|
The n Factor
By Dave Molta
Don’t give in to irrational exuberance over the latest Wi-Fi spec.
Cold, hard calculations are called for.
In most enterprises, Ethernet plus a little Wi-Fi equals network access. Sure, some industries thrive on wireless LANs, but more commonly, Wi-Fi is a convenience deployed to provide mobile access to information in public spaces.
Proponents of the emerging 802.11n standard say that dynamic is poised to change. They argue that the speed and range improvements in 11n will let organizations relegate wired Ethernet to the core and distribution layers of the network. Wi-Fi will emerge as the main access medium. And of course, this newfound mobility will forever alter the way people work, increasing productivity. If Ethernet is a conventional org structure, they say, Wi-Fi is the matrix.
The big question for IT: is this a compelling vision of the future or just another round of industry hype? Upward of 200 million Wi-Fi chipsets were sold in 2007, according to InStat. If you care to bet that those numbers will decline in coming years, you’ll find plenty of action. Meanwhile, the 802.11n standard is likely to be ratified by the Institute of Electrical and Electronics Engineers in mid- to late-2008, and 11n will soon make up the vast majority of Wi-Fi chipset and system sales. You do the math.
Still, time marches on. At issue for most enterprises is not whether they’ll deploy 802.11n, but when and how.
Standards Watch
Most organizations won’t consider adopting a network technology before standards are fully ratified. Experienced network professionals are content waiting for second-generation 802.11n offerings, thanks very much, while launching a few pilots.
For once, though, first adopters looking for competitive advantage won’t be putting too much on the line—there are important factors in play here that mitigate risk. First, while early phases of 802.11n standards development produced plenty of design debate, it’s been more than a year since most vendors have coalesced around a framework developed by a consortium of wireless chip developers and equipment makers. While the initial design on which 11n Draft 1 was developed had several notable deficiencies, the Draft 2 spec is reasonably solid. The IEEE has responded to thousands of formal comments related to both drafts, and any changes implemented between now and the final standard will likely be addressable through firmware updates. Even the worst-case scenario, in which a change is made that can’t be corrected in firmware, won’t render draft products inoperable with those based on the final spec. At worst, we’ll see minor problems related to performance or battery efficiency.
The second factor mitigating risk is the Wi-Fi Alliance’s decision to move forward with certification based on the Draft 2.0 specification. More than 100 products, mostly for small and home offices, have been ratified. While it’s true that the Wi-Fi Alliance certifies only a subset of standards inter-operability, it provides a de facto seal of approval that carries great weight with consumers, enterprise IT professionals and equipment manufacturers.
Finally, because the consumer market has been flooded with 802.11n Draft 2 offerings, momentum argues against changes that would diminish inter-operability. Now that prices for many consumer-grade 802.11n Draft 2 products have fallen below the magic $100 barrier, few will choose access points based on earlier standards. History has taught us that there’s a powerful relationship between the wireless technology people install in their homes and the systems installed in enterprises. The two must be compatible.
For many years, we’ve seen one simple principle predict trends in enterprise Wi-Fi adoption: follow the client. While early WLANs relied on PC Card radios, it wasn’t until Wi-Fi was embedded into client devices that the market began to take off. Users don’t want to deal with add-on network interfaces. More important, by embedding wireless cards in mobile devices, we achieve cost economies while also enhancing overall system performance and efficiency.
Intel’s Centrino is most emblematic of this trend, and the chipmaker has already delivered a Centrino 11n offering. In fact, if your organization buys higher-end enterprise-class notebook computers for employees, it’s likely that these systems will ship with integrated 802.11n support. However, if you try to save money by buying consumer notebooks, even high-end ones, you should explicitly request 11n support—assuming it’s even available.
While adoption patterns vary, three-year replacement cycles for notebooks are common. Thus, it’s unlikely that most organizations will have integrated 11n support in more than half of clients before 2009. This phased changeover makes sense, in part because there’s a natural aversion to pre-standard products, and in part because design and implementation of enterprise-class 802.11n is complex.
As far as smartphones and wireless VoIP devices are concerned, 11n is still on the drawing board.
Sea Change
Cisco Systems’ decision to jump in early with a Draft 2.0-compliant access point—the Cisco 1250—has helped legitimize the 11n market. Second-tier vendors such as Meru Networks and Colubris Networks also have moved aggressively into 11n, hoping to appeal to early adopters. But IT professionals must understand that supporting 11n isn’t nearly as simple as slapping a new radio inside an old AP model. A range of design and implementation challenges exist, in many cases forcing network engineers to re-examine their approaches to wireless deployment.
The central opportunity—and challenge—associated with 11n is higher performance. With APs capable of offering at least five times greater throughput, users are sure to be impressed...provided your overall architecture can support the additional traffic. Older designs may need to be reconsidered, and many vendors are making fundamental changes to their system architectures.
At the simplest level, access points will need faster wired backhaul interfaces. Aggregate wireless performance on a dual-band, dual-radio 11n AP will easily exceed 100 Mbps, so most enterprise APs will benefit from Gigabit Ethernet uplinks. To take full advantage of improved performance, you may need to upgrade access-layer switches. Even greater challenges exist with respect to powering these new 11n APs. Current draws exceed those in 802.11af power-over-Ethernet designs, no matter what vendors claim.
Wireless controller capacity is also a significant concern, prompting a number of vendors to introduce new multi-tier wireless switches. While many enterprises streamline deployment and management by centralizing controller functionality at the core layer—Cisco’s WISM-equipped Catalyst 6500 is a prime example—the higher performance of 11n may lead some network architects to reconsider a design in which all traffic is carried back to the core for processing.
Many vendors will simply beef up their centralized controllers to support higher speeds, but some, including Colubris, Meru, Nortel, and Trapeze, are touting new enterprise Wi-Fi architectures that distribute WLAN controller functionality out to the distribution layer, even to the edge switch or access point.
Now while such a design may have merit, changes in system architecture also introduce risk. Vendors may make fundamental changes in how and where packets are processed, for example. In the centralized-controller model, identification and remediation of bugs is much easier than is the case where controller intelligence is distributed. As the architecture becomes more complex, the potential for problems increases.
In the end, it’s likely that even vendors advancing a distributed design will also need to support higher-speed centralized controllers for customers that prefer the relative simplicity of such setups.
While controller capacity and wireless/wired integration challenges are notable, even more thorny issues relate to achieving inter-operability between 802.11n and older protocols like 11g without sacrificing performance. Our co-existence testing of Draft 2.0 802.11n and 802.11g, both running at 2.4 GHz, suggests that aggregate throughput decreases by about 40 percent.
Vendors like Meru and Extricom, which employ sophisticated scheduling algorithms to allocate capacity to clients, will likely do a good job managing 11g/11n co-existence at 2.4 GHz. However, most organizations will find that 11n provides an ideal opportunity for segmentation: legacy traffic can continue to operate at 2.4 GHz, while 11n traffic is moved to 5 GHz. That strategy lets IT more easily leverage the 40 MHz channels supported by 11n while still taking advantage of the much greater 5-GHz capacity.
In the past, organizations that were contemplating segmentation of traffic between 2.4 GHz 802.11g and 5 GHz 802.11a had to deal with the diminished transmission range at 5 GHz, forcing many to deploy expensive and complex microcell architectures. Because 802.11n significantly improves Wi-Fi’s transmission range, 5 GHz deployments are now much more practical. In fact, our early testing suggests that 5 GHz 802.11n performance far exceeds that of 2.4 GHz 802.11g at every distance up to 130 feet. It’s conceivable that many organizations will find it possible to upgrade their large-cell 802.11g deployments to dual-band 802.11n without having to relocate access points.
Performance Expectations
Here’s what you get for all this network rejiggering: the initial 802.11n design spec called for throughput in excess of 100 Mbps, three to four times what you’ll achieve with current 11a and 11g. Published performance tests of consumer-grade 802.11n products conducted by Craig Mathias of the Farpoint Group show that most products approach, or even exceed, that threshold.
In the enterprise market, Cisco has released early performance tests of its new 802.11n Aironet 1250 AP operating with a Lenovo T61 notebook and an integrated Intel Centrino 4965AGN wireless adapter. Using 40 MHz channels, Cisco measured peak TCP throughput of about 147 Mbps. Throughput with standard 20 MHz channels was about 89 Mbps. A baseline 802.11a/g test turned in throughput of 23 Mbps.
Although we haven’t yet brought a full set of enterprise-class 802.11n APs into our lab, we’ve spent the past several months running Apple’s Airport Extreme through a battery of tests, achieving peak throughput of 137 Mbps with a MacBook 5 GHz 11n client. Our decision to focus on Apple’s offering was predicated on three factors. First, Apple has a reputation for high-quality product design, and our briefings demonstrate that the company has an in-depth understanding of Wi-Fi. Second, the Airport Extreme is a dual-radio 11n AP, providing support for 2.4 GHz and 5 GHz operation. Finally, the product is based on the Atheros Wi-Fi chipset, a popular platform for enterprises. (For more details of our preliminary performance testing, see box, ‘Apple UPS 802.11 Performance Ante’)
Jump Or Hold?
We have no doubt that 802.11n will emerge as the predominant WLAN platform by 2009. Organizations considering a sizable deployment of Wi-Fi access points are thus understandably concerned that going with 802.11a/g WLANs in 2007 or 2008 means choosing legacy technology over a more strategic, powerful alternative. At the same time, jumping too quickly poses considerable risk, not so much because the final standard may change, but because you’re buying first-generation devices. This risk increases where vendors are touting new controller architectures to complement their new APs.
Our advice: if your existing WLAN serves your needs, stretch its lifespan as long as you can. Few organizations are bandwidth-constrained on their WLANs, so instead of spending time and effort on upgrades, most should re-evaluate their wireless security postures.
For greenfield deployments, the calculus is tougher. Unless compelling ROI can be demonstrated, we recommend stretching WLAN deployment out over a longer timeframe, say two to three years, using a mix of 11a/g and 11n APs. This won’t eliminate the trade-offs between 11a/g and 11n, but you’ll be in a better position to realize the full benefits of mature 802.11n offerings.
There’s little doubt that second-generation 11n products, likely to be available by late 2008 or early 2009, will offer significant enhancements. A longer-term phased deployment may prove the smartest approach, provided it doesn’t hamper application and mobility plans. Under this strategy, a limited number of 802.11a/g access points can be installed alongside a limited number of 11n APs where performance requirements dictate higher speeds and suitable client devices are available.
|