A QoS guaranteeing MAC layer protocol for the "underdog" traffic
© Sarkar and Paolini; licensee Springer. 2011
Received: 2 March 2011
Accepted: 12 October 2011
Published: 12 October 2011
With the tremendous boom in the wireless local area network arena, there has been a phenomenal spike in the web traffic which has been triggered by the growing popularity of real-time multimedia applications. Towards this end, the IEEE 802.11e medium access control (MAC) standard specifies a set of quality-of-service (QoS) enhancement features to ensure QoS for these delay sensitive multimedia applications. Most of these features are unfair and inefficient from the perspective of low priority (non-real time) traffic flows as they tend to starve the non-real time flows depriving them of appropriate channel access, hence throughput. To that extent, this article proposes a MAC protocol that ensures fairness in the overall network performance by still providing QoS for real-time traffic without starving the "underdog" or non-real-time flows. The article first presents analytical expressions supported by Matlab simulation results which highlight the performance drawbacks of biased protocols such as 802.11e. It then evaluates the efficiency of the proposed "fair MAC protocol" through extensive simulations conducted on the QualNet simulation platform. The simulation results validate the fairness aspect of the proposed scheme.
User priority to access category mapping
Such a mechanism facilitates differentiated QoS where HP, performance intensive traffic such as voice and video applications will enjoy less delay and greater throughput, compared to LP traffic (e.g., file transfer) [5, 6]. The QoS features in IEEE 802.11e raise two related concerns. First, these mechanisms can often be unfair and inefficient from the perspective of nodes carrying LP traffic. Second, selfish nodes can gain enhanced performance by classifying LP traffic as HP, potentially destroying the QoS capability of the system.
We envision a system where majority of traffic is non-real time, for example, in organizations like the healthcare industry, stock markets, and educational institutions, the bulk of the traffic still comprises of non-real-time flows. In these scenarios, it becomes essential to provide acceptable performance metrics for these non-real-time traffic in the face of growing real-time multimedia traffic. The 802.11e MAC scheme could have been justified if the majority of traffic in the system was real time. However, in these scenarios where the major chunk of network traffic is non-real time, the protocol will starve the non-real-time traffic which is the dominant traffic in most of these organizations and can present critical performance issues and diminish user satisfaction if not handled smartly [1, 2]. Even a lone real-time flow can hog the network and starve the non-real-time flows thereby drastically affecting the network performance . This article raises the following concerns: (i) will the standard still favor HP traffic at the cost of LP traffic starvation, especially when the network traffic is LP-centric? (ii) what will happen if the applications start falsely classifying their traffic as HP in pursuit of preferential service ? Such instances might destroy the QoS capabilities of the network. The research community has raised concern over these issues of fairness [8–11]. The standard does not address these issues as it mainly deals with HP traffic, for which it allocates a major share of its resources.
This motivates us to propose a MAC protocol that does not starve the LP traffic or "underdog" traffic in face of HP traffic. Our scheme especially prevents resource hogging by the few HP traffic flows even when the predominant traffic in the network is LP. In this article, we thereby propose a MAC scheme which imparts fairness to the traffic ("Underdog"), i.e., getting exploited at the cost of preferential service offered by the standard to real-time traffic. The purpose of designing this scheme is to prevent starvation of non-real-time LP data traffic while still maintaining an acceptable quality-of-service (QoS) performance for real time, delay sensitive HP traffic. We do so by introducing a transmission opportunity for LP traffic in the contention-free phase (CFP) of an IEEE 802.11e MAC protocol. Traditionally, IEEE 802.11e MAC would provision for only HP traffic transmission during the CFP. In our proposed MAC scheme, we advocate the introduction of transmission slots for LP traffic as well during CFP.
To explicitly understand the drawbacks of IEEE 802.11e (the standard which caters primarily to HP traffic) and thus motivate the need for a fair MAC protocol, we first analyze a hybrid-MAC scheme which mimics the 802.11e MAC in every essential respect. The analytical expressions attained for throughput and delay values of this hybrid MAC are discussed with the help of MATLAB simulation results. The drawbacks of an 802.11e-like MAC become apparent from these results. We thereby propose our fair MAC scheme. We perform extensive simulations on the network simulation platform QualNet to verify the feasibility and performance efficiency of our MAC scheme in comparison with the basic 802.11e protocol. Simulation results validate the performance efficiency of our scheme.
The rest of the article is organized as follows. In Section 2, we provide a system model for our 802.11e-like Hybrid-MAC and derive throughput and delay expressions for the MAC along with MATLAB simulation results. In Section 3, we present and discuss our proposed MAC scheme. In Section 4, we present QualNet simulation results and provide an analysis and a comparative study of our scheme with 802.11e. We finally conclude the article in Section 5.
2. Analysis of a hybrid-MAC
We intend to derive analytical expressions for modeling throughput and delay characteristics of a MAC protocol that mimics the IEEE 802.11e in every essential respect. We do so by first proposing a simplified model of the IEEE 802.11e MAC.
2.1 System model
We set out to analyze the 802.11e MAC protocol. We realize that an analysis of the exact scheme is cumbersome. We thus propose a hybrid-MAC model that resembles the 802.11e MAC in most essential respects. Our MAC model provides us with an abstraction of the essential features of 802.11e MAC, while avoiding the complex details of the latter. We believe that the insights obtained using our model are applicable to the 802.11e scenario. Our system model can be thought of as a hybrid MAC model which operates in both the contention and CFPs alternately, akin to a legacy 802.11 MAC protocol  with both its (a) distributed coordination function (DCF) and (b) point coordination function (PCF) modes enabled . While DCF is based on the contention-based CSMA/CA mode of channel access, PCF is based on the polling mechanism. Limited QoS support in the legacy 802.11 standard is available through the use of the PCF. The DCF phase mimics the enhanced distributed channel access (EDCA) mechanism which is a contention-based channel access scheme while the PCF mimics the HCCA which is based on a polling mechanism. EDCA and HCCA are used to provide prioritized and parameterized QoS services, respectively, in 802.11e.
2.2 Modeling throughput
which prevents the PCF from transmitting a poll frame in between a Data/CF-Poll and Data/CF-ACK transaction.
Given the definitions of SIFS and DIFS, Equation 10 can be understood as the sum of times required to conduct a successful packet transmission in the CP: the STA must first wait a DIFS amount of time to detecting a channel idle before proceeding to transmit, then an (H + P)/R amount of time to for an interface to transmit a packet consisting of H header and trailer bits and P payload bits at a data rate R, then a τ amount of time for propagation of the data packet, then a SIFS amount of time before the receiving STA's interface can transmit an acknowledgement frame, then (ACK/R) time to transmit the acknowledgement frame, and finally another τ amount of time for propagation of the acknowledgement.
where R is the fixed transceiver data rate.
2.3 Modeling delay
that describes the mean time a frame waits in queue to be serviced by the MAC, where the queue is modeled as a M/G/1 queue (a single server with frame arrivals having a Poisson distribution and service time having a general distribution). Total actual delay Dactual is modeled as the sum of (20) and an expression for the expected value of HOL delay which takes into account backoff delay.
since the number of different contention window sizes will be the exponent of the ratio of CWmax to CWmin. Equation (22) therefore gives the maximum number of retransmission attempts that will be made, if the initial transmission should result in a collision. For a FHSS based PHY, rmax is 6.
In Equation 25, we account for polling frames that may either be CF-Poll with no data (subtype 6 or 0110) or CF-Poll + Data (subtype 2 or 0010) as represents the mean length of polling frame bits transmitted by the point coordinator during the CFP. Similarly, we account for acknowledgement frames that may be CF-ACK with no data (subtype 5 or 0101) or CF-ACK + Data (subtype 1 or 0001) as represents the mean length of acknowledgement frame bits transmitted by all the stations during the CFP. The remaining terms in (25) follow from (15) and account for interframe delays, management and control frames, and propagation times.
2.4 Analysis of the hybrid-protocol simulation results
3 The proposed fair MAC scheme
Providing fair channel access opportunities to both HP and LP traffic such that adequate throughput is enjoyed by non-real time (or LP) flows while still supporting the QoS constraints of real-time traffic (or HP flows) is the main objective of this study, especially under scenarios where the bulk of network traffic is non-real time. Thus, we have designed a scheme that would be suitable for networks dominated by LP traffic and one that would eventually revert back to normal 802.11e functionality in the absence of LP traffic. Before we delve into the details of our fair MAC scheme, it is worthwhile to examine the existing IEEE 802.11e MAC protocol.
3.1 Examining IEEE 802.11e MAC protocol
To enhance the QoS support, IEEE 802.11e introduces a protocol called the HCF which includes two medium access mechanisms: contention-based channel access and controlled channel access which are referred to as the EDCA and HCCA. With 802.11e, there are two phases of operation within a superframe, i.e., the CP and a CFP. Each superframe begins with a control frame called the Beacon frame followed by the CP and then the CFP. Figure 2 pictorially depicts a typical 802.11e superframe.
The EDCA is used in the CP only, while the HCCA is used in both phases. QoS polling for HCCA can take place during CP as well. EDCF and HCCA together support up to eight priority traffic classes (TC). Each TC starts with a backoff after detecting the channel being idle for an arbitration interframe space (AIFS) period of time. The AIFS can be chosen individually for each TC and thus provides a deterministic priority mechanism between the TCs. Thus, a transmit opportunity (TXOP) almost always is given to the TC with the highest priority. During the CP, access is governed by EDCF, though the hybrid coordinator (HC--generally co-located within the AP) can initiate HCF access at any time. During the CFP, the HC issues a QoS CF-Poll frame to a particular station to give it a TXOP, specifying the start time and maximum duration. No station attempts to gain access to the medium at this time and thus the station to which the CFP-poll frame was sent has unhindered access to the medium. The HC has available, over time, a snapshot view of the per-TC, per station, queue length information in the cell, including that of the AP itself. This information is sent to the HC by stations periodically. With this information, the HC decides which station (including itself) to allocate TXOPs during the CFP. At minimum, the following needs to be considered: (a) priority of the TC, (b) required QoS for the TC (low jitter, high bandwidth, low latency, etc.), (c) queue lengths per TC, (d) queue lengths per station, (e) duration of TXOP available and to be allocated, and (f) past QoS seen by the TC. Thus, even during the HCCA (as during the EDCA), TXOPs are given to traffic of HP as well.
3.2 Motivating the need for a FAIR MAC scheme
3.3 Our FAIR MAC scheme
Conventionally, contention-based channel access schemes have been used for LP data transmission whereas "polling" and thereby dedicated channel access schemes have been thought of as the most appropriate way of transmitting HP (delay-sensitive) data. It is a well-established fact that if a MAC layer protocol has to cater to various types of traffic (both HP and LP), it is imperative that it employs both contention-based and polling channel access mechanisms. Thus, our fair MAC scheme alternates between a contention-based channel access mechanism, which we refer to as the CP, and a polling-based channel access scheme, which we refer to as the CFP as shown in Figure 6. Our system offers channel access opportunities to both traffic types (HP and LP) during the contention period, allocating higher preference to the HP traffic to grab the channel over the LP traffic. However, deviating from the norm, during the CFP, LP traffic is included in the polling list and thus polled by the HC along with the HP traffic. The duration of the CFP is equally distributed to allocate transmission time for all traffic flows in the network. The polling scheme is implemented in a circular queue such that all traffic flows gets polled almost equally. The HC, co-located with the AP, polls every station in the polling list starting with the traffic flow which has the highest priority and subsequently servicing the traffic flows on the polling list in the descending priority order till the lowest priority traffic flow is served. The HP flows still retain their precedence in the queue over the LP flows. However, such dedicated service during the CFP incentivizes LP traffic to deter from falsely classifying itself as HP and thus preserves system sanctity.
During the CP, a node with packets to transmit contends for channel access with a certain probability. QoS differentiation is enforced by allowing packets in HP queues to contend for channel access with higher probability than packets in LP queues. We assume that nodes are transmitting to an AP that can invoke a CFP by issuing a poll request to one or more nodes. These polled nodes can then transmit without any contention. Users can classify their applications as either HP or LP. Users are expected to take advantage of the MAC's QoS features by declaring their delay sensitive applications as HP, and delay tolerant applications as LP. The AP needs to decide what fraction of time the system will spend in the contention and CFPs. Our protocol is very similar to 802.11e's HCF, with the CP corresponding to 802.11e's random access or EDCA functionality and the CFP corresponding to 802.11e's polled access or HCCA functionality . More specifically, our system corresponds to the HCCA/EDCA mixed mode  of operation.
The vast majority of moderate-rate delay sensitive HP applications (such as VoIP and moderate resolution video streaming) and delay tolerant LP applications (e.g., file transfer and email) can be supported by the random access or contention functionality of 802.11e. If the number of users with delay sensitive traffic is relatively large, then polling or contention-free access is inappropriate because of the large delay incurred in waiting for one's turn . Therefore, from the HP user's viewpoint, it is more advantageous to operate in the CP rather than the CFP. On the other hand, operating the network mainly in a CP is both unfair and inefficient as far as LP applications are concerned. It is unfair because HP applications will obtain better throughput than LP applications as they contend for channel access more aggressively. It is inefficient because, even in the absence of HP applications, LP applications are forced to be conservative in accessing the channel. Thus, arises the interesting dilemma of how long should these CP and CFP periods be chosen such that system performance is maximized. We choose to investigate this problem in our future study.
It is also noted that polling is known to be very efficient throughput-wise, but leads to large delays because a user has to wait for his/her turn to transmit . Since LP traffic is delay-tolerant, polling is an efficient method to serve such traffic. Another consequence of the throughput efficiency of polling is that the system does not need to spend too much time in the CFP to serve LP users. Thus, the negative impact of our scheme on HP users is mild. Most of the time the system is in the CP where HP users can enjoy good delay performance of prioritized random access. Our incentive mechanism exploits the difference in performance required by HP and LP applications, to simultaneously satisfy QoS requirements for all users. HP applications, such as VoIP, have tight delay constraints but do not require very high throughput. LP applications such as file transfer have no particular delay constraints but require relatively high throughput for reasonable session completion times. Polling LP users during the CFP ensures that these users are guaranteed a certain minimum level of throughput, ensuring there is no motivation for LP users to falsely declare their traffic type as HP. This in turn implies that HP users encounter decreased interference from LP users during the CP leading to better delay performance.
It is to be noted that the duration of the CFP phase has a significant impact on the delay encountered by the HP traffic. This is because, the longer the CFP (to accommodate the several LP flows in a network), the more the time required for the system to transition into the CP, thereby making the HP traffic wait for a longer period of time to get an opportunity to transmit their delay sensitive data. We intend to address this issue in a quantitative manner in our future study. In Section 4, we provide a numerical analysis of the above fact. We want to emphasize that an absence of LP traffic flow in the network will make our scheme behave exactly in the standard 802.11e fashion. Thus, no undue delay will be encountered by the HP traffic flows. In summary, the extra opportunity to transmit data by the LP flows during the CFP phase leads to significant increase in their throughput with minor dent in the delay performance of the HP flows.
4 Simulation results
We evaluated our proposed fair MAC scheme using the network simulation platform QualNet 4.5 . Our network topology was comprised of several wireless stations (or nodes) and one AP, all located within each others' "hearing" range (i.e., every station is able to detect a transmission from any other station in the network). The nodes were placed in the default terrain with default dimension settings. Each simulation has been run for 600 s and each reported value has been averaged over 15 runs. The simulation results were analyzed using the QualNet analyzer.
802.11e with HCCA enabled
802.11b-data rate 2 Mbps
200 time units (TU)
50 TU, 160 TU
Type of traffic source
CBR with precedence 5, 6, 7 for HP traffic CBR with precedence 0,1 and FTP generic for LP traffic
System parameter values for three different PHYs as specified by IEEE 802.11 standard 
Slot time (μs)
4.1 Throughput performance for network scenarios I and II
In network scenario I, there is about an average 20% increase in LP traffic throughput in our scheme in comparison to 802.11e. This is expected, since our scheme provisions for extra time to transmit LP traffic in the CFP which 802.11e does not. This trend is evident in both cases where the CFP duration is set to 50 and 160 TU, respectively. However, it is to be noted that the throughput curve for LP users in both cases (CFP = 50 and 160 TU) shows a downward trend as the number of flows in the network increases. This can be attributed partly due to an increase in collisions during the CP. The major impact is however due to the thin time slicing of dedicated time slots allocated to each LP traffic flow as the number of flows increase in the network. This is a necessary evil since we have to accommodate all traffic flows in the network and yet not increase the total time duration of CFP. Also note that we present the results of the worst case traffic scenario. In our simulation, nodes have data to send all the time (constant bit rate--CBR--traffic) leading to a claim on the time slot during CFP always. In reality, not all nodes will have data to send at all times, thereby potentially preventing such thin time slicing for the data carrying nodes during CFP.
An increased duration of CFP in our scheme does lead to an increased throughput for both HP and LP traffic. If we increase the CFP duration to 160 TU, the throughput of our scheme increases about 50% more than the standard IEEE 802.11e scheme as depicted in Figure 8. When the CFP duration is increased to 160 TU (superframe duration = 200 TU), there is a severe drop in LP throughput in the standard 802.11e due to the drastic increase in collisions during the CP which now comprises of a small fraction of time of the total superframe duration. This proves the relevance of CFP duration on the network performance. However, a longer CFP duration also increases the delay which can be detrimental, especially for HP traffic (Figures 10 and 11).
There is a 3% reduction in HP traffic throughput in our scheme in comparison to 802.11e. This is due to the thinner time slicing for each HP traffic flow during the CFP in our scheme to accommodate the extra LP flows within the stipulated CFP duration (for example 50 and 160 TU in our simulations). It is noted that the increasing number of LP traffic flows also does not significantly affect the average HP throughput, since HP users contend for the channel more aggressively (better EDCA parameter set than LP) than the LP users and thus almost always gain access to the channel over LP users. The minimal decrease in throughput of HP traffic with an increase in the number of LP traffic flows (5 HP + 10 LP flows configuration versus 5 HP + 30 LP flows network configuration) in Figures 7 and 8 is attributed to the smaller time segment devoted to each HP traffic flow during the CFP as the number of traffic flows in the system increases. In Figure 9, the drop in HP traffic throughput with an increase in the total number of traffic flows in the network is attributed to the more aggressive channel contention that occurs between the increasing number of HP flows in the system, thereby leading to higher collisions and hence lesser overall HP throughput.
4.2 Delay performance for network scenarios I and II
Moreover since majority of the flows are LP, this deterioration is compensated by the almost 20% increase in throughput of the LP traffic. Figures 12 and 13 demonstrate the fact that the delay of LP traffic decreases considerably in the proposed scheme when compared to the standard IEEE 802.11e.
In Figure 11, as the number of HP flows increase, the HP delay increases both in the standard 802.11e protocol and our scheme as well. This is due to the increased number of collisions during the CP as HP flows in the network increases thereby leading to longer wait time for data delivery. In network scenario II (Figure 11), the increase in HP delay over standard 802.11e is approximately less than 7%. Once again, in scenarios where LP traffic predominates, this brunt in HP delay performance is acceptable, especially if we consider the havoc that can be wreaked in the network if selfish LP users start classifying their traffic as HP and thereby wreck the whole notion of QoS  in absence of a fair MAC scheme like ours.
This article focuses on "protecting" the "underdog" or non-real time data traffic in the face of the growing multimedia traffic that treads the wires in recent times. It provides analytical expressions to model the throughput and delay of a hybrid MAC scheme akin to IEEE 802.11e followed by MATLAB simulation results which highlight the drawback of protocols that are biased toward protecting and guaranteeing QoS for delay intolerant HP traffic thereby starving the delay tolerant non-real time flows. In addition, this article proposes a MAC scheme which provides performance (throughput) guarantees to non-real time traffic in face of real-time traffic such that they are not bandwidth starved. However, the new MAC protocol geared toward protecting the "underdog" traffic also aims to preserve the QoS requirements of delay-intolerant, HP. The performance of the new MAC scheme is compared against the standard IEEE 802.11e scheme using the QualNet simulation platform. The results prove that the proposed MAC scheme indeed boosts the throughput and delay performance of non-real time traffic (by as high as 50%) with a minimal dent in throughput (about 3%) and delay (about 6%) of real-time traffic though staying within the acceptable service range of such traffic. The QualNet simulation results show that the performance improvement of our proposed method is particularly significant when the traffic mix comprises of mainly delay tolerant traffic. Conventionally, using schemes like IEEE 802.11e, LP traffic would have been sabotaged by a small population of HP traffic which would have conventionally squeezed the majority of network resources to ensure its high performance. Our protocol alleviates this particular problem and proves that a fairer scheme is indeed possible and complete sabotage of non-real time traffic is not required to meet the demands of high priority traffic.
This study is based upon work supported by the National Science Foundation under Grant No. 0737048.
- itbusiness.ca-business advantage through Technology retrieved at[http://www.itbusiness.ca/it/client/en/home/News.asp?id=48078&PageMem=3]
- pcdistrict-news, retrieved at[http://www.pcdistrict.com/internet-traffic-is-still-dominated-by-p2p-file-sharing-review1510-16.html]
- Medium Access Control (MAC) Quality of Service (QoS) Enhancements IEEE 802.11e Working Group, New York;Google Scholar
- IEEE Standard for Information Technology-Part 11: Wireless LAN (MAC) and (PHY) Specifications IEEE Std 802.11 6/12/2007Google Scholar
- Ni Q, Turletti T: QoS support for IEEE 802.11 Wireless LAN. Nova Science Publishers; 2004.MATHGoogle Scholar
- Lindgren A, Almquist A, Schelén O: Quality of service schemes for IEEE 802.11 wireless LANs--an evaluation. J Mobile Netw Appl 2003, 8(3):223-235. 10.1023/A:1023389530496View ArticleGoogle Scholar
- Garg P, Doshi R, Greene R, Baker M, Malek M, Cheng X: Using IEEE 802.11e MAC for QoS over wireless. Proceedings of IEEE Performance, Computing, and Communications Conference 2003.Google Scholar
- Nuggehalli P, Sarkar M, Rao RR: QoS and selfish users: a MAC layer perspective. In Proc of IEEE GLOBECOM. Boston, MA; 2007.Google Scholar
- Obeid D, Sheeshia S: Modeling of a polling-based access scheme for 802.11. Proceedings of the 32nd IEEE Conference on Local Computer Networks 2007.Google Scholar
- Sarkar M, Ray M, Nuggehalli P: Evaluating a QoS-supportive MAC layer protocol for WLANs. In Proceedings of IEEE MILCOM 2008. San Diego, CA; 2008.Google Scholar
- Venkatakrishnan BA, Selvakennedy S: An enhanced HCF for IEEE 802.11e wireless networks. In Proceedings of the 7th ACM International Symposium on Modeling, Analysis and Simulation of Wireless and Mobile Systems. Venice, Italy; 2004.Google Scholar
- Khalaf R, Rubin I: throughput and delay analysis in single hop and multihop IEEE 802.11 networks. In Proc of 3rd International Conference on Broadband Communications, Networks and Systems, 2006 (BROADNETS 2006). San Jose, CA; 2006:1-9.View ArticleGoogle Scholar
- Truong HL, Vannuccini G: The IEEE 802.11e MAC for Quality of Service in Wireless LANs. IBM Research, Zurich Research Lab, Switzerland; 2003.Google Scholar
- Ansel P, Ni Q, Turletti T, Planete P, FHCF: a fair scheduling scheme for 802.11e WLAN. INRIA Research Report 2003., 4883:Google Scholar
- Mangold S, Choi S, Hiertz G, Klein O, Walke B: Analysis of IEEE 802.11e for QoS support in wireless LANs. IEEE Wirel Commun Mag 2003, 10(6):40-50. 10.1109/MWC.2003.1265851View ArticleGoogle Scholar
- Grilo A, Macedo M, Nunes N: A scheduling algorithm for QoS support in IEEE 802.11e networks. IEEE Wirel Commun Mag 2003, 10(3):36-43. 10.1109/MWC.2003.1209594View ArticleGoogle Scholar
- Gao D, Chai J, Zhang L: Physical rate based admission control for HCCA in IEEE 802.11e WLANs. In Proceedings of the 19th international Conference on Advanced Information Networking and Applications (AINA). Taipei, Taiwan; 2005:479-483.Google Scholar
- QualNet 4.5"[http://www.scalable-networks.com]
- Bianchi G: Performance analysis of the IEEE 802.11 distributed coordination function. IEEE J Sel Areas Commun 2000, 18(3):535-547. 10.1109/49.840210View ArticleGoogle Scholar
- Al-Sayyed R, Pattinson C, Dacre T: VoIP and database traffic co-existence over IEEE 802.11b WLAN with redundancy. In Proceedings of the International Conference on Computer, Information and Systems Science and Engineering. Barcelona, Spain; 2007:25-27.Google Scholar
- Agrawal DP, Zeng Q: Introduction to Wireless and Mobile System. Thomson, Brooks/Cole; 2003.Google Scholar
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.