Performance Evaluation of the IEEE MAC for QoS Claudio Cicconetti, Alessandro Erta, Luciano Lenzini, and Enzo Mingozzi IEEE Transactions On Mobile Computing, VOL. 6, NO. 1, JAN 報告者：李宗穎

Outline  Background  Simulation Environment  Performance Evaluation  Conclusions

Introduction  This paper focus following Frame-based point-to-multipoint mode The BS in a Time Division Multiple Access Full-duplex Subscribe Stations

IEEE  Bandwidth request mechanisms unsolicited requests unicast polls broadcast/multicast polls, and piggybacking

Simulation Environment  The simulator is event-driven and was developed using C++

Performance Metrics  gross subframe utilization The ratio between the OFDM symbols utilized in a subframe for data transmission  Throughput the overall amount of net user data  transfer delay a packet arrives at the MAC connection buffer of the source node to the next protocol layer at the destination node  backlog gap difference between the BS’s estimate of the backlog of a connection  notification delay a new SDU is received by an SS and the time instant at which the BS receives a bandwidth request for this SDU

BS and SS Schedulers  Uplink Weighted Round Robin  Downlink Deficit Round Robin

Bandwidth Requests Management  When BE or nrtPS becomes busy contention-based bandwidth request  When SS has a busy connections piggybacking  rtPS static allocation of periodic unicast polls ex: video 33ms VoIP 20ms  nrtPS with unicast polls every 500ms

Simulation Name  N = S x C x W W : identical basic data sources C : connections per direction S : overall number of stations

Simulation Parameters  Repeat 20 times  Run was 1200s  Warmup period of 360s  95% confidence interval

Performance Evaluation  Throughput and Delay Analysis  Bandwidth Request Analysis  Evaluation of Multimedia Traffic

Average delay VS number of SSs Minimum traffic unit is 147Kb/s (6 Web) Offered load is N x 147Kb/s (Best Effort)

Throughput VS number of SSs DL : Management overhead UL : Contention Slot overhead

Offered Load Partitioning  The offered load N increase 10 to 90  6 WEB source (24.5 x 6 ~=147Kb/s)

Utilization VS offered load Contention Slot BW min = 7

Throughput VS offered load Physical preambles

Bandwidth Request Analysis  Nc : a broadcast poll  Np : piggybacked on PDUs

Number of bandwidth requests per uplink subframe VS offered load N c : contention req. N p : piggybacked req. N > 50, N c Almost negligible

Average delay VS BW min Capacity reserved for Contention bw-req

Throughput VS BW min

Evaluation of Multimedia Traffic  The minimum traffic unit is 71.5Kb/s  VoIP traffic has not reported in the paper

CDF of the delay in the conn, source and SS cases with 30/60/90 videoconference sources the SS case incurs more overheads due to the transmission of a higher number of physical preambles compared to the conn and source cases in the conn case, the BS might schedule an uplink grant to another connection j before the unicast poll to connection i is due

Notification delay VS offered load piggybacking/bandwidth stealing mechanisms for source and conn

95% of the delay VS offered load

Backlog error VS time with 160 videoconference sources

Conclusion  There is a trade-off between average delay and throughput  SSs are able to request uplink bandwidth to the BS efficiently using piggybacked bandwidth request  Finally, paper have shown that rtPS outperforms nrtPS in terms of delay