WO2006130021A2 - Method and device for administration of traffic in communication system using a contention based access control - Google Patents

Abstract

A method and device for determining a starvation condition for an access category at an access point in an IEEE 802.11 based wireless network employing multiple wireless access categories. A first access point traffic load value is determined actual access point traffic, and compared with a predetermined second access point traffic load value. A starvation condition indicator is generated if said comparing shows that said first access point traffic load value is about or exceeds said predetermined second access point traffic load value. The first access point traffic load value is determined by obtaining for said actual access point traffic a number of successful transmissions and collisions, and dividing said number of successful transmissions and collisions by said number of successful transmissions and collisions and a number of empty slots. The predetermined second access point traffic load value is determined as l/(l+(AIFSN-2)), wherein AIFSN is a number of timeslots of an AIFS parameter of said access category. An access parameter of said access category is amended in response to said starvation condition indicator.

Description

Method and device for administration of traffic in communication system using a contention based access control

I. INTRODUCTION

During recent years the IEEE 802.11 WLAN standard [1] has been widely deployed as the most preferred wireless access technology in office environments, in public hot-spots and in the homes. For residential usage, the WLAN Access Point (AP) is not only used for Internet access, but vendors also envision using the AP as a wireless backbone that interconnects household devices, such as the TV set, the DVD player and the multimedia storage on the home computer, the phones and so forth. Due to the inherent capacity limitations of wireless technologies, the 802.11 WLAN easily becomes a bottleneck for communication. In these cases, the QoS features of the 802. lie standard will be beneficial to prioritize for example voice and video traffic over more elastic data traffic.

The IEEE 802.11 medium access control (MAC) comprises the mandatory Distributed Coordination Function (DCF) as a contention-based access scheme, and the optional Point Coordination Function (PCF) as a centrally controlled polling scheme. However, PCF is hardly implemented in any products, and DCF represents the commonly used MAC mechanism of 802.11. DCF adopts carrier sense multiple access ("listenbefore- talk") with collision avoidance (CSMA/CA) and uses binary exponential backoff. A station not only goes into backoff upon collision. It also carries out a "post-backoff after having transmitted a packet, to allow other stations to access the channel before it transmits the next packet. The IEEE 802.1 Ie standard [2] works as an extension to the 802.11 standard, and the Hybrid Coordination Function (HCF) is used for medium access control. HCF comprises the contention-based Enhanced Distributed Coordination Function (EDCF) as an extension for DCF, and the centrally controlled Hybrid Coordinated Channel Access (HCCA) as a replacement for PCF. EDCF has received most attention recently, and it seems that this is the WLAN QoS mechanism that will be promoted by the majority of vendors. EDCF is therefore the area of interest of this paper, and HCCA will not be discussed any further here. EDCF enhances DCF by allowing four different access categories (ACs) at each station and a transmission queue associated with each AC. Each AC at a station has a conceptual module responsible for channel access for each AC and in this paper the module is referred to as a "backoff instance". Hence each of the four transmission queues (and the associated ACs) on a station is represented by one backoff instance. The channel access between different backoff instances on a station is not completely independent due to the virtual collision handling between the queues on the station. If two or more backoff instances on the same station try to access the channel in the same timeslot, the station attempts to transmit the frame of the highest priority AC, while the lower priority frames will go through backoff. The traffic class differentiation of EDCF is based on assigning different access parameters to different ACs. First and foremost, a high-priority AC, i , is assigned a minimum contention window, W₀ ,- , that is lower than (or at worst equal to) that of a lower-priority AC. At a lightly loaded(or "unsaturated") medium, the post-backoff of the high-priority AC will normally be smaller than the post-backoff of a low-priority AC, resulting in an average higher share of the channel capacity. Moreover, as the channel gets more congested (or "saturated"), the high- priority AC will on average have to refrain from the channel for a shorter period of time than what the low priority AC has to.

Another important parameter setting is the AIFS value, measured as a Short Interframe Space (SIFS) plus an AIFSN number of timeslots. A high-priority AC is assigned an AIFSN that is lower than (or at worst equal to) the AIFSN of a lower-priority AC. The most important effect of the AIFSN setting is that the high-priority AC normally will be able to start earlier than a low priority AC to decrement the backoff counter after having been interrupted by a transmission on the channel. At a highly loaded channel where the decrementing of the backoff counter will be interrupted by packet transmissions a large number of times, the backoff countdown of the high-priority AC will occur at a higher average speed than that of the lower-priority AC. As the wireless medium gets more and more congested, the average number of empty timeslots between the frames transmitted by the higher-priority ACs might be lower than the AIFSN value of the low-priority AC. At this point, the AC will not be able to decrement its backoff counter, and all packets will finally be dropped instead of being transmitted. This is referred to as "starvation".

Other differentiation parameters that may be adjusted in 802.1 Ie (and which are also explicitly or implicitly included in the model proposed below) are the retry limit, L₁ (of short and long packets), the maximum contention window W₁ , L₁ and the TXOP-limit of each AC, i .

Most of the recent analytical work on the performance 802.1 Ie EDCF stems from the simple and fairly accurate model proposed by Bianchi [3] to calculate saturation throughput of 802.11 DCF. Later, Ziouva and Antonakopoulos [4] improved the model to find saturation delays, however, still of the undifferentiated DCF. They also improved the model by stopping the backoff counter during busy slots, which is more consistent with the IEEE 802.11 standard.

Based on this work, Xiao [5] extended the model to the prioritized schemes provided by 802. lie by introducing multiple ACs with distinct parameter settings, such as the minimum and maximum contention window. It is also straightforward to extend the model for the use of different TXOP sizes of different classes. Furthermore, this prioritized model also introduced a finite retry limit. This additional differentiation parameter leads to more accurate results than previous models. (A list of references for other relevant efforts and model improvements of DCF can also be found in [5].)

Priority based on differentiated Transmission Opportunities is not treated explicitly in this paper. For simplicity and to keep focus on the most important issues we have assumed that all traffic classes sends packets of equal lengths (i.e. of 1024 Bytes), and that each packet fits perfectly into one TXOP. Calculating the model with respect to different packet lengths is easy, as shown by Xiao [??]. It is also easy to extend our analyzes to contention-free bursting (CFB) within a TXOP by summing up the different SIFS occurring between subsequent packets of the burst.

THE PROBLEM AREA.

A differentiation parameter lacking in Xiao's model, however, is the important AIFS parameter. Xiao assumed equal AIFSN of all traffic classes, and the model does not correctly capture starvation. (In fact, a situation with only two different ACs was analyzed in the work.) In many cases the QoS-enabled Access Point (QAP) would need to predict when the starvation of an AC will occur, so that it will be able to know when to change the current parameter settings, e.g. to avoid that any AC is completely starved. In other words, predicting starvation can be paramount for admission control through dynamic parameter settings, because a QAP that is not currently sending traffic on a low priority AC may not know if an AC is facing starvation or if the other stations simply are not sending traffic of that AC. The model is also limited as a fully saturated channel is assumed. Due to the bursty characteristics of many types of data traffic, it is unlikely that the channel will be fully saturated all the time. The rate adaptation of TCP, for example, will often ensure that the total channel load will not be fully saturated. Hence, in many cases an access point will be more interested in knowing how to set parameters for a lightly saturated channel, and to adjust these parameters dynamically in this region. An analytical model that covers the full range from a non-saturated to a fully saturated channel, would be more useful.

The use of AIFSN to differentiate between ACs has two consequences. Assume two backoff instances with different AIFSNs. Both have a packet to send, but the channel is busy, so they have to wait. The first effect occurs when both backoff instances are not in backoff, i.e. they are not in binary backoff and the post-backoff is completed (or the post-backoff is not necessary because it is the first packet ever to be sent). In this case, when the channel is sensed idle, the backoff instance with the lowest AIFSN is the first to be allowed to make a transmission attempt. The other backoff instance will sense that the channel is busy during this time slot. It will have to wait until the transmission attempt of the first backoff instance is completed before it is allowed to transmit the packet.

The second effect occurs when both backoff instances are in backoff. To describe the effect, assume that the backoff counter of both backoff instances are equal. Each time a packet transmission is completed, the backoff continues countdown of idle time slots as soon as the AIFS interval is completed. However, after the packet transmission, the backoff instance with the lowest AIFSN will start counting down idle time slots before the other backoff instance is allowed to enter the count down procedure. Thus, the backoff instance with the lowest AIFSN will be able to count down the backoff window faster than a backoff instance with higher AIFSN.

THE INVENTION.

The accompanying drawings presenting figures 1 - 15 are considered part of the description of the invention. In the drawings, fig. 1 is a schematic representation of Markov chain description a traffic scenario relating to the invention, fig. 2 illustrates schematically AIFS differentiation, fig. 3 illustrates schematically a setup used to validate the analytical model, figures 4 - 6 are plots, fig. 7 illustrates a conceptual model of stations and a channel, fig. 8 illustrates schematically an example of an access point implementing the present invention, fig. 9 illustrates schematically a further example of an embodiment of the present invention, fig. 10 illustrates schematically a further example of an embodiment of the present invention, fig. 11 illustrates schematically a measurement on the channel for an embodiment of the present invention, fig. 12 is a flow chart illustrating a exemplary method embodiment of the invention, fig. 13 is a flow chart illustrating a exemplary detail of the method embodiment of the invention of fig. 12, fig. 14 is a flow chart illustrating a exemplary detail of the method embodiment of the invention of fig. 12, and fig. 15 is a block schematic overall illustration of an exemplary QoS enabled access point that includes an embodiment of the invention.

This paper extends Xiao's model to enhance it in a number of ways: • The presented model predicts the performance not only in the saturated case, but on the whole range from a unsaturated medium to a fully saturated channel. (Some works, such as [6] and [7] have explored unsaturated conditions, however, only of the one-class 802.11. They are also primarily focussing on the non-saturation part instead of finding a good descriptive solution for the whole range.)

• In the non-saturation situation, our model accounts for "post-backoff of an AC, although the queue is empty, according to the IEEE 802.11 standard. If the packet arrives in the queue after the "post-backoff is completed, the listen-before-talk (or CSMA) feature of 802.11 is also incorporated in the model.

• Our model describes the use of AIFSN as a differentiating parameter, in addition to the other differentiation parameters encompassed by Xiao's efforts and other works.

• A simple closed-form equation that predicts with satisfactory accuracy the starvation point (or "freeze point") of each traffic class is provided. The only prerequisite for a station (e.g. an access point) to determine that starvation of an AC has occurred, is to know the AIFSN value of the AC and to monitor the traffic load on the channel.

The remaining part of the disclosure is organized as follows: The next section summarizes the differentiation parameters of 802.1 Ie and provides the basis for understanding the analytical model provided in the subsequent two sections. After having presented the model, a section is allocated to the validation of the model against simulations. Our findings are finally summarized in our concluding remarks.

The invention provides a method and a device for device for determining a starvation condition for an access category at an access point node in a wireless network employing multiple wireless access categories, the features of which method and device are recited in the accompanying patent claims 1 and 5, respectively.

Further advantageous features of the method and device of the invention are recited in the accompanying dependent patent claims 2 - 4 and 6 - 8, respectively

Furthermore, the present invention intends to provide a method and a device for administration of traffic in communication system using a contention based access control. A first value for a prediction, estimate or measurement of the actual traffic load at an access point is determined. A second value for a predetermined traffic load is determined from system parameters or settings. From a comparison of the first and second values, a traffic starvation condition is a'predicted or detected, hi response to the predicted or detected traffic starvation condition, operational parameters of the system relating to starvation s modified. In one embodiment, modified parameters are forwarded to units of the system for them to take effect.

In a first aspect, the present invention includes a method for controlling a starvation condition for an access category at an access point in an IEEE 802.11 based wireless network employing multiple wireless access categories, the method including: determining from actual access point traffic a first access point traffic load value, comparing said first access point traffic load value with a predetermined second access point traffic load value, and generating a starvation condition indicator if said comparing shows that said first access point traffic load value is about or exceeds said predetermined second access point traffic load value.

The method according to the first aspect of the present invention may further include determining said first access point traffic load value by obtaining for said actual access point traffic a number of successful transmissions and collisions, and dividing said number of successful transmissions and collisions by said number of successful transmissions and collisions and a number of empty slots.

The method according to the first aspect of the present invention may further include a variant wherein the number of empty slots is a number of empty back-off slots in a channel associated with said access category of said access point.

The method according to the first aspect of the present invention may further include a variant wherein determining said predetermined second access point traffic load value as l/(l+(AIFSN-2)), and wherein AIFSN is a number of timeslots of an AIFS parameter of said access category.

The method according to the first aspect of the present invention may further include a variant with amending an access parameter of said access category on basis of said starvation condition indicator.

In a second aspect, the present invention includes device for controlling a starvation condition for an access category at an access point in an IEEE 802.11 based wireless network employing multiple wireless access categories, the device including: a first determining means adapted to determine, from actual access point traffic, a first access point traffic load value, a first comparing means operatively associated with said first determining means, said first comparing means adapted to compare said first access point traffic load value with a predetermined second access point traffic load value, and a first indicator generating means operatively associated with said first comparing means, said first indicator generating means adapted to generate a starvation condition indicator if said comparing shows that said first access point traffic load value is about or exceeds said predetermined second access point traffic load value.

The device according to the second aspect of the present invention may further include a variant wherein said first determining means, for determining said first access point traffic load value, includes (a) a first data acquisition means for acquiring, for said actual access point traffic, a number of successful transmissions a number of collisions, and a number of empty slots, and (b) a data computing means for dividing said number of successful transmissions and collisions by said number of successful transmissions and collisions and said number of empty slots.

The device according to the second aspect of the present invention may further include a variant wherein the number of empty slots is a number of empty back-off slots in a channel associated with said access category of said access point.

The device according to the second aspect of the present invention may further include a second determining means operatively associated with said first comparing means, said second determining means adapted to determine said predetermined second access point traffic load value as l/(l+(AIFSN-2)), wherein AIFSN is a number of timeslots of an AIFS parameter of said access category.

The device according to the second aspect of the present invention may further include an access parameter amending means operatively associated with said first indicator generating means, said access parameter amending means adapted to amend an access parameter of said access category in response to said starvation condition indicator.

Further features and aspects of the invention are disclosed in the following general description and further explanations, some of which are provided by way of embodiment examples of the present invention.

According to yet a further aspect of the invention, typically, based on comparing a traffic load measurement with a computed "starvation threshold", there are to options: a) a station (STA) amends its own channel access parameters; and b) if the station is an access point (AP), it is enabled to send the new parameters on the channel to inform other STA about the new settings (non-distributed control). In this case, a) would be relevant for the AP, while other STA leaves the responsibility for carrying out a) to the AP.

Note that 802.11 can be operated in "Infrastructure Mode" (with AP) as well as "Independent mode" (without AP). In the "Independent Mode", all STA will typically perform (a) described above and adapts their own parameters as a distributed solution, while b) will not be performed.

N is a number which reflects the AIFSN-assignments to the various access categories. Normally are the access categories of highest priority assigned by AIFSN = 2, thus, the minimum value of the assigned AIFSN value over all access categories is 2. Accordingly is the formula (AIFSN[i] - 2) derived as disclosed in the original text. The case is, however, that the access categories of highest priority not necessarily is assigned an AIFSN value of 2, but possibly a higher number, such that the "formula" should be adapted accordingly.

The designator i is used to number an access category. Each access category has its own AIFSN value, and AIFSN[J] designates the AIFSN value which has been assigned to access category number i. The starvation test is performed for each individual access category.

The AIFS parameters "minimum contention window" and "maximum contention window" are set for each category according to the 802.1 Ie standard. In addition comes the "TXOP- limit" (i.e., the period during which one is allowed to retain the channel after first having gained access to it) is another adjustable parameter. The "persistence factor" is a further parameter which principally may be adjusted (although it has been taken out of the standard).

π. IMPORTANT DIFFERENTIATION PARAMETERS OF 802. HE A. Priority based on Contention Windows (CWs) and Exponential Backoff For each AC, i(i = 0.,,,.3) , we let W_{t j} denote the contention window size in the j -th backoff stage i.e. after the j -th unsuccessful transmission; hence W₁₀ = CW_{i nin} + 1 , where the recommended values for CW_{i πin} are listed in table I. We also denote j = m_t as the j -th backoff stage where the contention window has reached CW_{i im]i} + 1 ; the window will no longer be increased in the subsequent backoff stages. Hence,

W₁. = log₂ i[cW_lιπm + 1)/ CW₁ ^ + 1) . Finally, we let L₁ denote the retry limit of the retry counter; if the transmission is unsuccessful after the L₁ -th backoff stage, the packet will be dropped. The 802.11 specification allows for different retry limits, DotllShortRetryLimit and DotllLongRetryLimit, for packets that are longer than and shorter than the Dotl IRTSThreshold, respectively. In this paper, however, we will assume that the sizes of all packets of a class i are either below or above the Dotl lRTSTreshold-parameter, so that Li is equal for all packets belonging to the same class.

9^JW j = 0,l,....,m,. -l

(1) ^l'^J {2""W_iβ = CW_it!mx j = «,,....,£,

In the special case where m_t < L₁ , eq. (1) is reduced to W_Uj = 2^JW_ifi for j = 0,1,...,L₁ .

TABLE J RECOMMENDED {DEFAULT) PARAMETER SETTtNCiS FOR 8.02. 1 j E

B. Priority based on Inter-Frame Spaces (IFSs)

When a backoff instance senses that the channel is idle after a packet transmission, it normally waits a guard time called the Distributed Inter-Frame Space (DIFS) during which it is not allowed to transmit packets or do backoff countdown. The duration of DIFS is the sum of a SIFS and two time slots. The two time slots of DIFS allow the Hybrid Coordinator (HC) on the QAP (or Point Coordinator with only plain 802.11) to access the channel with higher priority. The HC is allowed to enter the channel after only waiting one time slot (in addition to SIFS) and it does not need to go through "post-backoff before accessing the channel. Moreover, certain packets - including the Clear-To-Send (CTS) and Acknowledgement (ACK) packets - can be sent after only waiting SIFS. This gives maximum priority to this traffic, and ensures that a data exchange (such as a data transmission followed by an ACK) can be considered nearly an atomic transaction. Instead of using DIFS, each AC[i] of 802. lie uses an Arbitration Inter-Frame Space (AIFS [i]) that consists of a SIFS and an AIFSN[i] number of additional time slots. In this paper we define A₁ as:

A₁ = AIFSIi]- WJn(AIFS[I]) > 0 J = 0,...,N -l (2) where N is the number of different ACs (i.e. normally four). The 802.1 Ie standard mandates that AIFSN[i] _ 2, where the minimum limit of 2 slots corresponds to the DIFS interval. The use of AIFSN to differentiate between ACs has two consequences. Assume two backoff instances with different AIFSNs. Both have a packet to send, but the channel is busy, so they have to wait. The first effect occurs when both backoff instances are not in backoff, i.e. they are not in binary backoff and the post-backoff is completed (or the post-backoff is not necessary because it is the first packet ever to be sent). In this case, when the channel is sensed idle, the backoff instance with the lowest AIFSN is the first to be allowed to make a transmission attempt. The other backoff instance will sense that the channel is busy during this time slot. It will have to wait until the transmission attempt of the first backoff instance is completed before it is allowed to transmit the packet.

The second effect occurs when both backoff instances are in backoff. To describe the effect, assume that the backoff counter of both backoff instances are equal. Each time a packet transmission is completed, the backoff continues countdown of idle time slots as soon as the AIFS interval is completed. However, after the packet transmission, the backoff instance with the lowest AIFSN will start counting down idle time slots before the other backoff instance is allowed to enter the count down procedure. Thus, the backoff instance with the lowest AIFSN will be able to count down the backoff window faster than a backoff instance with higher AIFSN.

C. Priority based on Transmission Opportunities (TXOPs)

Priority based on differentiated Transmission Opportunities is not treated explicitly in this paper. For simplicity and to keep focus on the most important issues we have assumed that all traffic classes sends packets of equal lengths (i.e. of 1024 Bytes), and that each packet fits perfectly into one TXOP. Calculating the model with respect to different packet lengths is easy, as shown by Xiao [??]. It is also easy to extend our analyzes to contention-free bursting (CFB) within a TXOP by summing up the different SIFS occurring between subsequent packets of the burst.

m. ANALYTICAL MODEL

C. Priority based on Transmission Opportunities (TXOPs)

Figure 1 illustrates the Markov chain for the transmission process of on backoff instance of priority class i .

The state changes in the figure occur only when the backoff instance would be able to contend for the channel. If one or more stations transmit in a time slot, that slot is sensed busy by the backoff instances. For the subsequent time that the packet is under transmission, the backoff instance remains in the same state. The Markov process finally resumes at the first slot where the channel again is open for contention, i.e. after the transmission (or collision) is completed and after the real-time duration of an additional DIFS. Hence the (virtual) time scale is discrete and integral, and a backoff slot that is open for contention triggers each "clock tick". Measured in real-time the duration of a slot that leads to transmission is equal to the length of the transmission including the RTS and CTS (if being used), the data packet and the ACK as well as all the associated inter-frame spaces. The real-time duration of an empty slot is equal

:o the duration of a regular backoff slot.

[n the Markov chain, the utilization factor, p_t , represents the probability that there is a packet waiting in the transmission queue of the backoff instance of AC i at the time a transmission is completed (or a packet dropped). Now, the backoff selects a backoff interval k at random and goes into post-backoff. If the queue is empty, at a probability 1- p_t , the post-backoff is started by entering the state (i,Q,k,e) . If the queue on the other hand is non-empty, the post- backoff is started by entering the state (i,0,k) . Hence, p. balances the fully non-saturated situation with the fully saturated situation, and therefore plays a role to model the behaviour of the intermediate semi-saturated situation. We see that when p_t — > 1 the Markov chain behaviour approaches that of the non-saturation case similar to the one presented by Xiao [5]. On the other hand, when p_t — > 0 the Markov chain models a stochastic process where the backoff instance after transmission always goes into "post-backoff without a packet to send. As mentioned above, the states, (i,0,k,e) , in the upper row is entered when the channel is not fully saturated, and when the queue of a backoff instance is empty at the time a transmission is completed (or a packet dropped). The states (i,0,l,e) , ..., (i,0,W_{i 0} -l,e) represent a situation where the transmission queue is empty, but the station is counting down backoff slots. The probability that a backoff instance of AC i is sensing the channel busy and is thus unable to count down the backoff slot from one timeslot to the other is denoted with the probability p^* . (We use the asterisk to indicate that this is a probability related to the countdown process.) It undertakes a countdown at probability 1- p\ and moves to another state. If it has received a packet while in the previous state at a probability q^* , it moves to a corresponding state in the second row with a packet waiting for transmission. Otherwise, it remains in the first row with no packets waiting for transmission.

While in the state (i,0,0,<?) , on the contrary, the backoff instance has completed post-backoff and is only waiting for a packet to arrive in the queue. If it receives a packet during a timeslot at a probability q. , it does a "listen-before-talk" channel sensing and moves to a new state in the second row, since a packet is now ready to be sent. If the backoff instance senses the channel busy, at a probability p_t , it performs a new backoff. Otherwise, it moves to state

(/,0,0, e) to do a transmission attempt. The transmission succeeds at a probability \ — p_r

Otherwise, it doubles the contention window and goes into another backoff.

All other rows apart from the first row in the figure illustrate a situation with at least one packet in the system. Indeed, only these states are entered in the extreme case of a fully saturated channel and a non-zero traffic load on each AC, so that the transmission queue is always full. Hence, after successful transmission or after a packet has been dropped, the backoff instance proceeds directly into one of the post-backoff states (i,0,k) (for k = 0,1,...W- _j - 1 ). For each unsuccessful transmission attempt, the backoff instance moves to a state in a row below at a probability p_t . However, if the packet has not been successfully transmitted after L₁ + 1 attempts, the packet is dropped. Hence, the accumulated frame dropping probability, P_i<drop , can be estimated as:

Let b_{i j k} denote the state distributions. Since, the probability of a transmission attempt entering stage j (where j = 0χ....,L_t ) is pf , chain regularities yields:

K_j,o = Pι Ko,o ; 7 = 0,1,.., L. (4)

Furthermore, we observe that a backoff instance transmits when it is in any of the states (i, y^',0) where j = 0,1,..., L₁ . Hence, if we let T₁ denote the transmission probability (i.e. the probability that a backoff instance in priority class i transmits during a generic slot time, independent on whether the transmission results in a collision or not), we have:

In the following subsection we will find ways to express b_{tj 0} and p. in terms of T₁. Hence, a complete description of the system can be found by solving the above set of equations (one equation per AC i).

B. Markov chain analysis

First we will look at the post-backoff stage of the Markov chain for 7 = 0. From chain regularities, we observe that:

*M>,O = &U,.O + ∑ (1 - Λ)*UO <⁶)

By working recursively through the chain from right to left in the upper row, we get:

Furthermore, we see that: ^b .ifiAe - q- _wA-⁾w^{~ (i}- *<?;f'° ro W

From the upper left part of the Markov diagram we see that b_ifijWj ^₁ = b_(i0>0/((l - P^*Wi_,o)- ^BY working recursively and horizontally thorough the chain we also observe that:

W. — k Ko,_k = _w g _ . (^,0,0 + ) - hoxe for * = 1,2,...,WJ₁₀ -I (9)

Undertaking the same analysis for the rest of the chain, we get:

Ki* l,2,...,W;_i0 -l (10)

Finally, the normalization requires that:

This yields:

The first sum in the equation above represents the saturation-part, while the second part is the dominant term under non-saturation. Hence, the expression provides a unified model encompassing all channel loads from a lightly loaded non-saturated channel, to a highly congested, saturated medium. This full-scale model will be validated below.

The non-saturation part of the expression might require further explanation. First, the factor (1 - P₁) (where p_i is the utilisation factor of the backoff instance) represents the probability of having an empty queue after successful transmission. If this happens, it enters the empty- queue post-backoff procedure, which is represented by the states (/,0,0, e) , ... , (i,0,W_ifi -l,e) .

Second, the geometric sum 1 - (1 - q^*)^Wuo /(W_{i O}q^*) expresses the probability of not receiving any packets in the transmission queue while performing the complete empty-queue post- backoff. In other words, it is the probability of finally ending in the state (i,0,0,e) , instead of transitioning to any of the regular post-backoff states (/,0,0) ,..., (i,0,W_iβ - 2) where a packet is s waiting to be transmitted. In the geometric sum, q^* is the probability that such a transition will take place between any of the countdowns. Hence, q^* the probability of receiving a packet during the time-scale of one slot that is counted down. (The asterisk denote that the probability is associated with the countdown process.) Third, while waiting for a packet in the state (i,0,0,e) , q_ι (without the asterisk) represents the traffic generation probability. With a lightly loaded channel, the factor 1/qi will be the dominant part of the equation. At low loads the factor ensures the typical non-saturation behaviour where successfully transmitted traffic equals the traffic entering the transmission queue. Finally, the factor

(1 + (W^- ₀ — 1)<£ P₁- /2(1 — P_f)) appears as a consequence of the listen-bef ore-talk test in state

(i,0,0,e) (which can be replaced with 1 if the test is not implemented).

We also note that the saturation part of the equation can be written out. Performing the summation, we write the first sum as:

We let j?_fe denote the probability that the channel is busy. Since this means that at least one backoff instance transmits during a slot time, we have:

Here, n. denotes the number of backoff instances contending for channel access in each priority class i , and N denotes the total number of classes.

The probability of unsuccessful transmission p_t from one specific backoff instance (as described in the Markov chain), requires that at least one of the other backoff instances does transmit in the same slot:

At this point, we are ready to find all transmission probabilities T₁- ( i = 0,1,2,3 ) for the saturation condition (i.e. by setting p. = 1 ), without AIFS differentiation (by setting p^* = p_t ).

In the next subsection, we will look at AIFS differentiation, and then we will subsequently find expressions for p₍ , q_t , and in order to find transmission probabilities under non- saturation conditions as well.

C. The backoff countdown rate with AIFS differentiation

We see that the presented model does not encompass AIFS differentiation. In the following, however, we will include AIFS differentiation as an integral part of the count-down blocking probability, p^* .

The probability that a backoff instance senses a generic slot as idle is denoted p^* in the

Markov chain. Without AIFS differentiation, it equals the probability that all other stations do not transmit, /?• :

In this article, however, we argue that IFS -differentiation can be modelled with pretty good accuracy by adjusting the countdown blocking probability p^* . For the highest priority AC,

AC[3], we set pi = p₃ ). For lower priority ACs with a higher AIFS (for which A₁ ≥ 1 ) we reduce their countdown rate correspondingly. The additional A₁ slots, where lower priority backoff instances of class i have to suspend the backoff countdown, are model as being smeared out randomly and distributed uniformly over all slots. This assumption is required to be able to treat the A₁ slots within the Markov model. (In reality, an A₁ slot will only occur after another A₁ slot, or after a successful or unsuccessful packet transmission.) Using this simple assumption, it is possible to "scale down" the probability of detecting an empty slot.

This down-scaling can be illustrated in Figure 2. Here a large number of n slots are grouped into three groups; busy slots (due to successful transmissions or collisions), empty slots that are blocked (due to the AIFSN setting of the AC in consideration) and other empty slots that are not blocked. Each busy slot leads to a proportional share of empty slots being blocked, where A₁ yields the proportion of blocked slots for the AC in question.

By replacing p_b with the new "scaled" expression (A₁ +l)p_b , we derive a new "scaled" expression for p. , which is denoted p^* :

However, to maintain consistency in the model, a minimum bound is introduced:

K = HUn(I,/?,. +^) (18)

Thus, starvation occurs when p^* ≥ 1 in eq.(17) or when p^* = 1 in eq.(18). Since at this point T₁ — > 0 , starvation can be roughly predicted to occur when:

Although this approximate expression seems rather rough, its usefulness is striking. In the semi-saturated case, an AP is normally interested to adjust channel access parameters (such as contention windows, AIFSN etc.) for each traffic class to control the share of the channel allocated to each AC. The fundamental problem the AP faces is that by monitoring the transmitted packets it cannot know whether the absence of traffic of an AC is caused by starvation or the mere fact that the stations are not transmitting on this AC. However, by means of the above expression, the AP can simply measure the load on the traffic channel and determine whether the conditions for starvation of an AC are present or not. The expression will be validated below.

An alternative approach is to include AIFS countdown explicitly in the Markov chain [?]. However, the model becomes all too complex to be useful except for the simplest settings of the AIFS parameters. For example, in [?] the lowest priority AC is configured with AIFSN = 3, while for all other ACs the AIFS is set equal to DIFS. More flexible configurations of the AIFSNs seems complicated within this framework.

IV. ESTIMATION OF THE TRAFFIC PARAMETERS p, , q. AND

A. Delay and Service Time (under saturation conditions) In order to be able to determine an expression for p_t we first need expressions for the packet delays under both saturated and non-saturated conditions. Under saturation conditions, the queue is always full of packets ready to be transmitted (i.e. the utilization, p , of the queue is equal to 1).

To study delay under these conditions, we first deal with the delay associated with counting down backoff slots for the packets to be transmitted. The probability of a successful transmission exactly in the j-th stage is p/(l - p_{) . In each stage, h , the average countdown delay is T_e * (W^ _n - 1)/2 , and the accumulated delay for a packet sent in the j -th stage is found by summing all h stages up till j . In summary, the expected countdown delay, D_t ^CD , is:

W TXPUI- (20)

While the backoff instance is counting down, the probability of facing an empty slot is 1 - p^* while the probability of being blocked is p^* . Hence, p ^* /(I - p^* ) represents the share of slots where the countdown process is being blocked. While being blocked, the average delay is P_ST_S + (p_b - p_s)T_c . In summary, the expected blocking delay, Df , is:

Furthermore, each time a backoff instance goes from the (h - 1) -th to the h -th stage, there is a collision delay, T^* = T_C + T_τ0 , associated with the transmission attempt that a backoff instance must wait after experiencing a collision before it can sense the channel idle again. (Here, T_τo is simply defined as the difference between the collision delay, T^* , of a station that participates in a collision and that of a station not participating, T_c . When validating against simulations, its setting should reflect the implementation of the simulator.) The number of retry attempts is found by finding the average of j . In summary the expected retry delay, D* , is:

Moreover, we must add the average transmission delay, T_s , associated with the successfully transmitted packets (occurring at a probability of (1 - pf'⁺¹ ) , resulting in the delay, Df :

Finally, we must also take into account the delay Df^wp associated with dropped packets (occurring at a probability of pf'⁺¹ ). We have

ΨjB.dmp _ /"24)

D*^4wp = T_e\L_t +l)

, rCft.drop x

In conclusion, we find that the total delay under saturation conditions between packets that are either successfully transmitted or dropped, is:

p SAT ₌ p CD ₊ p B ₊ p R ₊ p T ₊ Q drop ₍₂₅₎

B. Delay and Service Time on a non-saturated channel

Under extreme non-saturation conditions, however, the post-backoff is completed before a packet arrives in the transmission queue to be transmitted. Thus, under these conditions the post-backoff will not add to the transmission delay, as we did in when calculating the saturation delays above. The easiest way to handle this, is to subtract the post-backoff delay from the expressions above.

C. Estimating p

First, for a G/G/l queue, the probability that the queue is non-empty, p is given by p = M , where λ represents the traffic rate in terms of packets per second and x is the average service time.

For simplicity, here we assume that the traffic rate faced by all backoff instances of a class is the same on all stations, and use X₁ to denote the traffic rate (in terms of packets per seconds) of traffic class i on one station. Then we have

n^ήniUp^-^) ≥ _Pi ≥ min(l,ΛA^SAr) (²⁷)

It is possible to use arguments to determine p_t with higher accuracy. In all scenarios we studied and validated, however, we did not experience any significant differences between setting P₁ =

= min(l,yl_iD₍ ^iV0Λf~&4r) . Hence, elaborating on this issue is beyond the scope of this paper.

If one measures the delay parameters in terms of μs and the rate, R₁ in terms of Mbps, and all packets are of 1024 bytes, one finds that:

M£EU. (28)

' 8 *1024 D. Estimating q_t and q^*

To estimate q. of the non-saturation model we assume that the traffic arriving in the transmission queue is Poisson distributed, i.e. that we have a M/G/l queue. q_t is the probability that at least one packets will arrive in the transmission queue during the following generic time slot under the condition that the queue is empty at the beginning of the slot. Starting out with the pdf of the length of a generic slot, b(t) , we have: b(t) = _PsS(t~T_s) + (l- p_b)S(t -T_e) + (p_b - _Ps)δ(t -T_c) (29)

Consequently, q. is calculated as: q. = l- )e-^λ>'b(t)dt = 1 - (p,e-W + (1 - p_b)e^~^ + (p_b - p_s)e^~^ ) (30) o

Noting, however, that the model is approximate of nature, it is possible to treat find the probability based on the average length of the timeslot. Hence, as a simplification, one might approximate:

Both expressions for q_t were validated in a number of scenarios, and we did not observe any significant differences between using either of them.

Ideally, q^* , should be estimated differently, since it expresses the probability of receiving a packet in the timescale of the countdown of one backoff slot, in contrast to q^* . We tested a number of different expressions for this, but observed that setting q^* equal to q. for simplicity worked as a good approximation in all the scenarios we explored.

V. THROUGHPUT Let Pi _s denote the probability that a packet from any of the backoff instances of class i is transmitted successfully in a time slot.

»A.o.od- Λ^A+1) (32)

Let also p_s denote the probability that a packet from any class i is transmitted successfully in a time slot.

N-I n-T N-I

,^•=0 \^L ~ ^τi) A=O Then, the throughput of class i , S_t can be written as the average real-time duration of successfully transmitted packets by the average real-time duration of a contention slot that follows the special time scale of our model:

T_e denotes the real-time duration of an empty slot, while T_s , T₀ denote the real-time duration of a slot containing a successfully transmitted packet and of a slot containing two or more colliding packets, respectively. The length of the longest colliding packet on the channel determines T₀. If all packets are of the same length, which we will consider in this paper,

T₀ = T_s (Otherwise refer to [12] to calculate T₀ based on the average duration of the longest colliding data packet on the channel.) Finally, T_MDSDU denotes the average real-time required transmitting the MSDU part of a data packet.

First, we notice that the share S₁ of the total data bandwidth (that is given by the current traffic load) allocated to a class i is given by:

VI. VALIDATION

A. Scenario

For validations, we compared numerical computations in Mathematica of the model presented above with ns-2 simulations, using the TKN implementation of 802.1 Ie in ns-2 [8].

In table II we can see the parameter settings for 802.11a, 802.11b, 802.11g.

TABLE H

PA RA METER SETTINGS FOK 802. i IA. 802.1 IB AND 802. i -G

Note however that parameters such as CWmin and CWmax are overridden by the use of 802. lie [2]. For our validations, we simply used the default 802.1 Ie values summarized in Table I. (Hence, a scenario where the HC adjusts these parameters and advertises them dynamically, was not considered.)

The scenario selected for validations used 802.11b [9], since this is the most widely deployed configuration per se. A configuration with the mandatory long preamble was explored [9]. According to the standard the long preamble and physical PLCP header are always transmitted at 1 Mbps, and takes 192μs in total. In our selected scenario, we also consider that all data payloads (i.e. MSDU) are of 1024 Bytes of length and are transmitted at the maximum 802.11b rate of HMbps. Furthermore, we consider a case with the basic transmission mechanism of sending a data packet followed by an acknowledgement (ACK) without the initiation RTS/CTS-mechanism. According to the standard, the MAC-part of the ACK shall be transmitted at the same rate as the proceeding frame, i.e. at 11 Mbps. However, in our scenario we consider an implementation where the MAC-part of the ACK is transmitted at IMpbs. The reason that we make this choice is to match with the implementation of the ns- 2 network simulator that is being used to validate our results.

For our validation, we consider a scenario with a number of different stations (QSTAs) contending for channel access in order to send upstream traffic to the access point (QAP). This configuration is depicted in Figure 3. This scenario correspond to the one of the two scenarios presented in the often cited paper by Mangold et al. [10], except that here we consider 802.11b instead of 802.11a.) The reason for selecting this scenario is that it represents a challenging case in terms of ensuring QoS for all the QSTAs and its backoff instances that are contending for the channel. The QAP is not actively initiating traffic. Its role in the simulation setup is to ensure that all MAC frames that are successfully transmitted on the channel are acknowledged by an ACK frame.

B. Determining the 802.1 Ib parameters for numerical calculations

With a transmission range in the order of 30m the propagation delay will be around 0.1 μs, and is neglected in the estimation of the parameters (which is often normal practice also with various simulator implementations). Conceptually, the propagation delay can be considered as an already included part of the value for the SIFS.

T, = 2βμs

TtMSDU = riow = ^frV* = rM⁾Λμ.H

T_s = {Tpjjy + TM _Aa + T_mi) + SIFS + (T_PHY + TACK-M_AC) + mini AI FS[Q], -, AlFS[Z])

= ømμs + ^^/^ + ^÷_jjr-μs) + Wμs + {I92μs + *ψμs) + SO/is = (957, Iμs) + iQμs + (βQAμs) + 50/(,s = 1321.1;«

% = (T_FHy + TMAa + 2Wi) + (EIFS)

= (957, ϊ μs) + (57FS + (TPHY + TAxm-MAcmMim + DIFS) - (957. Iμs) + (10μ» + W2μs + *ψ÷μs + 50μs) = 132.1...?._//.«

3? = ⁽Jp^' jjy + TuAo + Tm*) + (SIFS + T_PUy + T_ACκ-.rim_S + DIFS) e^β T,- = 1321.1μ^β (36)

Here T₀ denotes the time a non-colliding station has to wait when observing a collision on the channel, while T^* denotes the time a colliding station has to wait when experiencing collision.

A node-colliding station has to wait for a period determined by the fixed EBFS parameter, while a colliding station has to wait by a period determined by the configurable AckTimeout interval. For simplicity, we have set the AckTimeout equal to EIFS (which is also a normal practice with many simulators, such as ns-2), such that T₀ equals T^* , and T₇₀ = O . (Xiao [5] sets T₇₀ = EIFS - DlFS = 314μs.)

Note also that the calculation of T₀ (and T₀ ) includes the element T₁₀₂₄ since all packets on the air are of the same size. In a system where there are packets of different length T₀ (and T₀ ) should instead consider the transmission time of the longest packet, which is not difficult to estimate (e.g. see [12]).

If we had considered transmission with the RTS/CTS mechanism, on the contrary, we would have had the following changes: _τsτ⁸/oτs

_τiΩ's/σrs

_τ*^,m^<sfoτs ₌ T∞W + STFS + T_CΓS^****

C. Validation of the full-range model with starvation

Figure 4 compares numerical calculations of the analytical model with the actual simulation results. We observe that our full-range model, which models 802.1 Ie on the full range from a non-saturated to a saturated media, gives a qualitatively good match when compared with simulations. We also observe inaccuracies in the model in the saturation-part (right side) of the figure. Numerical results were also obtained for the original saturation model (omitted due to space limitations). They confirmed that the discrepancies between the curves could be mainly accredited to the inaccuracies in the saturation model.

D. Validation of the approximation for Starvation Prediction

Earlier, in Figure 4 we observed that the starvation of AC[O] and AC[I], experienced with simulations, is described with relatively good accuracy by the analytical model.

We will take a closer look at the importance of the expression to the analytical model. Figure 5 shows how the probability of a busy slot on the channel, p_b , changes as a function of the traffic load. Here, the curve for p_b is taken from exact same numerical calculations that were drawn in Figure 4. The horizontal lines illustrates the starvation conditions of the two classes according to eq.(19), i.e. when p_b = 0.167 and when p_b = 0.5. The vertical lines maps these freeze points down onto the x-axis, and translates them into the corresponding traffic loads. Despite integer resolution of the x-axis, we have linearly interpolated down to non-integer points on the axis, for illustrative purposes only. Returning to Figure 4 above, we observe that the starvation points were predicted with satisfactory accuracy.

E. Evaluating the importance of Virtual Collisions

A shortcoming of the original model, and also the model presented in this paper, is that it does not take virtual collisions into considerations. Hence, it gives a good model of a setting where each active transmission queue is allocated to a separate station. In other words, all stations are only transmitting traffic of one particular AC. (This was actually the model used in our validations above.) A scenario where all stations are transmitting traffic of all four ACs, calls for virtual collision handling between the queues on each station.

Figure (6) provides a comparison between the two scenarios, and thus gives indications of the effects of the virtual collision features of 802.1 Ie. For example, at the point in the figure where "Number of Queues pr. AC" equals 10, the virtual collision curves represent a scenario with 10 different nodes, each with four different queues. The curves without virtual collisions represent a scenario with 40 different nodes, and only one active queue of one particular AC at each station. Hence, the stations can be grouped into four groups of 10 nodes, and each group sends traffic of one AC.

It is possible to make modifications to take virtual collisions into account in the analytical model. Consider for example a situation with n nodes, and four active transmission queues on each station. A backoff instance can transmit packets if other backoff instances don't transmit, except the backoff instances of the lower priority ACs on the same QSTA. The reason for this exception is that the virtual collision handling mechanism ensures that upon virtual collision the higher priority AC will be attempted for transmission while the colliding lower priority traffic goes into backoff. This can be generalized by the expression:

If this expression is replaced with our original expression in (15), virtual collisions should be correctly incorporated in the model. The reason that we have not yet put effort into this in this paper, is that it makes the saturation model even less precise. Studying the effects of virtual collisions will be more appropriate when the accuracy of the saturation model has been improved.

vπ. CONCLUSIONS

This paper shows how analytical saturation models can be extended to cover the full range from an unsaturated to a fully saturated channel. It also provides an approximate expression to determine the starvation point of different ACs at a given traffic load and at given channel access parameters, such as the AIFSN assigned to each channel. (The other differentiation parameters also play a role in this expression as they indirectly influence the traffic load on the channel.) By measuring the channel load and by knowing the AIFSN assigned to each AC, the AP is able to tell when the starvation conditions are present for any of the ACs, independent of whether packets of these ACs are attempted to be transmitted. The model is calculated numerically and validated against simulations, using the default 802. lie parameter settings. We observed that the expansion of the model to cover unsaturated conditions gave a relatively good match with simulations. However, the accuracy of the extended model hinges on the accuracy of the original model for saturation throughput, and we observed that the accuracy of the original model have potential for further improvements. (In fact, we have also tested out some proposed improvements to the original saturation model, including the proposals in [13], without having succeeded in improving the analytical results significantly. The proposed improvements were not incorporated in our model presented in this paper, since the additional complexity seems not to pay off in terms of higher accuracy.)

The original model does not take virtual collisions into considerations. It is easy to incorporate virtual collision into the presented model. However, we observed that inclusion of it deteriorated the saturation results further. Further improvements of the original saturation model are issues for further work. With an improved model, we believe that the virtual collision features can be included without compromising the accuracy of the original model. The contributions of our work (e.g. the expansion of the saturation model into the non- saturation domain, the inclusion of AIFS differentiation and the prediction of starvation) will probably be useful and valid, even if the accuracy of the saturation model is improved, and valid, even if the accuracy of the saturation model is improved.

REFERENCES

[I] IEEE 802.11 WG, Part 11 : Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) specification, IEEE 1999.

[2] IEEE 802.11 WG, Draft Supplement to Part 11 : Wireless Medium Access Control (MAC) and physical layer (PHY) specifications: Medium Access Control (MAC) Enhancements for Quality of

Service (QoS), IEEE 802.11 e/D8.0, Feb. 2004. [3] Bianchi, G., Performance Analysis of the IEEE 802.11 distributed coordination function, IEEE J.

Select. Areas Commun., vol. 18, pp. 525-547. [4] Ziouva, E. and Antonakopoulos, T., CSMA/CA performance under high traffic conditions: throughput and delay analysis, Computer Communications, vol. 25, pp. 313-321 , Feb. 2002. [5] Xiao, Y., Performance analysis of IEEE 802.11 e EDCF under saturation conditions, Proceedings of ICC, Paris, France, June 2004. [6] Malone, D.W., Duffy, K and Leith, DJ. Modelling the 802.11 Distributed Coordination Function with Heterogeneous Load, Proceedings of Rawnet 2005, Riva Del Garda, Italy, April 2005. [7] Barkowski, Y., Biaz, S. and Agrawal P., Towards the Performance Analysis of IEEE 802.11 in multihop ad hoc networks, Proceedings of MobiCom 2004, Philadelphia, PA, USA, Sept.-Oct.

2004. [8] Wietholter S. and Hoene, C, Design and verification of an IEEE 802.11 e EDCF simulation model in ns-2.26, Technische Universitet at Berlin, Tech. Rep. TKN-03-019, November 2003. [9] IEEE 802.11 b WG, Part 11 : Wireless LAN Medium Access Control (MAC) and Physical Layer

(PHY) specification: High-speed Physical Layer Extension in the 2.4 GHz Band, Supplement to

IEEE 802.11 Standard, IEEE, Sep. 1999. [10] Mangold et. al., Analysis of IEEE 802.11e for QoS support in wireless LANs, IEEE Wireless

Comm, Dec. 2005, pp. 40-50.

[I I] Chuan, H. F., Juki,W.T. and Adel, B.M., Throughput and Delay Analysis of the IEEE 802.11e EDCA Saturation, Proceedings of IEEE International Conference on Communications (ICC 2005), Seoul, Korea, 16-20 May, 2005.

[12] G. Bianchi, Performance Analysis of the IEEE 802.11 Distributed Coordination Function, IEEE J-

SAC Vol. 18 N. 3, Mar. 2000, pp. 535-547. [13] Chuan, H. F. and Juki, W.T. Comments on IEEE 802.11 Saturation throughput analysis with freezing of backoff counters, IEEE Comm. Letters, VoI 9., No. 2, Feb. 2005, pp. 130-132.

Claims

C l a i m s

1.

A method for determining the occurrence of a starvation condition for an access category at a node in a network where multiple stations are transmitting data traffic on a shared channel and where transmitted data traffic belongs to one of two or more access categories, the method including: counting the number of successful transmissions, the number of collisions and the number empty backoff slots on the shared channel within a predetermined time frame; calculating the traffic load on the shared channel as the ratio of the sum of the number of successful transmission and the number of collisions to the sum of the number of successful transmissions, the number of collisions and the number of empty backoff slots; and determining that a starvation condition has occurred for said one of two or more access categories if the calculated traffic load is equal to or larger than l/(l+(AIFSN-2)), where

AIFSN is a number of timeslots of an AIFS parameter for said access category.

2.

The method of claim 1, wherein said node is a wireless access point.

3.

The method of claim 1, wherein said network is an IEEE 802.11 wireless network, and said network uses carrier sense multiple access (CSMA) to control access for said multiple stations.

4.

The method of any one of the previous claims, further including controlling a starvation condition for an access category at an access point in a wireless network by amending an access parameter of said access category on basis of said starvation condition indicator.

5.

A device for determining the occurrence of a starvation condition for an access category at a node in a network where multiple stations are transmitting data traffic on a shared channel and where transmitted data traffic belongs to one of two or more access categories, the method including: a first counting means for counting the number of successful transmissions, the number of collisions and the number empty backoff slots on the shared channel within a predetermined time frame; a first calculating means for calculating the traffic load on the shared channel as the ratio of the sum of the number of successful transmission and the number of collisions to the sum of the number of successful transmissions, the number of collisions and the number of empty backoff slots; and a first determining means for determining that a starvation condition has occurred for said one of two or more access categories if the calculated traffic load is equal to or larger than l/(l+(AIFSN-2)), where AIFSN is a number of timeslots of an AIFS parameter for said access category.

6.

The device of claim 5, wherein said node is a wireless access point.

7.

The device of claim 5, wherein said network is an IEEE 802.11 wireless network, and said network uses carrier sense multiple access (CSMA) to control access for said multiple stations.

8.

The device of any one of claims 5 - 7, further including an access parameter amending means operatively associated with said first determining means, said access parameter amending means adapted to amend an access parameter of said access category in response to starvation condition indicator generated by said determining means.