Historically, competitive testing has tended to focus on aggregate data throughput using file transfer as the means of establishing load on APs. The most recent public stress testing that included video was published in 2013.
Video is distinguished from most data applications, for example, email, file transfer, browsing, by its effect on end-user quality of experience. Whereas users are unlikely to notice or care if it takes a few extra seconds to download an email attachment, users immediately notice a stalled video. The likelihood that a user will experience poor video quality (stalling) increases when that user is in a high client density environment which, in this test suite, is defined as sixty (60) clients.
Video dominates network data traffic volume in many of today’s networks so it was chosen in a test suite as the primary data load.
All tests were run in the 5 GHz band, which is the industry-recommended best practice for high density environments. All clients were connected to the WLAN via a single SSID, secured with a PSK, using a 40 MHz-wide channel. Although 802.11ac supports higher data rates when using 80 MHz-wide channels, such large channel widths are not recommended for use in high-density environments due to channel contention and poor channel reuse.
All clients were given one minute to run and be counted for stalls before a data load was introduced. Since video stalls can be fleeting, a conservative method was used to define video stalling. In order to be considered stalled, a video must either not have started or be in a stalled state at the conclusion of each test phase.
Once the video clients were running, non-video data traffic was added to the WLAN for one minute by configuring Mac Mini clients as Ixia Chariot 7.3 EA end points (1 pair each). Enough network load was introduced to trigger competition for available bandwidth between the different classes of traffic (video and data). To allow for precise control and to create consistent loading, data traffic in the form of UDP was chosen.
In cases for which video didn’t begin immediately, it was retried twice. If it still failed to start it was considered stalled and counted against the initial tally (supported video clients without network load) and the number of stalled clients during network load (assuming it was still stalled).
At the end of the one-minute data-loading run time, the number of stalled video clients was counted again using the same stall criteria. The final aggregate throughput of the data clients (Mac Minis) was the value reported by Chariot.
Only one AP (Ruckus R610) was able to deliver stall-free video to all 60 clients both with and without a data network load.