Usefulness of Rate-Map Approach

The rate-map approach (section 4.1.2) is the foundation of the association and load balancing algorithms. This approach is based on the hypothesis that the signal strength of a client's probe request packets, as observed by a DAP (i.e. uplink packets), is a good approximation of the transmission rate a client can expect in both the up link and downlink directions in a dense DAP deployment. Higher transmission rates, generally imply higher throughput between the corresponding DAP and the client. Hence, the objective of the rate-map approach is to pick a DAP such that a client will get good throughput in both directions.

To validate our hypothesis we demonstrate a positive correlation between the RSSI of the probe request packets from the client to both uplink and downlink throughput via the following experiment. We set up a client laptop at a fixed location. The client attempts to associate with each of the 24 DAPs in turn. Prior to each association attempt with a DAP, we measure the signal strength of the client's probe request packets as observed at that node. This is the uplink signal strength (USS). After associating with a DAP, the client contacts a server on the wired network, and carries out a 2 minute TCP download, followed by a 2 minute TCP upload. We carry out this experiment from 6 different locations, and repeated the entire process 5 times. The experiment was performed on channel 64 of 802.11a band.

We found a correlation of 0.71 between USS and upload throughput and 0.61 between USS and the download throughput. These results indicate that USS can be indeed be used as a good predictor of upload and download throughput. This correlation is much stronger if we look at the throughput numbers against bucketized USS values. The rate-map approach bucketizes USS values and assigns a rate to each bucket (section 4.1.2). Figure 7 illustrates the strong correlation between throughput numbers and these bucketized USS values. The error bars indicate one standard deviation. It can be seen that the bucketized USS values are a good predictor of both upload and download throughput.

We have conducted these measurements over a number of clients and consistently found positive correlations thereby validating the hypothesis that USS values of probe packets from clients, can be used as good proxies for transmission rates between a DAP and the client.

Note that we have demonstrated a correlation between USS and uplink and downlink throughput. Detailed results that demonstrate a correlation between USS and transmission rates are available in [22].

Figure 7: Correlation between bucketized uplink RSSI and upload/download throughput