Cisco's 11n AP: Timing Is Everything

Responsible IT decisionmakers are wary of prestandard technologies, having discovered the hard way that "prestandard" is often code for "proprietary, premature and/or prone to vendor lock-in." Still, sometimes you have to walk on the wild side. Occasionally a nascent technology provides enough of a competitive advantage that it's worth taking a chance, or events make us comfortable that vendors will stay on the standards path. 802.11n Draft 2.0 is a perfect example, so when Cisco offered to let us test an early version of its 1250 access point in our Syracuse University Real World Labs, we were excited to get a glimpse of the future of enterprise WLANs.

From a performance standpoint, our testing revealed speed increases of four to six times what an 802.11a/g infrastructure can provide. And using an 802.11n access point for even legacy a/b/g clients delivered a measurable performance advantage thanks to MIMO (Multiple Input, Multiple Output) technology's ability to maintain high-bandwidth data rates for a larger portion of an AP's coverage area. For a/b/g voice over WLAN phones this greater reliability translates into higher quality calls and fewer dead spots.

We believe that 802.11n's bug shakedown period will be a less rocky version of what occurred when 802.11g was an IEEE draft. Unlike 802.11n, 802.11g's Wi-Fi Alliance interoperability certification occurred after the standard was set in stone; in contrast, products based on 802.11n Draft 2.0 are being certified today, before the standard's ratification by the IEEE. That's great news for enterprise IT: As of this writing 225 products, albeit predominantly SOHO/consumer oriented, have received the seal of interoperability from the Wi-Fi Alliance. Notable names on that list include Apple, Aruba, Atheros, Broadcom, Cisco, Intel and Meru—all big businesses with an economic interest in ensuring that the current crop of 802.11n Draft 2.0 products are both backward and forward compatible.

This isn't to say maximum performance won't increase as subsequent versions of 802.11n are endorsed by the Wi-Fi Alliance. But it does mitigate the risk that current devices will be incompatible with future revisions. The truth is, companies don't buy WLAN devices based on an IEEE standard, but on the Wi-Fi Alliance's endorsement of interoperability and the independent certification that 802.11n Draft 2.0 has received. In addition, Cisco is a member of Intel's more exhaustive "Connect with Centrino" interoperability testing program, which gives an added comfort level.Setting A 5GHz Strategy

The 5GHz frequency has been underutilized compared with the more popular 2.4GHz band, but with the advent of 802.11n that's poised to change. In the early days of WLANs, light user loads and a focus on maximizing coverage made the superior propagation characteristics of 2.4GHz a clear choice over 5GHz's more limited range. Now that those WLANs have grown from scattered hotspots to pervasive coverage blankets with many microcells supporting a multitude of users and high-bandwidth applications, the focus has shifted from coverage to capacity. And when you're talking capacity, nothing beats the massive amount of spectrum available in the 5GHz band, which encompasses 21 nonoverlapping channels when an AP implements full DFS2 (Dynamic Frequency Selection 2) support, compared with 2.4GHz's modest three nonoverlapping channels.

Interference is another differentiator. WLANs in the 2.4GHz band must contend with cordless phones, microwave ovens and other 802.11b/g devices, whereas 5GHz is relatively vacant except for cordless phones, occasional military radar sites, and a few 802.11a networks. Although the 802.11n standard describes support for either frequency band, 5GHz is the clear choice when channel bonding comes into play because it supports numerous nonoverlapping 40MHz channels, whereas 2.4GHz can support only one.

In addition to increasing speeds with wider 40MHz channels, 802.11n also boasts additional OFDM subcarriers, MAC Layer packet aggregation and MIMO. Core to the technology's increased performance, MIMO allows 802.11n to leverage environmental multipath reflections. Multipath has caused dead spots for 802.11a/b/g devices because they had a limited number of antennas, or RF viewpoints, to sort out reflections. But in 802.11n, multiple antennas and mulitple radios help to increase reliability and even allow multiple data streams or "spatial streams" to be sent concurrently. As for nomenclature, a MIMO device with 2 transmit and 3 recieve antennas supporting 2 spatial streams would be refered to as '2x3:2' MIMO following a 'TxR:S' convention. At its projected maximum, 802.11n using 4x4:4 MIMO will support 600 Mbps raw data rates. Current Draft 2.0 products top out at 300 Mbps using 2x3:2 or 3x3:2 MIMO.Peak Performance

We won't say this review was without hiccups—in fact, we experienced two show-stopping issues in the beginning of our evaluation period, involving how Cisco's 1250 handled an older Intel 2915 a/b/g WLAN client in high-traffic and abrupt-disconnection situations. But a few glitches are to be expected when testing prerelease software, and Cisco responded quickly. After receiving updated controller code, version 4.2.94.0, we had clear sailing; an official version of that code train was released as 4.2.99.0.

We benchmarked the Cisco 1250 AP against Cisco's 1240AG device using Ixia's Chariot High Performance script in both open air and Azimuth System's ADEPT-n isolated environment. We used a Cisco 4402 controller and found the 1250 AP was as easy to configure as any Cisco LWAPP access point. Our WLAN clients included a Lenovo T61 with an internal Intel 4965AGN chipset and an IBM T43 featuring Intel's 2915ABG silicon.

As a high-water mark, the Cisco 1250's maximum TCP Layer performance was 154.9 Mbps using a 40MHz channel in the 5GHz band—more than six times the performance of our 11a baseline of 22.8 Mbps. For a smaller 20MHz channel size, performance was 84.8 Mbps and 96.7 Mbps for 5GHz and 2.4GHz, respectively; even after spectrum analysis of our testing site, we're unable to explain why 2.4GHz is faster. That anomaly aside, Cisco's 1250 provided performance that was three times to four times greater than peak 802.11a/g throughput.

We also ran peak assessments with AES encryption enabled and found that Cisco delivered line-rate performance in all but 40MHz channel sizes, where throughput dipped to 144.5 Mbps, about a 7% loss. Cisco told us that optimizing its hardware encryption engine is a fundamental goal of the latest software release, and we expect this gap to narrow in the future. We also measured coexistence performance when both an 11n and an 11a/g client were associated to the same radio on the 1250. With both clients, overall cell throughput dropped to 60 Mbps, with an 80% to 20% speed split at 5GHz, a 90% to 10% split for 2.4GHz.

We performed additional mobile rate vs. range testing inside a 50,000 square foot warehouse, with cinderblock and sheetrock walls, on the outskirts of the Syracuse University campus. We selected this location for its RF isolation and lack of an incumbent production WLAN. Then, we upped the ante with a rotating Sherline Products digitally controlled turntable, which introduced movement in the client notebook's antenna orientation and systematically varied multipath conditions. We kept rotation speed at 1 RPM. After enabling AES encryption, we again used Chariot for traffic generation with bi-directional TCP pairs to assess overall performance with strong security; we took multiple test runs and averaged the results at distances of 15, 75, 80 and 130 feet. Across the entire location set, 802.11n's performance was four times better than 802.11g and better than 802.11a by three times and five times for 20MHz and 40MHz channel sizes, respectively.

We also saw increased reliability of an 802.11a/g client's connection across the farther distance locations, at 80 and 130 feet. In the 2.4GHz band, the 1250 AP maintained 20.8 Mbps as a minimum throughput, whereas the 1240 stepped performance down to 14 Mbps at the edge. In the 5GHz band, minimum performance was 15.7 Mbps vs. 11 Mbps for the 1250 and 1240, respectively. That means that even if your clients are mainly legacy a/b/g, an 802.11n infrastructure will still deliver a performance boost thanks to MIMO. If voice over WLAN is on your agenda, this will translate into much higher quality and coverage.

Our last measured performance metric, increased range, is one of the most heavily touted advantages of 802.11n. Although we aren't here to burst that bubble—our testing and subjective experience with Cisco's 1250 found slight enhancements over Cisco's 1240AG—understand that boosting range is, in itself, not a reason to upgrade.

We used AirMagnet's Survey Pro 5.1 with a Cisco CB21AG client card to perform walkabouts, measuring signal strength values around our 390 foot by 150 foot test facility. In the 5GHz band, the 1250 outshone the 1240AG in a few areas but exhibited a similar maximum range. In the 2.4GHz band we found coverage between the 1240 and 1250 nearly identical, with a measurable improvement over 5GHz only at the very extremes where numerous heavy walls provided challenging conditions. Cisco's internal range evaluation showed a 10% improvement in comparing the 1250 with the 1240AG, and of course, real-world increases vary based on environmental obstacles, distance and the quality of the WLAN client used to connect. Cisco recommends a 1:1 AP replacement strategy of 802.11n to 802.11a/g APs and employing the same microcellular approach as previous AP generations. Although in theory one could install a slightly fewer number of 11n APs to deliver the same aggregate performance as a denser 11a/g setup, we recommend continuing today's microcellular deployment approach to future-proof for tomorrow's bandwidth requirements.

The Power Problem

Of all the controversies within the 802.11n marketplace, nothing quite rivals the "he said, she said" vendor debates around PoE (power over Ethernet) support in 802.11n access points. Because MIMO requires multiple radio chains to work its magic, the power requirements of dual-band 802.11n APs generally exceed that which 802.3af PoE can provide, forcing vendors to come up with a variety of creative remedies.

Cisco's solution involves injecting additional wattage onto an AP's wired connection, either through 1250-specific power injectors or the enhanced PoE capabilities available in its flagship Catalyst 3750-E and 3560-E switches. Cisco also currently offers the option to run the 1250 platform with local AC power or just a single radio module on PoE. Among the competition, Siemens' recent announcement that it can fully power its 3x3 MIMO dual-band access point on standard-PoE created a bit of a stir, and Aruba states it can provide similar functionality. Given that Trapeze, Aruba and Siemens all use 802.11n silicon from Atheros, we'd be hard pressed to believe that power-savings modifications by one vendor can't be duplicated by the competition. Aruba also says that its AP will step down from 3x3 to 2x3 MIMO and slightly reduced performance when longer (greater than 100m) cable lengths reduce PoE current. While our labs have not assessed these vendor's power claims, the ability to fully power a dual-radio 802.11n access point with today's 802.3af PoE is an attractive prospect and would certainly give the competition a valuable differentiator over Cisco's current offering.

For a more permanent fix, a standards-based solution is currently being worked out in the IEEE that will offer more juice to these power-hungry APs. PoE Plus, or 802.3at, is on track for ratification in 2009 and will require edge switch replacements. Industrial Image

On the hardware front, the Cisco 1250 has a utilitarian and industrial design with a hardened white exterior accommodating two modular LWAPP radios when fully loaded. Although radio choices are currently limited to 2.4GHz or 5GHz 802.11n Draft 2.0 modules supporting MIMO 2x3:2 (two transmit, three receive with two spatial streams), additional options sporting enhanced MIMO capabilities may be available in the future. For our tests we used three omni-directional antennas on each radio module, but Cisco also offers a variety of compatible high-gain directional antennas.

Although the modularity of the 1250 series provides investment protection, much like that of the earlier Cisco 1230 platform, these APs are also larger than competitors' devices: Aruba Networks' AP-124 weighs in at 15 ounces, for example, compared with the Cisco 1250, which is on the bulkier side of the spectrum at a hefty 5.1 pounds with two radios installed, and an 8 inch by 9 inch footprint. We'd consider the device fine for manufacturing or industrial sites, but it will likely stick out if deployed in a plain sight in a corporate environment. We're hoping a non-modular 802.11n access point is on the way as a replacement to the Cisco 1130AG AP, with a demeanor that fits better into the carpeted enterprise.

Bottom line, at $1,299 for the dual-radio version, the 1250 is a bit costlier than current generation offerings, but the price/performance ratio is weighted in 11n's favor, especially if you've invested in 11n clients or are piloting voice over WLAN. Reseller discounts will keep the price under the $1,000 mark, but the highly recommended more powerful 1250 PoE injectors will add additional cost. Despite a few quirks with beta code, our overall experience with the product was positive, and we were impressed with its performance. The verdict: This is compelling technology worth a pilot or spot deployment where high-bandwidth or mission-critical applications demand the best.