Wireless Ethernet just got a boost: The A Team arrived, and it is significantly faster than the B Team. The first batch of 54Mbps 802.11a products is shipping, and we tested samples pointed toward enterprise and smaller installations. These products will be known by the Wireless Ethernet Compatibility Alliance designation Wi-Fi5 – when they pass WECA’s interoperability tests scheduled for this spring.
Products based on 802.11a are supposed to be very fast and not as rife with mistakes as earlier products in the 802.11b/Wi-Fi space. The first batch of products we tested have obviously benefited from the 802.11b experience. They were easy to set up and configure, but the choices varied widely. And while the 802.11a interoperability was good, we found the nonstandard high-speed modes (where available) didn’t interoperate. The temptation to use these modes will be high and will likely force buyers to source through one vendor unless a movement is made to achieve de facto rather than standards-based interoperability for these “turbo” (multichannel, nonstandard) modes.
We tested four early released products from three companies that represent two differing target markets (enterprise space and small to midsize businesses): Proxim Inc.’s Harmony (targeted at the enterprise space), Proxim’s Skyline Access Point, Intel Corp.’s Pro/Wireless 5000 and SMC Networks Inc.’s 2755W access point. All four are based on the same wireless chipset/reference model from Atheros, but unlike cookie-cutter 802.11b products, their flexibility, management and performance varied. And unlike former iterations in 802.11b access points, they all look aesthetically different.
In our tests, three of the four units performed consistently, Proxim’s two products and Intel’s 5000. The speed demon (over the widest broadcast range) in the bunch, SMC, had very fast performance – and occasionally had highly erratic behaviour until we downloaded “another Version 1.0” of its firmware. Because the markets that the products are reaching are different, and because the products are in their early stage, we didn’t score this as a traditional comparative review.
Compatibility between different access points and the CardBus adapters supplied was very good. In 802.11a mode, which runs at up to the 54Mbps data rate, all the adapters could connect to every access point – even with Wired Equivalent Privacy (WEP) encryption enabled. In ad hoc mode (connecting two PCs without an access point), all the adapters worked well together with the other access points, with some cards having better speed.
But when we used the access points’ turbo mode between an access point and a competing adapter, we couldn’t achieve a connection between that card and the access point.
For now, turbo modes shoot compatibility dead between different products, resulting in a disappointing performance.
With the older IEEE 802.11b cards, the stated data rate is up to 11Mbps, and typical performance (data throughput) is about half that rate under ideal conditions. Consider also that 802.11b cards send some signals at 1Mbps data rate, and have considerable overhead in traffic.
At the seemingly blazing data rates advertised for 802.11a, 54Mbps performance would make this about five times faster than 802.11b. This was true under ideal conditions.
Proxim clocked in the fastest modes, both in compatibility mode and with its own access point. Running against the Intel access point, the Proxim card turned in a 22Mbps score. Proxim also turned in the best score with its own access point, with 17.6Mbps as measured by our FTP program.
Things changed when we tested range. In close range, Proxim was fastest on average. But SMC’s access point had the best range. For consistent numbers, Intel’s was slowest on average overall, but was the most consistent across different ranges.
In our tests where we took the adapters and ran them against the other access points, SMC’s adapters were most consistently speedy over the widest geographical range. But we also had the most difficulty in completing our tests with these adapters. During the testing, one in seven tests failed with the SMC adapters when linked to the SMC access point until we reloaded the firmware. We got the feeling we had received prototypes instead of working units and later confirmed this. Oddly, links from the SMC notebook adapters to other hubs were otherwise consistent and performed well.
SMC’s 2755W access point also had the best range of the devices tested. In the multi-storey office building tests, the SMC access point was the only one that gave usable (if very slow) results on the floor above the access point. Antenna positioning, available on Proxim access points and the SMC access point, didn’t have much impact on the results.
Of the access points tested, Intel’s was the only unit that shipped with mounting hardware and was also the heaviest.
Proxim says the Harmony line will soon support Power-On-LAN (POL), a method that puts power onto unused pairs in a Category 3, 4 or 5 cable connecting an access point, but this feature wasn’t available in the units we tested. SMC and Intel do not claim support for POL. All three units shipped with transformers for power, and the SMC access point had the shortest power cord.
Proxim’s Harmony product line uses a central wireless hub management appliance to control the Harmony access points. The management appliance then controls the Proxim access points (802.11a or 802.11b) on a subnet.
We found the Web-based access point controller application to be incredibly thorough to use with multiple Harmony access point deployments.
If the Harmony access points are located on the same subnet, an access point controller can be set to apply rule sets to any newly joining Harmony access point. Indeed, we found that when the controller goes offline, the access points stopped within a few seconds – they’re slaves to the controller.
Proxim also produces an access point in its Skyline product line – its characteristics were virtually identical in performance. The main difference between Skyline and Harmony is the management relationship between the access point controller and Harmony access points.
The Harmony line also includes a proprietary encryption algorithm that’s different from WEP, although Proxim refused to say how it’s different. This encryption method is in addition to the 54-bit through 128-bit WEP encryption methods available and common to all the access points we tested.
Another Harmony feature is a forced VPN tunnelling method that can also force logons to a single server. This adds much security to mobile networks, which potentially could dramatically reduce drive-by logon risks. Additional security imposed at the server that is the target of the tunnel requires the drive-bys to also get past whatever security is imposed at the server.
The Harmony access point controller also supports Remote Authentication Dial-In User Service (RADIUS) for added authentication, media access control (MAC) layer address admittance control and SNMP for monitoring. Multicasts and IPX debris such as Routing Information Protocols and SAP server.
The Intel access point has fixed antennae. An option is available to limit the antenna pattern to “half-circle front” from an omnidirectional pattern. We found only a nominal boost in performance when we chose this option.
Like the other access points tested, the Intel unit has a wide choice of channels. No provision is made for traffic or protocol filtering.
While slow on performance, it was consistent across the ranges. We tested the Pro/Wireless 5000 less than the others because it had no errors and no additional modes to test.
Sometimes boring has its place in connectivity; Intel’s product was consistent and satisfying in its consistency.
We got a hint of trouble with SMC’s 2755W access point when the Web interface kept asking if we wanted to install a Chinese language pack. We had to consistently decline, even though the interface was in English at every page change. This, with other inconsistencies (a sample that wouldn’t update and highly erratic performance) led us to believe we had prototypes and not production units. We were right and received production units from SMC. A new firmware update from SMC’s Web site for the 2755W units removed the aforementioned problems.
Installation was simple. SMC has a set-up wizard that is similar to Intel’s rapid deployment method. Help screens that were offered were only nominally useful in describing functionality.
Like the others, SMC’s access point has provisions for choosing data rate minimums (for acceptance as a client to the access point) and a power selection. These features allow tailoring of multiple access point coverage. A lower power rate might prevent interference with another access point, and data rate minimums prevent accidental cross-access point logons in some installations. Performance is more affected with power choices in the SMC 2755W than the antenna directionality choice in the Intel Pro/Wireless 5000.
The SMC access point has similar security constraints as the Intel Pro/Wireless 5000 by not supporting VPN (as a client stack only), tunnelling, homed servers, RADIUS or encryption beyond WEP security. Like the Intel Pro/Wireless 5000, it does support access control lists by MAC layer and WEP key control.
SMC’s turbo mode is designed to run at up to 72Mbps. We found this mode fastest of all tested. Of particular note was that at higher data rates, the range vs. data rate wasn’t affected as much as with the Proxim Harmony access point.
Range was less affected in the nonstandard turbo mode than Proxim’s range with its turbo mode. SMC had the best performance for all the tests, despite occasional inconsistencies. However, we also were concerned that we might be transmitting illegally because of the frequency ranges.
Some Ready for Prime Time
We feel this fresh crop of access points is ready for use. Proxim’s Harmony system is time-proven, and we feel the difficulties we saw are related to the new 802.11a software (drivers for the access point and notebook adapters). SMC’s access points were eventually consistent, as the range was good and the turbo mode seemed to work very quickly over a wide range. Where the simple criteria of speed and range are concerned, SMC shines.
However, we’d like to see better security from SMC and Intel. The lack of RADIUS support hurts companies that would depend on them. Clearly, Proxim’s Harmony product is the most enterprise-ready of the early A Team access points. Proxim’s access point controller creates additional expense in a wireless network, but its many security features (while not enhancing interoperability) and rapid deployment capability are compelling.
Another failing is the lack of a PCI-bus adapter. We were warned not to use CardBus/PCI adapter cards that we use with 802.11b because the cards might not survive. We could find no PCI bus adapters for the current crop of 802.11a CardBus adapters, and this shuts out desktops from participating in the higher-speed wireless networks – at least until PCI bus adapters and Universal Serial Bus adapters come to the marketplace.
How We Did It
We tested the access points and cards in a single-floor office building (with steel wall studs and a sheet metal roof) and a nine-storey office building. In the single-storey building, we set three test points: 3 feet, 60 feet through two walls, and 72 feet through four walls from the access points, which were mounted at the top of a free-standing rack.
We configured two identical Compaq Presario 700 notebooks, each running Windows XP Professional, to be clients. During our initial test phases, we immediately found problems getting good results. It appears that the client IP stack on XP Professional has problems – especially with the SMC drivers. We reconfigured one of the Presario 700 notebooks to run SuSE Linux 7.3 and serve an FTP server, and achieved consistent results.
The test procedure used the XP Professional Client’s FTP application to get a 36MB file. The data rate performances recorded were the data rates specified by the XP command line interface-based FTP application on an average of 10 iterations after we threw out the two worst scores (which seem to come from aligning cache through the delivery chain). Note that a baseline wireline measurement through the Presario 700’s built-in Realtek 100Base-TX network card yielded a rate of 78.4Mbps.
Tom Henderson and Tim Ritchey are researchers at ExtremeLabs. They can be reached at[email protected]and[email protected].