CHANNEL- AND QoS-AWARE SCHEDULER DESIGN FOR WiMAX NETWORKS

In this section some design guidelines are given for a “channel-aware” and “QoS-aware” scheduler to be implemented at the BS of a WiMAX PMP network for the delivery of DL traffic to a set of distributed SSs with active connections of different traffic nature. The scheduling algorithm, which is running at the WiMAX base station, needs to fulfill the following requirements:

§  Efficient link utilization: The scheduler shall take opportunistic decision and not assign a transmission opportunity to a flow with a currently low-quality link, because the transmission will be wasted.

§  Delay bound: The algorithm shall be able to provide delay bound guarantees for individual flows, to support delay-sensitive applications; besides, it shall prevent too late packet transmissions from wasting bandwidth.

§  Fairness: The algorithm shall redistribute available resources fairly among flows; thus providing short-term fairness to error-free flows and long-term fairness to error-prone flows.

§  Throughput: The algorithm shall provide guaranteed short-term throughput to error-free flows and guaranteed long-term throughput to all flows.

§  Implementation complexity: A low-complexity algorithm is necessary to take quick scheduling decisions.

§  Scalability: The algorithm shall operate efficiently when the number of flows sharing the channel increases.

To match all the above-mentioned needs, (1) a class-based wireless scheduling is recommended to fulfill the WiMAX service class differentiation needs, and also to simplify interworking with the Internet (supporting class-based differentiated services); (2) “per class” service differentiation must be achieved, and simple “per flow” mechanisms (e.g., lead/lag counters per each flow) must be provided to guarantee fair service to traffics within a class; (3) channel awareness must be exploited to make efficient use of wireless resources, and a compensation technique for missed transmission opportunities must be provided to guarantee proportional fairness in sharing the bandwidth; (4) useless compensation must be avoided through simple measures, e.g., periodic buffer cleaning of over-delayed packets could be used; and (5) simple expedients must be provided to avoid monopolization of the scheduler by a lagging flow after the recovery of its channel, and to guarantee graceful throughput degradation of leading flows; this could come at a very low cost by using combination of lead/lag counters and queue parameters.

In Figure 1, the reference channel- and QoS-aware scheduling architecture is illustrated. It is considered to be implemented at the MAC layer of a WiMAX BS, which manages local traffic queues for packet delivery over the DL channel. The illustrated framework must be able to provide per-class differentiated QoS and fair service to flows in the same class through the operation of the following QoS-support modules:

§  An error-free service scheduler that decides how to provide service to traffic flows based on an error-free channel assumption.

§  A lead/lag counter for each traffic flow that indicates whether the flow is leading, in sync with, or lagging its error-free service model and in which extent.

§  A compensation technique that is used to improve fairness among flows. A lagging flow is compensated at the expense of a leading flow when its link becomes error free again. The system maintains credit/debit counters for each lagging/leading flow.

§  Separate per-class packet queues used to support rtPS, nrtPS, and BE traffic flows.

§  A means for monitoring and predicting the channel state of each backlogged flow.

Figure 1: Channel-aware QoS scheduling architecture in the BS.

SOME OPEN RESEARCH ISSUES | Scheduling in WiMAX

In the foregoing discussion we have alluded to various incomplete studies and opportunities for further research. In this section, we shall list some open research issues related to QoS scheduling in WiMAX, viz.,

§  An open issue is to how to provide tight delay and throughput guarantees in rtPS and ertPS traffic classes while using round robin schedulers, i.e., WRR and DRR. Also, the applications of other variations of round robin, such as smoothed round robin and stratified round robin, need to be further investigated.

§  Integration of time-stamped schedulers, such as WF²Q, SCFQ, and SFQ, with IWFQ type of algorithms for wireless networks, thus designing schedulers that are specific for high priority traffic classes of WiMAX.

§  The use of IWFQ within rtPS and ertPS traffic classes needs to be investigated further, as it requires significant support from the MAC layer.

§  The CIF-Q algorithm has higher complexity, and authors have argued that the wireless part has a lower bandwidth, which is obviously not true in the case of WiMAX. Thus, an interesting area for further work is to investigate the behavior of CIF-Q with rtPS and ertPS traffic classes.

§  Design of dynamic and efficient priority-based mechanisms for distribution of the allocated grants among the various classes at SS. The issues of inter- and intraclass fairness need to be investigated further.

§  Development of intelligent bandwidth requests, admission control, and bandwidth allocation schemes at the BS. Also, an investigation of efficient and adaptive techniques for traffic policing and shaping at BS/SSs is required.

§  Transforming the schedulers designed for PMP mode into the mesh mode of operation for WiMAX.

§  It is important to investigate the use of RED-based buffer management for BE at SSs. Also, the use of modern scheduling techniques, such as multiuser diversity, that exploit variations in wireless channels must be investigated. Similarly, cross-layer approach, adaptive bandwidth requests, and queue length based scheduling are needed to be investigated further. One of the basic issues with the cross-layer approach is how to cater for multiple SSs in the same time frame.

§  Performance measurements of WiMAX based services is also very important. For this purpose a test bed has been developed, which employs the well-known Packet-E-Model for estimation of voice quality. It is also important to develop models that can estimate performance of various services (voice/video/data) offered in WiMAX. Hence, this area also needs to be investigated further.

§  One of the important issues is cooperation between various current technologies, for the benefit of customers. In this regard, the interoperation WiMAX with 3GPP’s High Speed Packet Access based technologies would be very worthwhile to investigate.

PERFORMANCE MEASUREMENTS FOR WiMAX NETWORKS

It has been noted that, at the time of writing, there is not much published work available regarding real-life performance evaluation of WiMAX networks. However, some of the currently available literature will be briefly described in this section.

WiMAX test-bed based results for measurements in the field have been reported. The Alvarion test bed, BreezeMax http://www.alvarion.com/, operating in the 3.5 GHz licence band and fully IEEE 802.16-2004 compliant, has been employed for measurements. Experimental data has been collected for four nodes operating in PMP mode in a rural residential environment. A sectorial antenna with a gain of 14 dbi, covering all three SS (FDD half duplex), has been deployed. All nodes run using a Linux distribution and are attached to WiMAX equipment through the Ethernet. Data flow and CBR VoIP are generated by the freely available tool known as D-ITG, http://www.grid.unina.it/software/ITG/;

It has been observed that the performance of a G.711 codec is far too low to be acceptable and SS cannot support more than two high quality calls. The G.723.1 codec outperforms G.729.2. It has been pointed out that UL measurements contradict simulation results and its earlier version, where larger delays in UL are ascribed to bandwidth request mechanisms and PHY overhead. Due to activation of piggybacking for bandwidth reservation. It has also been pointed out that the R-factor, E-model, needs to be considered for scheduler design, though, nothing has been mentioned about the type of scheduler that was being used at SSs and BS during the test-bed measurements.

Recently, a performance study of UL Scheduling algorithms for PMP WiMAX has been carried out. It has studied major scheduling algorithms using the NS-2 simulator. Also, it provides pseudocode for various schedulers in a simple and accessible manner. The existing scheduling mechanisms have been divided into three categories: homogenous, heterogenous, and opportunistic types. It has been reported that EDF and (EDF+WFQ+FIFO) result in the lowest average delay for rtPS and ertPS; WRR, WFQ, and (EDF+WFQ) provide a fair distribution of bandwidth; the (EDF+WFQ) hybrid setup is fairer than (EDF+WFQ+FIFO). In addition, it has been concluded that most of the legacy schedulers are not very suitable for WiMAX applications. However, WFQ, cross-layer, and queueing theoretic-based schedulers are seen to be promising candidates for applications in a WiMAX network.