The iPhone 4 is the first to support 802.11n, which offers the highest Wi-Fi data rates and throughput. But it runs only on the crowded 2.4GHz band, and at one university, which is deploying hundreds of the new devices, that poses some big Wi-Fi challenges for IT.
The iPhone deployment at Abilene Christian University, in Abilene, Texas, is unusual, possibly unique: there can be up to 500 or more iPhone 4 handsets in a big lecture hall, all trying to connect to the hall's collection of wireless access points. It's especially frustrating because ACU's IT group had successfully deployed hundreds of 11g iPhones, in the same lecture hall, on the same 2.4GHz band with minimal problems.
For now, in the areas with 802.11n access points, 11n has in effect been turned off, and the new iPhone 4s will run as 11g clients, at least for a few weeks until the kinks get worked out. The WLAN setup based on ACU's 11g experience includes the unique idea of using the student's as RF signal attenuators.
At this point, it's not clear if the WLAN instability is an issue of: configuring the access points; the mix of 11g and 11n clients, which triggers 802.11 protection mechanisms adding overhead; the limited channel assignments; a possible iPhone 4 Wi-Fi bug; or some combination of these.
Other colleges and universities so far are reporting no similar problems, though iPhones are popular on nearly all of them. University of Washington, in Seattle, has seen iOS devices soar from 1,400 in the fall of 2007 to 17,000 as of July 2010, according to statistics tracked by David Morton, mobile strategies director for the university, on his blog. So far, he says, they've had no issues, though he's not run any tests on iPhone 4 (UW is just starting its 11n upgrade).
A university in Georgia has upgraded its residence halls to 11n, with classrooms to follow. So far, about two-thirds of students on the "ResNet" use 11n, evenly split between 2.4 and 5GHz, with no problems; the remaining third are still using 11b/g, according to a university RF engineer, who asked not to be identified.
Yet few of these other schools have the kinds of classroom concentrations that ACU does, since ACU courses increasingly are designed to require the use of iPhones in actual class work.
Three years ago, as part of a major mobile learning initiative, ACU decided to equip each successive incoming freshmen class with the current iPhone model, as part of systematic, campus-wide exploration of "mobile learning" course content and teaching methods. The project included a campus-wide 802.11abg WLAN, both 2.4 and 5GHz, based on Alcatel-Lucent products (which are from Aruba Networks). They persuaded AT&T to put a 3G cell site on campus.
When the iPhone 4 was announced, ACU decided to begin upgrading to 11n starting with dorms, the library and some lecture halls. The IT group assumed the new phones would support the 5 GHz band, which is much less crowded and supports 11 vs. three non-overlapping channels. The extra channels mean more channel assignment options, making deployment and management simpler, and give the option for merging two standard 20 MHz channels into one "fatter" 40 MHz channel, to boost throughput, with plenty of channel options remaining.
When they discovered the iPhone 4 only runs on the 2.4 band, "it was kind of deflating," says Arthur Brant, ACU's director, networking services. That meant the iPhone 4 would join its 3G and 3GS brethren on the same crowded frequency, contending for the same three channels. Of 1,000 incoming students, 800 chose iPhone 4 (roughly 300 faculty also have it), the remainder chose the existing 11g-only iPod touch (Apple’s just-announced new iPod touch has 11n).
The new phone's Wi-Fi chip is Broadcom's BCM4329, a low-power, single-chip system designed specifically for mobile devices. It can run in either Wi-Fi band.
It appears to be what's called a single-stream MIMO (multiple input multiple output) implementation, which sounds like a contradiction in terms. (Through a spokesperson, Broadcom declined to explain its products features or the implementation options for device makers like Apple. Apple also declined to comment on its 11n implementation.)
Basically, MIMO breaks up a data stream into two or more substreams, each which is usually but not necessarily associated with a corresponding antenna. These multiple inputs and outputs have a kind of multiplicative impact on data rates and throughput, and the same technology is part of mobile WiMAX and LTE.
Single-stream 11n can't achieve the same effect, but it does realize higher throughput, greater range, and more robust signals compared to 11a and 11g due to other optional features of the 11n specification, including space time block coding, which improves transmission redundancy and reliability. The Broadcom chip supports many of these options. (For more information on 11n optional features see "802.11n set for final approval")
So what kind of 11n performance does iPhone 4 actually get? We asked WLAN test vendor VeriWave to run a simple trial, with its WaveDeploy product suite.
The test was run at VeriWave on a separate but real 802.11n network, based on Cisco gear, designed to behave as a typical enterprise production WLAN. An iPhone 4 was tested for upstream and downstream throughput at 14 locations, sending and receiving TCP traffic, with 1,460-byte frames which are "representative of typical Web traffic," says Eran Karoly, VeriWave's vice president of marketing.
Upstream throughput ranged from 3.9Mbps (at one access point; the next lowest was 9.1Mbps) to 15.6Mbps, for an average of 12.3Mbps. Downstream throughput ranged from 9.4Mbps to nearly 13Mbps, for an average of 10.5Mbps. No significant changes were found using 20MHz or 40MHz channels, according to Karoly. For comparison, an iPod touch using 802.11g was also tested: "the results were more or less the same" as for iPhone 4, he says.
Some schools are reporting better results. Scripps College, in Claremont, Calif., finds that the iPhone 4 and iPad, which also has 11n, are averaging about 28Mbps according to various bandwidth testers, says Jeffrey Sessler, director, Information Technology.
So in terms of data rates and throughput, and depending on lots of variables, often there may not be too much difference between single-stream 11n and a well-designed 11g network. But, 11n still can increase the reliability of the signal, and the range at which a client can sustain a given data rate, both of which can have a big impact on the user's experience.
What does this all mean for ACU's deployment? It means that in the most stressed areas – a big lecture hall packed with 500 freshmen armed with iPhone 4 handsets, 11n has been de-activated.
The campus has two big lecture halls, with stadium seating, one of 6,400 square feet seating 300 students, the other of 7,800 square feet seating 800.
In the smaller of the two halls, ACU previously had refined its 2.4GHz strategy for iPhone on 11g: packing in access points mounted on the side walls, but just 3 feet off the ground; using four channels (so, including one not considered "non-overlapping"); and actually increasing the number of connections allowed per access to 50 from 30; running the radios at 60% of full power. Wireless clients such as laptops, which could run on the 5GHz band, were configure to automatically choose that option.
The low position of the access points means the radio signal has to pass through numerous bodies, deliberately weakening it in an attempt to encourage the iPhone to find and hold onto a very nearby access point. That let the four-channel arrangement work adequately.
Initially, ACU used the same approach for 11n in the larger hall: 16 11n access points total, 10 around the side walls, six in the center (two each at front, middle and rear); mounted 2 feet instead of 3 above the floor; four channels; 50 connections per access point; radio power settings to be determined.
In late August, before freshmen arrived, ACU's Brant staged what he calls a BreakIT session, inviting staff, faculty, some students to the lecture hall for a Wi-Fi stress test. About 160 folks turned up, with a mix of iOS devices, as well as laptop PCs. The access points have two radios, and the plan was to have one radio for 11g and one for 11n clients.They were asked to enter a specific URL on their Web browsers. The effect was dramatic.
11n and 11g clients
"Almost immediately, we started experiencing [network] instability," Brant says. In short order the BreakIT session indeed broke the WLAN. Reviewing the logs, the IT team found that every access point but one rebooted, some of them several times. "What was perplexing was that there didn't appear to be any reason for these access points to have, essentially, crashed," he says. "The number of connections per access point were within reasonable limits, and the bandwidth was within reason."
One likely issue is the mix of high numbers of 11n and 11g clients on a relatively small number of nearby two-radio access points. That situation triggers two problems. One is the increased difficulty in assigning one of the three available non-overlapping channels to each radio, coupled with the fact that the 11g clients hang on to the channel for longer time periods per frame transmission
The other problem is that the 11g clients activate one of several 802.11n protection mechanisms, designed to allow a mix of 11g and 11n clients to associate to the same 11n access point. AirMagnet, for example, has a whitepaper on “Impact of Legacy Devices on 802.11n Networks” (see page 6, "802.11n Preamble Types" and page 7, "802.11n Protection Rules"; Another slightly more recent version, retitled "Ensuring 802.11n and 802.11a/b/g Compatibility" is available on the Airmagnet site, registration required. The complexity of these operating modes and protection mechanisms is noted in this June 2009 blog post at the Certified Wireless Network Professional Web site by Kevin Sandlin.
The key issue: these mechanisms add greatly to the network overhead, reducing throughput for the 11n client and for the WLAN overall. The AirMagnet study found that for 20-MHz channels this "overhead tax" could range from 7% to over 70%, depending in part on the aggregated frame size.
Brant plans to keep sleuthing, tinkering and testing to find the optimal Wi-Fi configuration for dense iPhone 4 deployments.