US Robotics 802.11g Wireless Turbo Multi-Function Access Point - Page 2
October 08, 2003
When it came to performance, the unit performed well in some scenarios and not so well in others. One thing is for sure--no reasonable interpretation of the results can let one to believe that it delivers performance equivalent to 100Mbps, or that its 100% faster than the competition.
G-only mode Chariot performance testing on the 5450 revealed results that were within range of products based on competitive chipsets. In native mode with a USR 5410 PC Card client adapter for example, the USR5450 exhibited throughput of 20.67Mbps at 10 feet, which dropped to just under 9Mbps by 125 feet. Enabling the "Turbo" packet bursting mode (on both the AP and the client) did improve performance measurably. With the feature turned on, peak throughput at 10 feet rose from 20.67 to 26.52 Mbps, an improvement slightly more than 25%.
On the other hand, during distance testing the USR5450 exhibited an odd performance characteristic at ranges beyond 25 feet. In these tests, the throughput levels as plotted by a Chariot graph resembled a compressed sine wave, with the throughput alternately rising and dropping substantially at almost exactly at two second intervals, about 30 times over the course of the test. This phenomenon exhibited itself at only at distances beyond 25 feet, which happens to be the approximate point in my environment where line-of-sight with the access point is lost.
Re-orienting the 5450's antenna or the unit itself sometimes reduced the effect, but it didn't eliminate it entirely. Observing the client during the test indicated that the signaling rate did fluctuate at times, but not in a pattern that correlated to the Chariot graph. This behavior was not displayed by other 802.11g products I've recently looked at.
The volatility of the USR5450's throughput rate did have a clear impact on performance, but since it was transmitting at a high rate through most of the test it suffered only about a 15% throughput penalty as a result of the fluctuation. Typically, the existence of persistent interference would cause an access point to consistently operate at a lower rate or perhaps adjust the rate as conditions varied, but these results still suggest that the USR5450 may have difficulty maintaining a consistent signal quality at distance in some environments. USR wasn't able to induce the same behavior.
In the mixed mode test, the access point produced throughput of 10.31Mbps for an 11g client with a 11b client associated. That figure jumped to 17Mbps with Turbo mode enabled, an increase of almost 70%. A run with 11b and 11g clients transmitting simultaneously dropped aggregate throughput to 7.96Mbps (with 5.27Mbps for the 11g client and 2.71 for the 11b client). These results certainly didn't represent the best throughput among 802.11g products, but neither are they the lowest. Enabling the Turbo mode in this scenario improved total throughput to 11.7Mbps, and gave the 802.11g client a big boost to 9.72Mbps. The 802.11b client throughput in turn dropped slightly, to 2.05 Mbps.
Since one of the purported benefits of the TI chipset is compatibility with the 22Mbps signaling rate of 802.11b+ clients, I did an additional run substituting a TI-based 802.11b+ client for a conventional 11 Mbps client. In this test, aggregate throughput rose from 7.96Mbps to 9.60Mbps. And while the throughput rate for the 11g client climbed only slightly, to 5.82Mbps (from 5.27), the 11b+ client fared well with 3.81 Mbps of throughput, compared to the 2.71 the b client had posted.
My final test was to remove the 802.11g client and run the 802.11b+ client alone (again with 4x enabled). In this case, it actually posted throughput numbers in excess of 10 Mbps. That's almost twice what the b+ client would be expected to do while connected to a native 802.11b+AP, indicating that the 5450 can breathe new life into older PBCC clients.
Attempts to associate a non-TI 802.11g client to the 5450 went off without a hitch, and the client (in this case, and Intersil-based card) performed comparably to the TI client with Turbo mode inactive.
USR touts the TI chipset's ability to provide superior performance and support the widest variety of clients at their native signaling rates. In actuality, while TI-based 11b and 11g clients can benefit greatly from Turbo/4x mode, the test results suggest that when this mode is not used, the 5450 doesn't provide any better performance for 802.11g and plain-vanilla 802.11b clients than access points based on other chipsets.
Features of the USR5450 like bridging, repeating, and 802.1x support (and soon WPA) compare favorably with most of its peers, and the performance benefits of the Turbo mode are clear. Therefore, if you're starting a WLAN from scratch and plan to stick with TI-based hardware, or if you want to gradually upgrade to 802.11g technology but still have a large installed base of TI-based 802.11b+ clients that you want to take full advantage of, then the US Robotics 5450 is clearly a good choice.
On the other hand, non-TI clients won't be able to reap the performance rewards of the 5450's Turbo mode. This fact, combined with the product's very high $200 street price makes it a less compelling purchase in many cases. If you don't have a significant investment in existing PBCC hardware, you will likely find an equally capable access point at a lower cost. A lower price, better documentation, and the inclusion of WPA support, would tip the scales towards the 5450's favor.