Rates were generally highest for upstream traffic, as in the 802.11n tests. However, unlike the 802.11n tests, rates tailed off dramatically both for bidirectional and downstream traffic. Further, even the very highest upstream result was nearly 30% lower than with 802.11n clients only, despite two radios being active and thus twice as much capacity theoretically being available. (Read a recent test on wireless LAN management gear.)
Because few applications involve traffic in one direction only, and since few enterprises will run 802.11n alone on day one, our results suggest that network managers shouldn't expect the same high rates from mixed-mode deployments as with pure 802.11n setups. To be sure, rates in these tests are again far higher than would be possible with 802.11g or 11a access points; but they're nowhere near as fast as in the pure 802.11n tests.
One possible counter argument is that network designers should deploy all 802.11n clients on the 5-GHz radio and dedicate the 2.4 GHz radio to legacy 802.11g and 802.11b clients. This may be a sound network design practice, but a quick look at frame rates belies the argument that overall performance would improve. None of the systems came anywhere close to meeting theoretical limits of around 1.5 million frames per second with short frames or 178,000 frames per second with long frames (measured across eight access points). The big gap between theoretical and observed frame rates suggests that access points' CPUs and buses will be limiting factors well before the systems hit any bandwidth bottlenecks.
We also measured latency and jitter for mixed-mode traffic. Average latency was generally slightly lower than in the 802.11n tests, not surprising considering the lower loads involved. Once again there were big differences between average and maximum latencies, with the latter jumping well above 1 second in three cases involving the Aerohive and Siemens access points.
Certainly these high maximum delays can adversely affect applications. However, because jitter remained relatively low across the board it's likely that the high maximum latencies were caused by only a few stray frames (something we verified in the case of Aerohive's access points by examining capture files).
Power consumption is a key concern with 802.11n. The central question is whether the new 802.11n access points will draw more than the 12.95-watt maximum permitted in the existing 802.3af specification for power over Ethernet (PoE).
Some of the more power-hungry access points may need even more juice than the 15.4-watt maximum that today's PoE power sources provide. (The 2.45-watt difference between device and power source limits exists to account for power dissipation in cables and voltage fluctuations; in practice, actual dissipation is much smaller, even over maximum-length cables, typically a few hundred milliwatts.)
Power usage is a major issue for some enterprises, especially those that only recently put PoE switches or injectors in place. For network designers, the question is whether it's necessary to trade off some performance to stay within the power budget. The IEEE is working on a higher-wattage version of the PoE spec, but work isn't yet complete.
To determine maximum power draw, we enabled both radios on one access point from each vendor and associated 20 802.11n clients to each radio. We also configured the access points to use channel bonding, ensuring maximum bandwidth and thus the highest possible power draw.
Working with a Fluke multimeter and probe, we took three measurements: Once with no traffic to determine power usage when idle, and again with downstream flows of 88- and 1,518-byte frames, each offered at the throughput rate. The Fluke multimeter recorded the maximum power used in each test.
Clearly, the greenest of all the access points came from Siemens. When idle, the Siemens' HiPath Wireless AP 3620 used only 6.3 watts, less than half the limit for the existing PoE spec. Even under the heaviest load, the Siemens access point drew less than 11 watts, again well under the 12.95-watt limit. These results validate Siemens' claim that its 802.11n gear does not require a forklift upgrade of existing PoE infrastructure.
At the other end of the spectrum, the Aerohive HiveAP 340 was over the 12.95-watt line in all three tests, drawing as much as 18 watts when forwarding 1518-byte frames. Aerohive access points have a "SmartPoE" feature that can dynamically adjust power consumption to match that available from an 802.3af-compatible power source, but we did not test this. After reviewing its PoE test results, Aerohive said SmartPoE would have resulted in significantly less power draw, roughly equivalent forwarding rates and a smaller coverage area, but again we did not verify this.
Motorola's AP-7131 also exceeded the current PoE limit but only when handling 1,518-byte frames. While it's probably possible to run the Motorola access points with existing PoE gear (because of the 2.45 extra watts of headroom between devices and power supplies, it's safest to use new "PoE-plus" power sources with either the Aerohive or Motorola access points to supply power at levels above 15.4 watts).
As noted, there are power/performance tradeoffs involved in assessing PoE. Traffic rates for Aerohive's access points were much higher than others in this event, but then again so was power usage. For enterprises looking for the absolute fastest system, adding new power supplies may be worthwhile. On the other hand, enterprises looking to leverage existing PoE infrastructure are safe with either the Bluesocket or Siemens access points, as both stayed under the 12.95-watt limit in our tests.
The Siemens access points offered the best combination of power and performance: They delivered more traffic faster per watt used than any other system tested, while at the same time staying well under the power budgets of existing PoE gear.
Even though all systems implement the same 802.11n protocol, and use the same Atheros radio chipset, we saw very different results in testing. The new 802.11n systems already offer vastly higher performance than their predecessors, and with further refinement of their software they could represent a real step toward making wireless the default when it comes to enterprise connectivity. (Compare other wireless products in Network World's buyers' guide.)
Newman is president of Network Test, an independent test lab in Westlake Village, Calif. He can be reached at email@example.com.
Thanks: Network World gratefully acknowledges the support of test equipment vendors that made this project possible. VeriWave supplied not only its WaveTest WT-90 test system but also considerable engineering support. Those at VeriWave supporting this project included Tom Alexander, Tim Bennington-Davis, Carl Brown, Eran Karoly, Jerry Perser and Hardev Soor. Thanks too to Fluke, which provided a Fluke 87V multimeter and i30 DC clamp meter for measuring power consumption.
Newman is also a member of the Network World Lab Alliance, a cooperative of the premier reviewers in the network industry each bringing to bear years of practical experience on every review. For more Lab Alliance information, including what it takes to become a member, go to www.networkworld.com/alliance.
With Jupiter Broadcasting’s announcement that it is shutting down the Linux Action Show podcast, Bryan...
Jared Kushner has had a twitter account for eight years but uses it nothing like his famous...
A review of 18 companies that offer free cloud storage
The FBI warns that attackers are targeting vulnerable FTP servers used by small medical and dental...
10 early-warning signs of a failing vendor or product.
Scammers leverage an alleged iCloud account leak that also is likely not real.
Following a breach, organizations should focus on mitigating damage and data loss and providing...