Dynamics of Contention Free Period
Reservation in IEEE 1901 Networks
Brad Zarikoff and David Malone
Hamilton Institute, NUI Maynooth and Latitude Technologies.
29 March 2012
Overview
• IEEE 1901, broadband, in-home.
• Options for contention-based and contention-free access.
• Contention part looks like WiFi.
• Contention free arises from previous work1 2
• Good for traffic with QoS requirements?
1
H. Hrasnica and R. Lehnert, Reservation Domains in MAC Protocols for
Broadband PLC Networks, ISPLC 2005.
2
Y.-J. Lin, H. A. Latchman, J. C. L. Liu, and R. Newman, Periodic
Contention-Free Multiple Access for Broadband Multimedia Powerline
Communication Networks, ISPLC 2005.
Contention Free Access
B
CFP
CAP
B
CFP
For each flow wanting to use CFP:
• Station must make request to BSS manager in CAP.
• BSS manager must update CFP schedule.
• Schedule is announced by BSS manager in beacons.
• Station begins use of CFP, until reservation is canceled.
Schedules are transmitted with a lifetime; to expire a schedule you
must wait for the lifetime CSCD (= M frames) and transmit a
preview of the new schedule.
Sources Reservation of Delay
• Contending for access (backoff, collisions, ACKs, etc.).
• Waiting for preview schedule to become current.
• Waiting for modification to current schedule.
Has to be repeated if reservation is canceled.
• BSS manager can cancel the reservation.
• The station can request the cancellation.
• There is an inactivity limit (Til ).
Setup
• Focus on reservation delay.
• Simulate with discrete event simulator.
• N stations making reservations.
• Run with beacon interval of 40ms for 80s (2000 intervals).
• Start with defaults of TCAP = 4ms and M = 3.
1
0.8
Signals
Success
Success Success
Failure
0.6
0.4
Back-off
Slots
0.2
0
0
0.5
1.0
1.5
2.0
2.5
Time (ms)
3.0
3.5
4.0
Saturated Traffic
1400
M =7
M =3
M =1
1200
E[d] (ms)
1000
800
600
400
320
200
160
80
0
0
10
20
30
N
40
50
How big should CAP be?
2.0
N = 50
N = 10
N =5
E[d] (s)
1.6
1.2
0.8
0.4
0
1
2
3
4
5
6
TCAP (ms)
7
8
9
10
How about voice?
• Saturated traffic won’t time out, delay is one off.
• Saturated traffic usually not too delay sensitive.
• How about voice?
• Simple model, 64kbps, on-off exponential mean 1.5 with talk
clamped below at 240ms3 .
• Note, delay budget on the order of a few frames.
3
A. P. Markopoulou, F. A. Tobagi, and M. J. Karam, Assessing the quality
of voice communications over internet backbones, IEEE/ACM ToN, vol. 11, no.
5, 2003.
Saturated vs. Voice with long timeout
1400
1400
M =7
M =3
M =1
1200
1000
E[d ] (s)
1000
E[d ] (s)
M =7
M =3
M =1
1200
800
600
800
600
400
320
400
320
200
160
80
0
200
160
80
0
0
10
20
30
TCAP (ms)
Saturated
40
50
0
10
20
30
40
50
TCAP (ms)
Voice
Really measuring setup. How about with more realistic timeouts?
How big should Til be?
216
18e3
N = 50
N =2
208
16e3
14e3
12e3
192
E[E ]
E[d] (ms)
200
184
176
10e3
8e3
6e3
168
4e3
160
2e3
160
320
Til (ms)
480
0
Delay
160
320
Til (ms)
Timeouts
M = 3, TCAP = 40 ms
480
Mixed Saturated and Voice
Saturated would usually live in contention period.
4.5
480
N = 50
N = 20
N=8
N=2
400
4.3
240
4.2
MOS
E[d ] (ms)
320
Nbg = 0
Nbg = 4
Nbg = 8
Nbg = 2
Nbg = 6
Nbg = 10
4.4
160
480
4.1
4
400
3.9
320
3.8
240
160
3.7
0
2
4
6
Nbg
Delay
8
10
0
10
20
30
40
50
N
E-Model MOS
MOS: 4.34 Very satisfied; 4.03 Satisfied; 3.60 Some users dissatisfied.
Conclusion
• Contention-free access looks useful.
• Reservation delays can be significant.
• Use small M if possible.
• Use long Til if possible.
• Careful use of prioritisation may help.
• Matching application may help.