September 2002 Mathilde Benveniste, Avaya Labs Slide 1 doc.:IEEE 802.11- 02/525Ar0 Submiss ion Simplifying Polling Simplifying Polling Mathilde Benveniste benveniste @ieee.org Dongyan Chen Avaya Labs Research
Apr 01, 2015
September 2002
Mathilde Benveniste, Avaya LabsSlide 1
doc.:IEEE 802.11-02/525Ar0
Submission
Simplifying PollingSimplifying Polling
Mathilde [email protected]
Dongyan ChenAvaya Labs Research
September 2002
Mathilde Benveniste, Avaya LabsSlide 2
doc.:IEEE 802.11-02/525Ar0
Submission
BackgroundBackground
• Avaya is interested in QoS (e.g. VoIP) and channel throughput in the enterprise space
• Minimal delay for time-sensitive applications and efficient channel use are the focus -- regardless of data rate
September 2002
Mathilde Benveniste, Avaya LabsSlide 3
doc.:IEEE 802.11-02/525Ar0
Submission
MotivationMotivation
• RRs serve QoS while promoting efficient channel use– RRs keep polling list short – limited to active
stations– RRs permit the polling mechanism to work
for burstry traffic as well as periodic traffic
• Simple implementation is attractive in an RR mechanism
September 2002
Mathilde Benveniste, Avaya LabsSlide 4
doc.:IEEE 802.11-02/525Ar0
Submission
TerminologyTerminology
• CC/RR the reservation mechanism for polling, as it appears in D3.0
• EDCF/RR alternative reservation mechanism for polling – no CCI; RRs comprise top priority EDCF class– the RR could be the QoS-Null frame
September 2002
Mathilde Benveniste, Avaya LabsSlide 5
doc.:IEEE 802.11-02/525Ar0
Submission
Simulations in Simulations in 01/571r0 01/571r0
• In 01/571r0 the results showed a better performance of CC/RR compared to ‘straight’ PCF– Straight PCF polls all associated stations
• These results were not sufficient to validate the CC/RR mechanism– The results simply showed that a managed polling list improves
performance
September 2002
Mathilde Benveniste, Avaya LabsSlide 6
doc.:IEEE 802.11-02/525Ar0
Submission
Simulations in Simulations in 02/305r0 02/305r0 • 02/305r0 results showed better performance with
CC/RR compared to EDCF/RR while using a short CP– CFP uses up to 18 ms in a 20 ms cycle– Only 2 ms provided for CP; a fairer allocation of channel time would
have averted the high collision rate affecting the RRs
• 02/305r0 results showed long access delays with EDCF/RR caused by DCF contention despite use of PF differentiation– A Persistence Factor was included, but not set at lowest value; a
better choice of EDCF parameters would have protected the RRs
• The results in 02/305r0 do not establish the superior performance of CC/RR over EDCF/RR– CC/RR is compared to a sub-optimal EDCF/RR specification
September 2002
Mathilde Benveniste, Avaya LabsSlide 7
doc.:IEEE 802.11-02/525Ar0
Submission
Our objectiveOur objective
• To present an efficient specification of EDCF/RR
• To show simulation results comparing EDCF/RR to CC/RR
September 2002
Mathilde Benveniste, Avaya LabsSlide 8
doc.:IEEE 802.11-02/525Ar0
Submission
Simulation ScenariosSimulation ScenariosUplink and downlink transmission through HCF in CP The performance of the two reservation mechanisms
– CC/RR: CCI period=5 ms; CCOPS=50 – EDCF/RR: PF=0.5 [01/409r2]; AIFS=PIFS; CWmin=8; Cwmax=2
is compared under two scenarios• Under low load (LL)
The load consists of time-sensitive bursty traffic (Load1, p 8)
• With heavy DCF load (HL)A best effort load of 5 stations, 1 Mpbs per station, is added
(Load2, p 8)
The same HCF implementation used on the downlink for both scenarios
September 2002
Mathilde Benveniste, Avaya LabsSlide 9
doc.:IEEE 802.11-02/525Ar0
Submission
Traffic ScenariosTraffic Scenarios
Two loads considered on 11Mbps DS channel Load1 is transmitted by polling; it generates frequent RRs
– 8 stations engaged in two-way exchange with AP of time-sensitive bursty traffic
– Payload [excludes MAC & PHY overhead] per packet =120 bytes; inter-arrival fixed at 10 ms when ON; exponential 342 ms ON /650 ms OFF (example: silence suppressed voice)
Load2 contends for the channel through DCF– 5 stations sending low priority data uplink– Payload per packet=724 bytes; inter-arrival exponential
at 7 ms
September 2002
Mathilde Benveniste, Avaya LabsSlide 10
doc.:IEEE 802.11-02/525Ar0
Submission
Statistics ReportedStatistics Reported
• RR ETE (end-to-end) delay• Load1 uplink ETE delay –
– Simple poll scheduling– Optimized poll scheduling
• Load1 downlink ETE delay• Load2 load ETE delay
September 2002
Mathilde Benveniste, Avaya LabsSlide 11
doc.:IEEE 802.11-02/525Ar0
Submission
RR end-to-end delay (sec)RR end-to-end delay (sec)
CC/RR-LL under Light Load
EDCF/RR-LL
CC/RR-HL under Heavy LoadEDCF/RR-HL
EDCF/RR gets the RRs out sooner
September 2002
Mathilde Benveniste, Avaya LabsSlide 12
doc.:IEEE 802.11-02/525Ar0
Submission
Load1 uplink end-to-end delay (sec)Load1 uplink end-to-end delay (sec)
CC/RR-LL under Light Load
EDCF/RR-LL
CC/RR-HL under Heavy LoadEDCF/RR-HL
Without optimization
EDCF/RR leads to lower uplink delay
September 2002
Mathilde Benveniste, Avaya LabsSlide 13
doc.:IEEE 802.11-02/525Ar0
Submission
Load1 uplink end-to-end delay (sec)Load1 uplink end-to-end delay (sec)
CC/RR-LL under Light Load
EDCF/RR-LL
CC/RR-HL under Heavy LoadEDCF/RR-HL
With optimization
Even with optimization EDCF/RR does at least as well
September 2002
Mathilde Benveniste, Avaya LabsSlide 14
doc.:IEEE 802.11-02/525Ar0
Submission
Load1 downlink end-to-end delay (sec)Load1 downlink end-to-end delay (sec)
CC/RR-LL under Light Load
EDCF/RR-LL
CC/RR-HL under Heavy LoadEDCF/RR-HL
Downlink performance is the same for the two mechanisms
September 2002
Mathilde Benveniste, Avaya LabsSlide 15
doc.:IEEE 802.11-02/525Ar0
Submission
Load2 end-to-end delay (sec)Load2 end-to-end delay (sec)
CC/RR-HL
EDCF/RR-HL
under Heavy Load
•CC/RR dedicates channel time (20% in this example) to CCI that could be otherwise used for transmission
•EDCF/RR uses channel more efficiently
September 2002
Mathilde Benveniste, Avaya LabsSlide 16
doc.:IEEE 802.11-02/525Ar0
Submission
ConclusionsConclusions
• EDCF/RR gets the RRs out sooner• EDCF/RR uses channel more efficiently• EDCF/RR performs at least as well as
CC/RR under all situations– Light load– Heavy load– Optimized and non-optimized poll
scheduling
• EDCF/RR is simpler to implement
September 2002
Mathilde Benveniste, Avaya LabsSlide 17
doc.:IEEE 802.11-02/525Ar0
Submission
MotionMotion
• Remove CCI from D3.0• Devote top priority EDCF class to RRs
exclusively • Re-introduce the persistence factors
and use lowest PF value for RRs only