150 likes | 155 Views
This study aims to clarify the usefulness and applications of CC/RR in specific situations where it is the most effective mechanism. Simulation results demonstrate that CC/RR is particularly useful for servicing bursty traffic, initiation of signaling, and delay-sensitive traffic. Alternatives such as round-robin polling and EDCF have their limitations and cannot fully replace CC/RR. The recommendation is to keep CC/RR due to its better performance in certain cases.
E N D
CC/RR Performance Evaluation Yonghe Liu, Jin-meng Ho, Matthew B. Shoemake, Jie Liang Texas Instruments Y. Liu , et al
Objective • To clarify and illuminate the usefulness of and applications for CC/RR • CC/RR is not a panacea, but there are specific and important situations in which CC/RR is the most effective mechanism • CC/RR is complementary to HCF in its ability to enhance controlled contention functionality • To show simulation results that show specific and valid situations where CC/RR is the most effect mechanism to use Y. Liu , et al
Applications of CC/RR • CC/RR is useful for servicing bursty traffic: • Initiation of signaling • E.g, voice call setup • It is unacceptable to have long delays in setting up voice calls or even worse dropped signaling packets due to heavy network load • On/Off delay sensitive traffic • E.g., telnet, chat, web browsing • These types traffic are very bursty yet require low latency • CC/RR provides an efficient mechanism to determine which stations have traffic to send on a heavily loaded network without large overhead • Better control • Through CC sending frequency and CCI length Y. Liu , et al
Possible Alternatives to CC/RR • In the cases where CC/RR is most effective, alternative mechanisms have problems • Round robin polling • In the case of a heavily loaded network with bursty traffic not all devices have data to transmit • Likewise there may be many stations, e.g. VoIP clients, that have no data to transmit, but require a guaranteed low latency response when they do have traffic • In either case, round robin polling becomes inefficient due to the fact that many clients on the network do not have traffic • Likewise, when the number of clients with traffic is small percentage, round robin polling yields large response times than does CC/RR • EDCF • In servicing delay sensitive traffic, EDCF has little control • Making EDCF traffic more aggressive to minimize delay results in the risk of dropped packets Y. Liu , et al
EDCF Replacing CC/RR? • EDCF • Request frames sent using EDCF • Piggyback requests with data • Possible • High latency for requests • Simply cannot tolerate this! • Large jitter • Dropped requests Y. Liu , et al
Simulation Setup • Two downlink streaming flows • Contention Free • 2.4Mbps each • 1500 Bytes packets • Exp. inter-arrival time • 14 STAs using EDCF • CWmin = 31, AIFS=DIFS • Exp. ON/Off model • 1.2 Mbps each • 1 STA sending TXOP requests • Simulations compare latency of TXOP requests using: • EDCF with various CWmin/CWmax • CC/RR Y. Liu , et al
Simulation Result: EDCF • Request sending every 2s • Using EDCF • CWmin = 15, CWmax=1023, AIFS=DIFS Large latency and jitter Request dropped after retry limits (7) Delay (seconds) Y. Liu , et al
Simulation Result: EDCF • Request sending every 2s • Using EDCF • CWmin = 7, CWmax=1023, AIFS=DIFS Slightly lower latency but jitter still high Still have dropped requests after retry limits (7) Delay (seconds) Y. Liu , et al
Simulation Result: EDCF • Request sending every 2s • Using EDCF • CWmin = 3, CWmax=1023, AIFS=DIFS Latency no better and jitter still high Still have dropped requests after retry limits (7) Delay (seconds) Y. Liu , et al
Simulation Result: EDCF • Request sending every 2s • Using EDCF • CWmin = 15, CWmax=63, AIFS=DIFS Latency marginally better and jitter still high Still have dropped requests after retry limits (7) Delay (seconds) Y. Liu , et al
Simulation Result: EDCF • Request sending every 2s • Using EDCF • CWmin = 7, CWmax=63, AIFS=DIFS Latency marginally better and jitter still high Still have dropped requests after retry limits (7) Delay (seconds) Y. Liu , et al
Simulation Result: EDCF • Request sending every 2s • Using EDCF • CWmin = 3, CWmax=63, AIFS=DIFS Latency marginally better and jitter still high Still have dropped requests after retry limits (7) Delay (seconds) Y. Liu , et al
Simulation Result: CC/RR • Request sending • Using CC/RR, CC Priority<Streaming • Every 20ms starting at 0 High delay due to large CF burst, EDCF suffers as well Delay (seconds) Y. Liu , et al
Simulation Result: CC/RR • Request sending • Using CC/RR, CC Priority>Streaming • Every 20ms starting at 0 0~20ms Offset possible depending on periodic CC’s start time Delay (seconds) Y. Liu , et al
Conclusion • CC/RR an effective method for collecting requests for delay sensitive • Signaling • Bursty traffic • EDCF not capable of replacing CC/RR • Large latency and jitter • Dropped packets • AP has no control over delay of RR request • HCF with Round Robin Polling • Inefficient in case where CC/RR is effective Recommendation Keep CC/RR, because there are important cases where it provides better performance than any other mechanism Y. Liu , et al