sfi

NDP

ENTERPRISE IRELAND

NCNRC

Communications Network Research Institute

Framework for Radio Resource Management (RRM) on 802.11 WLANs

Mark Davis

MAC Bandwidth Components

The concept of MAC Bandwidth Components described in the publication A Novel Framework for Radio Resource Management in IEEE 802.11 Wireless LANs provides a framework for developing radio resource management schemes for IEEE 802.11 WLANs.

Essentially, this framework describes the way in which the L2/MAC mechanism operates in controlling how the bandwidth of the wireless medium is utilized. In particular, the bandwidth utilization is described in terms of three coupled components:

The Access component is probably the most important of these as it relates directly to the operation of the CSMA/CA mechanism that lies at the heart of the 802.11 MAC mechanism. Specifically, the Access component includes the inter frame spacing (IFS) and the time required to get the back-off counter (BC) to zero. The Access component is to a large extent determined by the overall level of contention for access to the medium. For example, the value of the Access component increases as the number of competing stations increases owing to the increased number of deferrals required.

One of the key parameters to emerge from this framework is the access efficiency ηa which is defined as the ratio of the Load to the Access components:

ηa = Load / Access

The access efficiency is essentially a measure of how efficiently a station is making use of the bandwidth of the medium. Since the access efficiency is directly proportional to the Load component it will have a dependence on the size of the transmitted frame where the larger the transmitted frames, the larger the access efficiency. The access efficiency is inversely proportional to the Access component and hence will also be dependent on the level of competition or contention for access from all the stations on the medium. As the level of contention increases, the access efficiency decreases which reflects the fact that a station has to wait longer in order to gain an access opportunity to transmit its frames.

In addition to describing the utilization of bandwidth on a WLAN, the MAC Bandwidth Components framework can be used to estimate the availability of bandwidth or capacity. However, the term capacity needs to be carefully defined here as it is not a fixed quantity and instead relates to a station’s ability to seize bandwidth from other stations.

Under the MAC Bandwidth Components framework, the bandwidth availability or capacity is defined in terms of three key values:

The two capacity values Cmax and Cmin recognize the fact that the network load is continually changing and serve to set bounds on the maximum and minimum value of the available capacity. The dominant factor in determining the capacity of a station/AC is its access efficiency and the Cmax and Cmin values predict the upper and lower bounds on the achievable capacity for a given access efficiency value.

Essentially Cmax and Cmin recognizes the fact that although stations and ACs compete with one another for access to the medium (and hence can seize bandwidth from one another), there is an upper limit on how much bandwidth they can seize from another station/AC. Furthermore, Cmin can be thought of in terms of a “ protected” bandwidth or a minimum bandwidth that will always be available irrespective of the network load.

Bandwidth Provisioning

The access efficiency ηa is the key parameter that emerges from the MAC Bandwidth Components framework in determining the ability of a station to seize bandwidth. Since the medium (i.e. the bandwidth) is shared on a WLAN, this means that a station with a large access efficiency factor is actually gaining bandwidth at the expense of another station with a smaller access efficiency factor. Therefore the basis of bandwidth “provisioning” is to control the values of the access efficiency factors in order to put an upper limit on the ability of a station to seize bandwidth from other stations. The access efficiency factor can be controlled via the value of the Access bandwidth component.

Under the original MAC described in the IEEE 802.11 standard, it is not possible to control the Access component as all stations use the same MAC parameters, i.e. DIFS and Contention Window (CW) size. However, under the newer IEEE 802.11e/WMM MAC Enhancement standard, it is possible to have up to four different MAC (or more correctly EDCA) settings that can be used. These four settings correspond to the four traffic classes or access categories (ACs) supported under the standard, namely:

These ACs are implemented as four separate queues each with their own CSMA/CA MAC mechanism and associated EDCA parameters. Essentially, there are three main EDCA parameters of interest here:

802.11e QAP

By allowing the ACs to use different EDCA parameter settings it is possible to realize a differentiated service in terms of prioritized access to the medium, i.e. it is possible to arrange to have the higher priority ACs wait for a shorter period of time than lower priority ACs in accessing the medium.

This allows the Access component and hence the access efficiency to be controlled and this is the basis of the bandwidth provisioning scheme described here.

Default values for these parameters have been defined in the standard for each of the four access categories (ACs) supported. However, the effectiveness or otherwise of these default settings is to a large extent determined by the level and nature of the contention for access. Therefore, the main challenge in any IEEE 802.11e/WMM based bandwidth provisioningscheme is how to adaptively tune these parameters in response to changing network conditions.

However, in the context of bandwidth “provisioning” described here, it more accurately relates to “protecting” a station’s bandwidth by limiting the ability of the other stations to seize bandwidth from it. In terms of IEEE 802.11e/WMM WLANs this means tuning the values of EDCA parameters (i.e. AIFSN, ECWmin, and ECWmax) for each of the access categories (ACs) in such a way as to ensure that each AC enjoys a minimum bandwidth that is “protected” from the other ACs.


The CNRI has developed a measurement-based radio resource control application for IEEE 802.11e/WMM WLANs that can adaptively tune these EDCA settings in such a way as to ensure that sufficient bandwidth can be allocated to each of the ACs in order to satisfy a user's QoS requirements.

Controlling WLAN Capacity

It is possible to tune the Cmin value through adjustments to the EDCA parameters. Consequently, designing a RRM scheme for IEEE 802.11e/WMM networks can now be viewed in terms of tuning the EDCA parameter values in such a manner to ensure that each station/AC has a sufficiently large Cmin value to support its anticipated load.

This can be best illustrated with the following example: Consider a situation where there are two stations operating under the original 802.11b standard where the EDCA parameters are fixed and are the same for both stations. Furthermore, suppose that station STA1 is carrying a low priority service, while station STA2 is carrying a high priority service.

Using 8-2.11b Standard - Fixed MAC
parameters

In this case, the low priority station STA1 has a current load that is less than its Cmin value which essentially means that its service is protected from other stations on the network. On the other hand, the high priority station STA2 has a load that is greater than Cmin and lower than Cavail which means that there is sufficient bandwidth to make the service available, but not enough to protect it. Essentially, the yellow region between Cmin and Cavail represents bandwidth that has been “borrowed” from other stations and happens to be currently available. If these other stations were to increase their load, this “borrowed” bandwidth may be taken back.

Clearly this is an unsatisfactory arrangement as one would prefer for the situation to be reversed, whereby the high priority station STA2 would have a larger value for Cmin (i.e. Cmin > Load) to protect its service, while the low priority service on STA1 is forced to make use of “borrowed” bandwidth (i.e. Cmin < Load < Cavail).

By utilizing the IEEE 802.11e/WMM EDCA mechanism, it is possible to tune the {Cmin, Cavail, Cmax} values via the EDCA parameter settings, i.e. the AIFSN, ECWmin, and ECWmax values, as illustrated below.

Using 802.11e QoS Standard - Adjustable MAC parameters

WLAN Radio Resource Management – Simple Example

This framework for radio resource management on IEEE 802.11 WLANs is best illustrated with the following simple example.

Consider a simple two access category (AC) case where we wish to allocate a minimum capacity (Cmin) to each of the ACs as follows:

The minimum access efficiency factors (AEFs) required to achieve these values of minimum capacity are the following:

The figure below shows the locus of the operating points (AEF1, AEF2) that results in a capacity (C) for AC1 equal to 0.1, i.e. where C(AC1) = Cmin(AC1).

Capacity Contours

Since these operating points represent the minimum values of AEF1 required to achieve the minimum capacity for AC1, all points to the right of this curve will result in a capacity C(AC1) > Cmin(AC1), i.e. the hatched region in the figure below represents valid operating points that satisfy the minimum capacity requirement for AC1.

Capacity Contours

Similarly for the access category AC2, the locus of operating points that results in a capacity C(AC2 ) = Cmin(AC2) is shown in the figure below.

Capacity Contours

As these operating points represent the minimum values of AEF2 required to achieve the minimum capacity for AC2, all points above this curve will result in a capacity C(AC2) > Cmin (AC2), i.e. the hatched region in this figure represents valid operating points that satisfy the minimum capacity requirement for AC2.

In order to satisfy the minimum capacity requirements for both access categories, the valid operating points (i.e. the values of AEF1 and AEF2) must lie in the region bounded between these two curves, as illustrated by the hatched region in the figure below.

Capacity Contours

However, not all of this region can be used since there will usually be an upper limit imposed on the value of the access efficiency factor, usually determined by such practical factors as the size of the transmitted frame.

For example, suppose that there is a maximum value imposed on the values of access efficiency factors (possibly as a result of the nature of the load from their respective nodes) as follows:

Consequently, the system is now limited to the region bounded by the minimum capacity curves and the upper bounds on the access efficiency factors. This region is illustrated by the hatched region in the figure below.

Capacity Contours

Essentially, the goal of the RRC algorithm is to place the operating point (i.e. the access efficiency factors) at the center of the valid region while respecting the minimum capacity requirements and limits on the values of the access efficiency factors.

Suppose that we now wish to accommodate a number of streams of access category AC1 and the obvious question that arises is: How many streams of access category AC1 can be supported?

The figure below shows the resulting capacity contours as the number of AC1 streams is increased from 1 to 6 streams. The minimum capacity requirement for AC1 remains at 0.1 and there is still only a single AC2 stream present.

Capacity Contours

The first feature to notice on this figure is how the minimum AEF values increase as the number of AC1 streams increases. This is to be expected as the increased capacity requirement results in larger values of AEF being required. Another important feature is how the area of region between the capacity curves decreases as the number of AC1 streams increases which means that there is a greater restriction imposed on the values of AEF1 and AEF2 that can be used.

Furthermore, if there is an upper limit imposed on the values AEF1 and AEF2, then this will determine the maximum number of streams that can be accommodated. In this case where the maximum value of AEF1 is 0.3, then the maximum number of AC1 streams that can be accommodated is 5. Moreover, the values of AEF1 and AEF2 are restricted to a tight range of values, i.e. (AEF1, AEF2 ) ~ (0.3, 0.5). Unfortunately, more than five AC1 streams cannot be accommodated as the minimum access efficiency factor that is required exceeds that which is achievable.

This simple example serves to highlight the need to control the access efficiency factors of the various access categories in order to correctly manage the bandwidth resource and ensure that each access category experiences its required quality of service.

WLAN Radio Resource Management – Practical Example (i.e. VoWLAN Call Capacity)

In this next example, we consider a more practical example where we wish to investigate how many voice over IP (VoIP) calls can be accommodated on a WLAN, i.e. we wish to study VoWLAN call capacity. The call capacity is very much a function of the other traffic on the network and in this example we will consider the capacity under a number of different traffic types.

We begin by investigating the call capacity in the presence of video streaming traffic and here we assume that all the VoIP streams are transmitted using the AC_VO access category, while the video streams are transmitted using the AC_VI access category.

Furthermore, in order to ensure that an acceptable quality of service (QoS) is experienced by each stream we specify a minimum capacity requirement for the streams. From our experimental work we have determined that the minimum capacity requirements for VoIP and video streaming services are as follows:

Furthermore, owing to the nature of traffic streams generated by these services, there is an upper limit imposed on the access efficiency factors that can be achieved as follows:

In the figure below, the minimum access efficiency factor for the AC_VO streams is plotted as the number of AC_VO streams is increased. When there is no video stream present, a maximum of 24 VoIP streams can be accommodated before the value of the minimum access efficiency factor reaches its maximum achievable value.

Minimum Access Efficiency Factors for
AC_VO streams

When there is a single video stream present, the capacity drops to 16 VoIP streams and when there are two video streams present, the capacity drops back further to 7 VoIP streams. When there are three or more video streams present, then no VoIP calls can be accommodated as the minimum access efficiency factor required exceeds that which can be achieved.

Having determined the minimum values of the access efficiency factor that need to be used, the next step is to consider what are the actual values of the access efficiency factors that should be used. Here we use the algorithm for calculating the target access efficiency factors described in the section on the design of the WLAN resource controller (WRC). Essentially, the goal of this algorithm is to place the operating point (i.e. the access efficiency factors) at the center of the valid region while respecting the minimum capacity requirements and limits on the values of the access efficiency factors. In other words, we are seeking to “tune” the access efficiency factors for each of the access categories in order to satisfy their minimum capacity requirements.

The figure below shows how the access efficiency factors need to be tuned as the number of AC_VO streams is increased (and where there is a single AC_VI stream present). This figure also shows the performance of the AC_VI stream.

Tuning Access Efficiency Factors for
AC_VO and AC_VI

The AEF(AC_VO) value needs to be tuned in the range 0.08-0.1 close to its maximum value, while the AEF(AC_VI) value needs to tuned in the range 0.45-0.49. The larger values in these ranges are required as the amount of the allocated capacity increases. Initially the AC_VI stream is allocated a capacity well in excess of its minimum requirement of 0.25. However, this excess capacity allocation is gradually given up to the AC_VO streams as their requirement increases. When the AC_VO traffic saturates at 16 VoIP streams, all access categories are operating at their minimum capacity requirement, i.e. C (AC_VO) = 0.1 and C(AC_VI) = 0.25. The value of AEF(AC_VI) needs to be limited to 0.49 to prevent it from taking more capacity away from the AC_VO stream while at the same ensuring that its minimum capacity requirement is satisfied.

The figure below shows the tuning required for the access efficiency factors in the case where there are two video streams present.

Tuning Access Efficiency Factors for
AC_VO and AC_VI

The AEF(AC_VO) value needs to be tuned in the range 0.09-0.1 close to its maximum value, while the AEF(AC_VI) value needs to tuned in the range 0.43-0.48. The larger values in these ranges are required as the amount of the allocated capacity increases. The initial capacity allocation to the AC_VO streams is greater than that required. However, the allocation is reduced to its minimum requirement as the number of AC_VO streams increases to its saturation point at 7 streams.

Finally, we consider the case where all four access categories allowed under the IEEE 802.11e/WMM standard are utilized. We assume the following characteristics for these access categories:

The results from the algorithm for calculating the target access efficiency factor values are plotted in the figure below. In this case the number of AC_VO streams that can be accommodated is 8.

Tuning Access Efficiency Factors for
AC_VO and AC_VI

The tuning ranges for the access efficiency factors are as follows:

Initially all access categories are allocated a capacity greater than that required. However, as the number of AC_VO streams increase, each access category has its allocation reduced to its minimum requirement in order to meet the growing demand from the AC_VO streams.