used and can be modeled by a two-state Markov chain as shown in Figure 3.1. The model is actually an ON/OFF stochastic process. The ON (talkspurt) and OFF (silence) periods are exponentially distributed with mean values 1/ α and 1/ β, respectively. During talk periods a voice flow is represented as an isochronous source with fixed inter-arrival time T that are determined by the audio codec. In other words, the packet generation rate λ is fixed. During silence period, nevertheless, no packets are generated.
Figure 3.1: A two-state Markov chain voice activity model.
The talkspurt-silence alternation characteristic of voice traffic makes RR polling mechanism not efficient at all. The reason is that the HC will neither add/delete station IDs in the polling list, nor change the sequence according to stations which are polled. As a result, although a certain voice source enters silence period, the still HC continues to poll it even the station responds NULL frame each time. This will cause wastage of
Talkspurt Silence
β
λ α 0
valuable bandwidth wastage and incurs unnecessary delay to other stations in talkspurt state [17].
We see that when the HC enforces fair access to all stations regardless they were talking or silent, it wastes the resources that should be reserved for the talking stations with higher priority than the silent stations.
3.3 Polling overhead
In the IEEE 802.11e draft, overhead in HCF is due to frequent poll frames from the AP to mobile stations. The AP sends a QoS CF-Poll frame to each station (according to its polling list) one by one to grant the TXOP. However, this polling method is inefficient due to the large overhead. According to the IEEE 802.11 standard, the CF-Poll frame is required to be transmitted at the basic rate (2 Mb/s of IEEE 803.11b) regardless of the data rate. The size of CF-Poll frame is 36 bytes in the 802.11e draft, including 10 bytes for frame/sequence/QoS control and frame check sequence (FCS), 24 bytes for station ID, and 2 bytes for duration ID. Its transmission time at 2 Mb/s is 36 * 8/2 = 144 µs.
Furthermore, considering the PHY overhead (192 µs), the total transmission time is 336 µs. The transmission time for a 196-byte voice packet (160-byte payload with a G711 codec and 36-byte MAC header) at 11 Mb/s data rate is 196*8/11 = 142.55 µs. Compared to the voice packet transmission time, the overhead contributed by the CF-Poll frame is relatively large. For a WLAN accommodating N voice users, the total overhead contributed by CF-Poll is 336 * N µs, which is significant [2]. Consider 40 voice users, half of them are talking and the other half are listening, to give any of them the permission to access the medium AP must send a QoS CF-Poll first, for the silent users who has no voice packets to send this poll is wasted and they are polled for nothing, and this poll is considered as overhead, 20*336 = 6720 µs. So we need a mechanism to avoid polling those stations with no buffered packets on their voice TC.
3.4 Delay
Polling silence stations wastes the time of receiving the null packet in addition to the time required to send the CF-Poll packet to the silent station, and by polling those silent stations frequently, longer and longer time is wasted, as a consequence of this
wasted time, talking stations will suffer long access delays, because they are enforced to wait until the HC poll all the stations before them in the polling list, which may be silent.
Long delays may cause a high packet drop ratio on the station side. Talking station that is generating voice packets continuously may start dropping voice packets because of the long waiting time it may suffer.
Delay Average VS Network sieze
0 5 10 15 20 25 30 35 40
1 2 3 4 5 6 7 8 9
Number of stations
Delay Average (ms)
Round-robin
Figure 3.2: Delay average against network size for RR using HCF
We can notice in Figure 3.2, which we got from our simulation, the increasing delay due to the increase in the network size, which can be explained by the wasted polls to the silent stations and due to the high polling overhead. While Figure 3.3, which is a result of our simulation too, shows the packet loss ratio of RR for different network sizes, the increasing in packet loss ratio is noticeable; this increase is a result of the high access delays when the number of stations in the network is increased.
Packet Loss Ratio VS Network size
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18
1 2 3 4 5 6 7 8 9
Number of stations
Packet Loss Ratio
Round-robin
Figure 3.3: Packet loss ratio against network size using HCF