Scheduling Algorithms in Broad-Band Wireless Networks YAXIN CAO AND VICTOR O. K. LI, FELLOW, IEEE IEEE PROCEEDINGS OF THE IEEE, VOL. 89, NO. 1, JANUARY 2001 報告者 : 李宗穎

Outline Introduction System Model Major Issue in Wireless Scheduling Different Scheduling Methods Compared and Conclusion

Introduction The characteristics of wireless communication pose special problems that do not exist in wireline networks high error rate and bursty errors location-dependent and time-varying wireless link capacity scarce bandwidth user mobility power constraint

Wireless Network Model Downlink the base station has full knowledge of the status of downlink queues Uplink The base station performs uplink scheduling based on these requests and related information

Wireless Link Model good (or error-free) bad (or error) the wireless link is assumed to be error-free bad (or error) packets transmitted on the link will be corrupted with very high probability

Major Issues Wireless Link Variability Fairness QoS Data Throughput and Channel Utilization Power Constraint and Simplicity

Wireless Link Variability wireless channels are more error-prone and suffer from interference, fading, and shadowing some mobile hosts may enjoy error-free communication with the base station, while others may not be able to communicate at all

Fairness wireline media may be considered error-free ,the wireless link is actually in an error-state the packet will be corrupted and the transmission will waste transmission resources in error-state

QoS at least prioritized scheduling service for aggregated traffic with QoS differentiation per-flow-based guaranteed QoS performance, such as delay or jitter bound

Data Throughput and Channel Utilization minimize unproductive transmissions on error links maximize the effective service delivered and the utilization of the wireless channels

Power Constraint and Simplicity minimal number of scheduling-related control messages the scheduling algorithm should not be too complex

Channel state dependent packet scheduling (CSDPS) Bad state LSM mark Waiting time out It does not have any mechanism to guarantee bandwidth and the algorithm does not provide any guarantees on packet delay

CSDPS + CBQ (class-based queueing) A class is called unsatisfied if it has persistent backlogs, and the service it recently received is less than its allocated fraction When class exceeds its allocated bandwidth share and contributes to any other class’ unsatisfied state. Such a queue is called a restricted queue

Idealize Weight Fair Queue (IWFQ) (1/3) Queue size leading lagging in sync

Idealize Weight Fair Queue (IWFQ) (2/3) When a packet of sequence number of flow arrives, it is tagged with virtual service start time Si,n and finish time fi,n Si,n = max{v(A(t)), fi,n-1} fi,n = Si,n + Li,n/ri The scheduler always chooses to serve the packet with the smallest finish time Li,n : packet size of the arrived packet V(A(t)) : system virtual time defined in WFQ ri : service rate allocated to flow

Idealize Weight Fair Queue (IWFQ) (3/3) Lagging bound all flows that will be compensated is bounded by B bits A flow i with weight ri is allowed to compensate a maximum of Leading bound for more than li bits, it will only surrender up to li bits of service share to other flows later on To implement this bound, the scheduler checks each leading flow after transmitting one packet

channel-condition independent packet fair queueing (CIF-Q) (1/4) Each flow has its own queue, and the real error-prone scheduling system is S associated with an error-free system Sr Arrived packets are put into queues both in S and Sr (virtual queue) No link error, packet is chosen in Sr and served in both S and Sr Link error, the real packet in the queue of S is kept, but the virtual packet in the queue of Sr is still served

channel-condition independent packet fair queueing (CIF-Q) (2/4) lagi is flow i serving different between S and Sr To achieve graceful degradation, a parameter α is used to define the minimal average rate (αri)

channel-condition independent packet fair queueing (CIF-Q) (3/4) packet in S is transmitted unless one of the following situations occurs a) Link is an error state b) Leading flow and receive more than αri Lagging flows have higher priority to receive additional service in a) and b) the compensation is distributed among the lagging flows If no lagging flow, the additional service is distributed to nonlagging flows

channel-condition independent packet fair queueing (CIF-Q) (4/4) Compared with IWFQ, CIF-Q improves scheduling fairness by associating compensation rate and penalty rate with a flow’s allocated service rate and guaranteeing flows with error-free links with a minimal service rate

Server-based fair approach (SBFA) (1/3) SBFA allocated to some compensation server(s), called long-term fairness server The scheduler maintains two queues, packet queue (PQ) and slot queue (SQ) for each flow SQ is the virtual queue in this system

Server-based fair approach (SBFA) (2/3) Round Robin Policy

Server-based fair approach (SBFA) (3/3) Problem LTFS needs preallocated network resources the algorithm does not work well if the packet size of a flow is variable

Improved channel state dependent packet scheduling (I-CSDPS) (1/2) deficit counter (DC) keeps a record of the total credit received less the credit used compensation counter (CC) CC keeps track of the amount of lost service for each flow quantum size (QS) Determines how much credit, in number of bits or bytes, is given to a flow in each round

Improved channel state dependent packet scheduling (I-CSDPS) (2/2) bad state At the beginning of each round αCC amount of credit is added to DC, and CC is decreased by the same amount, where 0 < α < 1 QS1 = 100, QS2 = 50 α 1 = 1/3, α2=1/2

Comparison Delay Bound Long-tern Throughput Guarantee Short-tern fairness Pre-allocated Resource for Compensation CSDPS CSDPS + CBQ ☆ IWFQ CIF-Q SBFA I-CSDPS

Future Work Adaptive Error-Correction Coding and Deferment of Transmissions Scheduling in CDMA Networks—Multiple Servers and Multiple Link States Integration of Admission Control, Scheduling, and Congestion Control

Conclusion This paper presented a comprehensive and in-depth survey on current research in wireless packet scheduling. The major issues in wireless scheduling were discussed