Airport Wi-Fi Challenges

In depth analysis of the traffic of a large commercial Wi-Fi network & accurate predictions of network state for several days ahead. 

 

The context


wifi-in-airportsSome international airports are among the liveliest and most varying wireless environments. The Wi-Fi network of such hotspots are challenged with the demand of a mixture of users with different requirements and habits. Business travelers, tourists, workers of ground handling facilities, office employees, deployed wireless sensors and fleets of special vehicles are only an indicative subset of the numerous users.

To make things more complex, airport wireless broadband traffic properties are characterized by high volatility; unexpected events as flight cancellations or delays as also planned high load situations make network demand from per area of an airport rapidly oscillate between different extreme states.

Let’s assume a typical departure control hall of an airport terminal with 10 X-ray machines serving passengers in parallel. At normal state people walk quickly from the entrance of the hall till the entrance of the X-ray machine, where they empty their pockets and place their hand-luggage on the conveyor belt. They pass through the control and then pick up their stuff and walk quickly towards the stores, restaurants, coffee shops and eventually the departure gate. At such a state only few users will generate traffic on the wireless broadband.

Airport AtLow

In brief, the traffic properties for this network state are summarized below:

  • Considerably low number of concurrent Wi-Fi users
  • High mobility
  • Low usage density

In this state, the overall control hall can be easily supported by 1 or 2 access points transmitting at maximum power in order to cover the maximum possible space. In this way the network topology will assure complete area coverage, robust network service for all users (moving or not), at the least necessary CAPEX and OPEX. An indicative network topology for this state is depicted in the figure below.

WiFi AtLow

Airport AtHigh

However things can change rapidly. Due to a weather problem or a small technical problem many more than originally planned passengers may end up queueing at the very same control hall. A hall that normally has short queues and quickly passing passengers is transformed to a very crowded place of people that spend several minutes standing and waiting for their turn and some seconds walking through the control. Obviously Wi-Fi traffic bursts. Not only because so many more people are there but because they wait standing, they need to inform their associates about the delay, utilize the wasted time to communicate or work or hear the news or amuse themselves ... The Wi-Fi traffic properties change dramatically to the following:

  • The number of users grows at 5x, 20x or even 100x
  • Very low mobility
  • Usage density at extreme levels

At this state the optimal network setup is quite different. The number of users requires a denser network with access points transmitting at low power so that to cover small areas in order to keep the number of users per access point at bearable levels. Of course such topology comes with a completely different distribution of channels (frequencies) to the access point. The optimal network topology now looks as the figure that follows:

WiFi AtHigh

Obviously this configuration, although able to serve the high load extreme is not recommended for the rest of the times, which will most probably be the dominant ones at a well performing airport. If we tune the network for the worst of the two states we shall end up ...

  • decreasing QoS of moving users that will end up constantly changing serving AP
  • overspending on energy and AP use

If we tune for the normal state we won’t be able to serve most users adequately at the high load extreme. Changes from one state to the other are not rare however. They occur within minutes of time, due to disturbances on flight flow or other sudden operational challenges.

This is an indicative case where knowledge and prediction can dramatically improve performance of already existing management mechanisms as load balancing, deep packet inspection etc.

A use case

In this use case, over 200 Wi-Fi Access Points (APs) are deployed to cover several buildings and floors and offer internet services to both paid (business) and free (regular) users. In specific times and places, the network faces high contention conditions which can lead to congestion and QoS degradation or interrupted service. This eventually leaves room for improvements in the network and performance and user satisfaction.

Incelligent integrates with the Wi-Fi equipment in order to gather real-time information on the Wi-Fi network and environment. This dataset is further enriched with information on arrivals/departures, location and user movement behavior within the airport.

By analyzing these data, the Incelligent management system first characterizes (clusters) APs with respect to their traffic profile and/or faulty behavior in time and space and second, it provides very accurate predictions of the expected behavior e.g. traffic load per cluster of APs.

Based on these predictions, Incelligent proactively identifies network contention before users are affected and acquires the ability to support intelligent management decisions, so as to preserve Quality of Experience.

Indicative results

After processing a week’s data Incelligent system is able to accurately predict all key metrics of each access point of the network for at least the next days. The figure that follows presents the comparison of predicted and actual metrics for an indicative access point at such an environment. As shown, the predictor mimics the real behavior very efficiently, enabling intelligent decisions as far as transmission power, bandwidth and frequency allocation are concerned. 

 Results1

 

This use case was based on a purely Cisco Wi-Fi – based network. The key benefits derived are in the areas of service quality improvements and network management automation / simplification.