nav search
Data Center Software Security Transformation DevOps Business Personal Tech Science Emergent Tech Bootnotes BOFH

Boffins explain why it takes your Wi-Fi so long to connect

And there you were thinking WiFI interoperability was done and dusted ... NOT!

By Richard Chirgwin, 24 Jan 2017

You'd think by now that Wi-Fi has been around so long, nobody could get it wrong, right?

Wrong, according to Changhua Pei of Tsinghua University and collaborators from Tencent and the Tsinghua National Laboratory for Information Science and Technology: Wi-Fi implementations vary so much that 45 per cent of attempted connections fail, even if all you see is a frustratingly slow connection.

That quite surprising number comes from a study of the kind of scale that happens so easily in China: five million users in four cities, connecting to seven million access points and starting off 400 million Wi-Fi sessions.

The researchers collected their data from the WiFi Manager Android and iOS app.

Of the 45 per cent of connection attempts that fails, they write in this Arxiv paper, one-third led to users waiting between five and ten seconds to connect. The most common connection stage that fails? – scanning for networks, which popped up 47 per cent of the time.

Wifi setup process

In the four-stage connection process shown above, the researchers say scanning fails quite simply – the access point doesn't respond to the probe from the device. The next most common failure point, association, fails because the access point doesn't send Association Response packets.

The researchers say connection “time cost” is influenced by, frankly, practically anything in the environment:

  • Device chipset – slower chipsets take longer to connect;
  • Operating system – different Android versions on the same device show different connection time cost;
  • Access point deployment – private access points respond better than public, and perhaps surprisingly, not because of the number of users. The paper notes it could be because different models are bought for different deployments, since some units only turned up in public networks, while others only appeared in private networks.

The cure they offer is a machine learning algorithm that helps users' devices select access points as slow or fast, rather than choosing the access point based on the strongest signal. This, they claim, cuts the total connection failure rate from 33 per cent to 3.6 per cent, and since unnoticed failures lie behind slow connections, users could see their devices connecting ten times as quickly. ®

The Register - Independent news and views for the tech community. Part of Situation Publishing