Changes between Version 25 and Version 26 of 802.11/wlan_exp/app_notes/dcf_with_multiple_flows
- Timestamp:
- Apr 17, 2014, 10:52:24 AM (10 years ago)
Legend:
- Unmodified
- Added
- Removed
- Modified
-
802.11/wlan_exp/app_notes/dcf_with_multiple_flows
v25 v26 78 78 79 79 || [[Image(wiki:802.11/wlan_exp/app_notes/dcf_with_multiple_flows/figs:connected_symmetric.png, width=600)]] || 80 || '' Throughput Timeline''' ||80 || '''Throughput Timeline''' || 81 81 82 82 The above figure shows the timeline of throughputs achieved by each of the four flows during the experiment with a one second rolling window average. These timelines were constructed from the event log using a technique identical to that presented in the [wiki:../../examples/txrx_log_analysis Log Analysis Example]. Furthermore the script used for this experiment is provided in the [#Resources Resources Section]. … … 120 120 == Experiment 2: Symmetric with Hidden Stations == 121 121 122 In this second experiment, we modify the attenuation values in the experimental setup such that each flow sees the same path loss as the first experiment, but the path loss between the two STA nodes is dramatically larger. Topologically speaking, this is an extreme hidden node problem:STA 1 and STA 2 can each "see" the AP, but they cannot see one another. Note: this experiment resets the CW changes we made to the AP at the end of the prior experiment. Each node is running a completely standard 802.11 Reference Design.122 In this second experiment, we modify the attenuation values in the experimental setup such that each flow sees the same path loss as the first experiment, but the path loss between the two STA nodes is dramatically larger. Topologically speaking, this is an extreme [http://en.wikipedia.org/wiki/Hidden_node_problem hidden node problem:] STA 1 and STA 2 can each "see" the AP, but they cannot see one another. Note: this experiment resets the CW changes we made to the AP at the end of the prior experiment. Each node is running a completely standard 802.11 Reference Design. 123 123 124 124 ==== Experiment Details ==== … … 163 163 164 164 || [[Image(wiki:802.11/wlan_exp/app_notes/dcf_with_multiple_flows/figs:hidden.png, width=600)]] || 165 || '' Throughput Timeline''' ||165 || '''Throughput Timeline''' || 166 166 167 167 Notice that Flows 3 and 4 are very unstable. They do not simply oscillate around some nominal throughput like they did in the first experiment. It appears that they alternate between nearly turning off with near-zero throughput and achieving very high instantaneous throughput. How can we explain this? 168 168 169 The event log of the WLAN Experimental Framework is very powerful. It can reveal much more than throughput timelines. Another set of data it can give us is what the contention windows of each node in the network as a function of time. In this way, we can visualize the binary exponential backoff. 169 The event log of the WLAN Experimental Framework is very powerful. It can reveal much more than throughput timelines. Another set of data it can give us is what the contention windows of each node in the network as a function of time. In this way, we can visualize the binary exponential backoff of each node in the networks 170 171 || [[Image(wiki:802.11/wlan_exp/app_notes/dcf_with_multiple_flows/figs:connected_symmetric_cw.png, width=600)]] || [[Image(wiki:802.11/wlan_exp/app_notes/dcf_with_multiple_flows/figs:hidden_cw.png, width=600)]] || 172 || '''Experiment 1 (Fully Connected) Contention Window''' || '''Experiment 2 (Hidden Stations) Contention Window''' || 173 174 Comparing the contention data from the two experiments reveals what is going on with the throughput behavior in the hidden node case. Whereas the first experiment saw each node's contention window hovering around similar values, the second experiment's hidden station nodes have radically different behavior. The contention window wildly swings and rails between maximum and minimum values many times. Notice that when STA 1 has a large CW, STA 2 has a small CW and vice versa. The gap of idle time created by one station choosing a large contention window allows the other station many opportunities to successfully transmit and minimize its own contention window. These contention window selections explain why the throughputs of Flows 3 and 4 vary between high and low values with such frequency. 175 176 177 == Conclusion == 178 179 In this application note, we used the WLAN Experimental Framework to study behaviors of the DCF. We used the framework's event logs to dive deeply into this behaviors and describe transient effects -- we are not limited to coarse aggregate measures of performance. An extension to this experiment would be to remove the RF cabling altogether and investigate over-the-air performance. In this scenario, there is much more to deal with. For example, there will be interference from other ISM band transmitters and there will also be multipath fading. The WLAN Experimental Framework can help decipher the guiding forces behind the performance that is observed. For example, the [wiki:../../examples/chan_est_viewer Channel Estimate Viewer] shows how to extract channel 170 180 171 181