CERN Accelerating science

Jump to related content

Following the success of CINBAD (CERN Investigation of Network Behaviour and Anomaly Detection), a project launched in 2007 in collaboration with HP which ended in 2010, HP and CERN started on a new joint research project in 2010, under the codename WIND (Wireless Infrastructure Network Deployment). WIND focuses on Wi-Fi networks in large campus environments.

Wi-Fi challenges

In a perfect world, the Wi-Fi networks would be as efficient as the wired ones. One could deploy such networks without arduous planning. The need for laborious site surveys would be significantly reduced. Once deployed, the lightweight control logic would adapt to the environment’s changes in real-time to squeeze the best possible performance out of the given setup. The network would react to unpredictable congregations of users, radio interference, and other events. Furthermore, the network itself would be able to suggest deployment changes that could improve the performance, reliability, and overall user experience.

Reality is regrettably much harsher and Wi-Fi networks will probably never be as efficient as the wired ones. Indeed, painstaking planning and simulations precede the typical installation phase. Site surveys consume time and money and do not result in a perfect network as they provide only a static snapshot of the environment.Reinforced fire doors, microwave ovens, Bluetooth users to name but a few are very likely to cause interferences and, nevertheless, are generally not taken into account in site surveys.

As many network operators neglect the task of monitoring and tuning network parameters to adapt to new requirements and changes, a large number of deployments suffer from performance problems. Network operators cannot easily tell about the current health of the wireless network from the end-user point of view. No clear procedures help pinpointing potential problems or even find the root cause of existing ones. In many cases, the diagnosis and remedies are part of a random trial-and-error process. Last but not least, the end users do not understand and do not want to know about the intricacies of the radio wave propagation, wireless deployment, and operation. For all these reasons, the objective of providing wire-like quality proves to be extremely challenging.

CERN network

Today, CERN provides Wi-Fi service to more than 2 600 concurrent wireless users. This service is available in various types of environments, from big auditoria, to offices, warehouses, crowded control centres, and underground caverns. The installation currently consists of over more than 700 dual channel access points.

Challenges are foreseen as the current installation is expected to grow in the near future: in terms of coverage as the goal is to have most of the CERN campus covered, but more importantly also in terms of complexity. For instance, over the last few years, the number of wireless devices literally exploded. It is no longer uncommon for a single user to have several Wi-Fi clients such as a laptop and a smartphone, for instance. The number of devices alone is not a serious challenge but the way the devices are being used is definitely a critical one. Indeed, users expect Wi-Fi networks to be natural extensions of wired networks and thus demand similar performance (e.g. sufficient to stream HD video) and reliability (e.g. no disconnections, no roaming problems).

Reliability is taken for granted for wired networks and when performance is at risk, simple solutions such as adding more cables can solve the issue. Solving issues with wireless networks is far more complex. Indeed, the number of available wireless channels is limited, multiple interferers are present, and the radio propagation parameters are by no means constant. These are just a few examples of the limitations Wi-Fi network users are facing. More information on this topic is available on the CERN openlab website, in the presentations sub section, and in the Networking Competence Centre section.

WIND project and first results

Given the variety of its premises and the number of Wi-Fi users on the large CERN campus, the Organization is keen on collaborating in a research activity focusing on wireless networks and provides the perfect environment to carry it out. The WIND project attempts to offer new algorithms, guidelines, and solutions for large-scale installations. In particular, it aims at monitoring, optimizing, and troubleshooting the existing installations as well as providing feedback for Wi-Fi design. Results of the joint research could possibly be incorporated into HP hardware and software to provide even more robust and efficient networking solutions.

In the first phase of the project and its first year, the WIND team investigated the state-of-the-art in the area of wireless networking. The results of this study demonstrated that the research on the operation and optimization of large-scale wireless networks is still in its infancy and that current standards have so far largely overlooked the issues related to monitoring and client control. The report called ‘Wireless Control and Optimisation’ was published at the end of 2010 and is available in the technical documents sub section of the CERN openlab website.

In parallel, the WIND project team worked on identifying the data sources that could provide information about the current state of the network. These data sources proved to be rather scarce, the only data sources available at the moment being the access points since the current standard does not allow for querying the clients for monitoring purposes. However, HP access points do an adequate job in providing detailed information about the associations, connected clients, and radio visibility by means of the SNMP (Simple Network Management Protocol).

Following this research, a dedicated system for collecting and storing the Wi-Fi related information was created. This scalable system combines the information from SNMP tables (e.g. client and visibility statistics) with the association events signalled by SNMP traps. Stored data is not only used for testing new algorithms but also for helping CERN’s network engineers in their daily operations. New features can easily be integrated to the collector. It is envisioned to add an sFlow input module as soon as the access points start providing the additional information.

Several tools using the collected data have already been created. One of them attempts to detect the clients that are most likely to cause network slowdowns by correlating the transmission time of each client with the client’s average transfer rate. Another tool tackles a different but equally important aspect of wireless networks by providing a visual representation of roaming events between different access points installed on the campus. This helps to quickly identify roaming «hotspots» or troublesome devices.

During the second year of the project, the WIND team will continue working on algorithms for analysing the state of Wi-Fi networks and for detecting anomalies such as slow devices and deployment problems. Another aim of the WIND project will be to identify the key performance indicators that can reliably characterize the state of the network from the end-users’ point of view. Finally, in the coming year, the WIND team will carry out some experiments on selected fragments of live networks to test chosen algorithms for optimizing performance (e.g. by forcing clients to associate with a given access point or on the 5 GHz band). The project promises interesting findings while the tools created up till now are already providing valuable information to CERN’s network engineers.