The project has yielded the first passive Access Point side measurement for monitoring achievable TCP throughput over wireless links for any client in the network. Virtual speed test can passively estimate the outcome of a speed test for any of the associated STAs (stations) in the network. While doing so, virtual speed test does not require any active measurements such as probing, no client-side co-operation or modification and no additional hardware infrastructure to collect more measurements. Virtual speed test can make an estimate solely based on passive AP-side observations.
The objective of our project is to develop models and frameworks that can enable the estimation of download and upload speeds of associated STAs (stations) in the network. While doing so, the framework cannot perform any active measurements, cannot seek any STA side co-operation and does not have access to any additional information except what can be obtained from the logs of the AP.
TCP speed tests are end-to-end tests of network health and are available via a plethora of online apps. As part of the measurement process, a client performs an active TCP download and an active TCP upload to a server to measure the download and upload TCP throughput respectively. Since more than 80% of current Internet traffic is transmitted via TCP, the performance of numerous online applications is crucially dependent on the maximum TCP throughput achievable over an underlying network path.
If a client’s speed test uses a nearby server (i.e., a server with minimum possible latency to the AP), the WLAN becomes the key part of the end-to-end path. Consequently, the results would be valuable to the network manager to assess WLAN performance and make decisions on network infrastructure alterations to improve the quality of service experienced by the end user. However, the results can only be seen by the end user and are unavailable to the administrator without seeking end user co-operation. Moreover, regularly performing such speed tests imposes additional traffic load on the network and hence doing so can potentially disrupt user traffic and drain the battery of mobile devices.
We have designed a framework called virtual speed test which enables an AP side estimation of the download and upload speeds that any of its associated STAs would achieve upon running a speed test. Virtual speed test employs a novel measurement-based model which we call L2 Edge TCP model which characterizes the speed test download and upload throughput in terms of parameters that can be directly observed at the AP. We have validated virtual speed test using extensive experiments performed on commodity hardware. Overall, virtual speed test demonstrates throughput estimation error under 5% yet with outcomes available at the AP.
We evaluated virtual speed test via extensive tests performed on an off-the-shelf Wi-Fi chipset. Specifically, we deployed our commodity hardware based APs in two environments – a University building and a residential apartment and conducted extensive tests spanning multiple days. Across all our experiments, virtual speed test has demonstrated a mean estimation error under 5% with outcomes available at the AP.