A New MAC Scheme for Very HighSpeed WLANs

1
A New MAC Scheme for Very High-Speed WLANs

• Tianji Li, Qiang Ni, David Malone, Douglas Leith,
  
• Yang Xiao and Thierry Turletti

Presenter Tianji Li Hamilton Institute, National
University of Ireland June 28, 2006
2
Outline

• Goal
• To design a new MAC with high efficiency for very
  high-speed next-generation WLAN (e.g. 802.11n)
• Difficulty
• Overhead at MAC and PHY
• Solution
• aggregation at MAC

3
Goal

• Now
• 802.11b PHY rate 11Mbps, MAC throughput
  7011 7 Mbps
• 802.11a PHY rate 54Mbps, MAC throughput
  5054 27 Mbps
• Future
• PHY rate gt 216 Mbps (up to 648 Mbps),
• MAC throughput ???

4
DCF The Current MAC

• Overhead DIFS, backoff, SIFS, PHY headers, and
  ACKs.

5
What if using DCF in Very High-Speed ?
MAC throughput lt 50 Mbps for ever !
6
Why DCF so Slow?

• Tframe frame size / R, it scales with 1/R.
• Tack ack size / R, it scales with 1/R.
• But, other items in denominator are constant,
  which leads to
• Solution We need to make all in denominator
  scale also with 1/R.

7
Prior Work (1/2)

• Burst ACK proposed in early versions of 802.11e
• Tdifs and Tbackoff scale with 1/R.
• Block ACK in the current 802.11e
• Tdifs , Tbackoff and TACK scale with 1/R.

8
Prior Work (2/2)

• Aggregation from Ji et. al.
• Tdifs , Tbackoff , Tack and Tsifs scale with 1/R.
• Aggregation from Kim et. al.
• All in denominator scale with 1/R, then why I am
  here

9
What are still missing?

• How to have very large frames?
• Wait or not if no enough information?
• How much time to wait for?
• Is there a limit for the frame size? What is the
  best size?
• What is the best size for retransmission?
• What the delay will look like?

10
Our Sample Scheme AFR
The Aggregation with Fragment Retransmission
(AFR)
11
Zero-waiting

• Question
• how much time should we wait for enough
  information to aggregate?
• Answer
• Zero-waiting transmit immediately
• Why
• In heavily loaded networks, aggregation happens
  automatically
• In slightly loaded networks, AFR degenerates to
  the legacy DCF
• Zero-waiting is proven to be stable where
  feasible

12
Maximum Frame Size
Constant throughput is possible with increasing
frame sizes Maximum frame size 65536 bytes
13
Fragment sizes (1/2)
Fragmentation is necessary with large frame in
bad channels
14
Fragment sizes (2/2)
A single fragment size can be found for
near-optimal efficiency
15
MAC Delay
CSMA/CA delay for a frame is worse than in DCF
16
MAC Queue Delay
Total delay is much better due to pipeline-like
ability
17
AFR vs DCF
18
HDTV (simulation)
19
Thanks!