97 | | The 802.11 DCF employs a binary exponential backoff (BEB) |
| 97 | The 802.11 DCF employs a [http://en.wikipedia.org/wiki/Exponential_backoff binary exponential backoff (BEB)]. Roughly speaking, each packet transmission is separated by a random wait time that may be paused by other transmissions from other nodes. This random interval is chosen uniformly over a contention window (CW). The CW itself is dynamic. It increases when nodes do not receive an acknowledgement that their transmission succeeded (an indication that the medium is under heavy contention). This makes the node less aggressive about transmission attempts and serves to reduce the likelihood of collisions. Conversely, the CW resets to its minimum value when nodes successfully transmit and receive an acknowledgement. For 802.11g, the minimum contention window over which a backoff is drawn is [0, 15]. After failures, this window doubles to [0, 31], [0, 63], and so on. The 802.11 standard caps the maximum CW to [0,1023]. |
| 98 | |
| 99 | ==== Reducing the CW of the AP ==== |
| 100 | |
| 101 | One way that we can modify the performance of the AP is to let it play by a different set of rules than the two connected stations. In the 802.11 Reference Design, the CW selection is purely in C-code. By making a simple change to this C-code, we have modified the AP such that it's minimum CW is [0, 7] instead of the standard [0, 15]. After a failure, its next CW is [0, 15] instead of [0, 31] and so on. We expect that this change will tend to make the AP win access to the medium in front of the two stations on average since it is usually drawing from a set of smaller values. |
| 102 | |
| 103 | Re-running the experiments with this change yields the following. |
| 104 | |
| 105 | || || Isolated Throughput (Mbps) || Simultaneous Throughput (Mbps) || |
| 106 | || Flow 1 || 14.84 || 3.7 || |
| 107 | || Flow 2 || 14.83 || 3.7 || |
| 108 | || Flow 3 || 14.07 || 2.72 || |
| 109 | || Flow 4 || 14.08 || 2.71 || |
| 110 | || Sum Throughput || n/a || 12.83 || |
| 111 | |
| 112 | When the flows are run separately, we can see that Flow 1 and Flow 2 achieve slightly higher throughputs than they used to because of the CW change. When each flow is run simultaneously, we can see that their relative share of the sum throughput is a very different story than what it used to be. |
| 113 | |
| 114 | || [[Image(wiki:802.11/wlan_exp/app_notes/dcf_with_multiple_flows/figs:modified_cw_simultaneous_xput.png, width=300)]] || |
| 115 | || ''Relative Share of Network Sum Throughput''' || |
| 116 | |
| 117 | Whereas Flow 1 and Flow 2 used to split 1/3 access to the medium, they now get a much larger share. The AP gets to be much more aggressive about transmitting with the change to the contention window behavior, thus compensating for its extra burden of needing to source two traffic flows. |
| 118 | |