30 January 2009

Wireless LAN Architecture

This year is the tenth anniversary of the 802.11 Wi-Fi standard. A true success story, 802.11 has completely transformed how we access data networks. What had been a market filled with islands of proprietary wireless implementations has morphed into a community of standards-based vendors.

The "micro-cell" architecture at the heart of 802.11 blankets the coverage area with radio signals operating at different channels, like a colorful patchwork quilt. As Wi-Fi clients - laptops, smart phones, PDAs, barcode scanners - roam, they maintain connection with the network by moving between adjacent cells. This architecture was modeled after the overwhelmingly successful micro cell architecture employed universally throughout the cellular industry.

Standards-based Wi-Fi clients have a well defined set of characteristics, allowing them to properly associate and interoperate with standards-based Wi-Fi infrastructure. Interoperability can be validated by independent, third-party testing organizations - such as the Wi-Fi Alliance - providing end users with peace of mind that products work as expected. Component vendors - such as Wi-Fi radio IC manufacturers - can produce a common set of part in high volume, driving down the cost to users.

That is not to say that all Wi-Fi networks are identical. There are very significant throughput, security, ease-of-use, scalability, and life-cycle cost differences between solutions from different vendors. Adaptive Wi-Fi solutions - ones that optimize performance automatically as the RF environment changes - offer significant advantages over less capable non-adaptive networks, for instance. However, even these different solutions all start from a standards-based Wi-Fi micro cell architecture. Client interoperability is a given.

Not so with the handful of vendors that have bucked the standards trend and opted to deliver proprietary Wi-Fi networks. The most egregious example is the so-called single channel architecture. Instead of making the best use of all available Wi-Fi channels, the single channel architecture operates all of the clients and infrastructure on a single channel. The argument is that doing so allows clients to be switched instantaneously by a central controller between radios covering adjacent areas as the client roams. This is stated to be especially important for voice communications, where a fast transfer improves the user experience, and cellular networks are held up as an existence proof of this assertion.

What goes unstated is that cellular networks migrated to a micro cell architecture because they could not otherwise scale network size. The coverage and fast switching "issues" cited by single channel vendors actually aren't issues at all - they've all been solved by adaptive micro cell networks.

Worse yet, by forging a proprietary solution the single channel vendors have gone against the tide with regard to client interoperability. Wi-Fi clients expect to move between cells operating on different channels, and many operate erratically when they hear multiple signals from different radios on the same channel. For example, the Frances Xavier Warde School noted that their single channel network was unable to work with Intel-based clients (http://www.arubanetworks.com/company/news/release.php?id=142), a horrific situation given the prevalence of these devices.

The other issue with single channel architectures is the inability of independent analysts to understand the limits and integrity of underlying architecture. The vendors hold close-to-the-vest the "secret sauce" that makes these networks function, and the revisions they make to add features over time.

A case in point. One single channel vendor offers two completely different ways of broadcasting the identity of radios with which the clients associate, one called shared bssid and the other per-station bssid. Shared bssid broadcasts the same radio name from all radios, a characteristic that in part is responsible for confusing Intel clients. Per-station bssid sends a different name from each radio on a client-by-client basis, putting considerable burden on the controller that has to track all of the name/client associations. Per-station bssid and shared bssid scale differently, work only with select radios within the vendor’s product line, and have very different performance limitations.

It seems inappropriate to ask a customer to understand the ramifications of these different approaches. Especially when adaptive micro cell architecture doesn't so burden the customer and offers superior performance.

As we celebrate the 10 year anniversary of the 802.11 Wi-Fi standard, and watch as the adoption of adaptive micro cell Wi-Fi networks grows apace, one can only wonder why anyone would choose a proprietary alternative. Does anyone remember the Wang word processor? There's a reason why it disappeared from the landscape, and no doubt it's only a matter of time - perhaps a very short time - before the same fate meets the single channel architecture.