From a chronological order perspective, the Wi-Fi infrastructure industry has progressed through several distinct phases, and is now entering a new phase. The order of progression looks like this, ending in where the industry is now headed:
- Autonomous - standalone APs, managed individually
- Autonomous - standalone APs, managed with a Wireless Network Management System (WNMS)
- Controller-based - dependent APs, co-ordinated and managed by a single controller
- Controller-based - dependent APs, co-ordinated by one or more controllers, which are managed by WNMS
- Controller-based - dependent APs, co-ordinated by a group of controllers, managed by WNMS, that are adapted to help transition the WLan to high-throughput, mission-critical support
- Controller-less - intelligent APs, managed by WNMS, that use inter-AP protocols to co-ordinate all control tasks for high throughput, deployment flexibility, and mission-critical support
It no longer makes one look like a visionary to publicly announce that WLan infrastructure will ultimately have fully-distributed intelligence. In fact, many suppliers have already shown enough of their cards that it's quite simple to see where their roadmap leads. The internet has taught us that a distributed intelligence model is scalable, high-performance, and resilient - something that mission-critical 802.11n Wi-Fi networks need. This is no big secret among Wi-Fi infrastructure suppliers.
For controller-based firms, the march toward a decentralised architecture began with a return to distributed forwarding (from the days of Autonomous APs), continuing into branch-office solutions, but it's in the branch office where they realised significant technical and financial challenges. These challenges immediately sparked additional development to push more features into the AP itself, but this too presented a roadblock in that controller-based APs were not designed with the processing power to replace a controller. This meant re-architecting AP hardware and WLan client handling, which is currently in progress across much of the industry and is introducing considerable developmental growing pains for controller-based suppliers.
Distributed intelligence may eventually take a variety of forms, such as controller-less or integrated edge switch/controllers, but in the end, manufacturers will push WLan intelligence to the network edge because it's the most scalable, reliable, and cost-effective way to design a WLan infrastructure.
Controller-less architecture, in particular, allows for linear and unlimited scalability, eliminating the stair-step model where additional or larger controllers must be introduced. A network design can start as small as a single AP, and linearly scale with one step: adding more APs. Moore's law has made it possible to introduce powerful AP hardware at low cost, enabling the removal of the controller hardware. Now, the control plane can exist directly between APs, as protocols, which are both resilient and free. Elimination of the controller removes a bottleneck and a single point of failure while increasing security and decreasing costs.
In addition to the scalability, reliability, security, and cost factors, a distributed-intelligence architecture introduces some additional key advantages. Intelligent APs can discover and self-configure for whatever role they need to play within the network without manual intervention, and failover/failback, best-path forwarding, and dynamic mesh routing features allow for a highly-resilient infrastructure. The speed of the Ethernet infrastructure can be leveraged through distributed forwarding, without feature limitations, since all of the intelligence of today's most feature-rich controller resides within each AP.
So is the future of enterprise Wi-Fi on the edge? I think the industry, and more importantly the market, believes that it is.
Devin Akin is chief Wi-Fi architect at Aerohive Networks
This was first published in May 2010