He-Ha: Netgear's 802.11a offerings

by Fred Coffman

Written by: Rob Flickenger and Fred Coffman

Well folks, the future is here. 802.11a access points have been shipping for a couple of months in the U.S., and are available online and in stores from vendors we know and love (including Proxim, Agere, Linksys, D-Link, SMC, Intel, and Netgear). Widely regarded as "Wi-Fi's better cousin", the 802.11a standard is poised to answer
many of the shortcomings of 802.11b: Higher data rates (54 vs. 11Mbps), better security, and implementation in the largely uncrowded 5.8GHz UNI band (far away from the overcrowded 2.4GHz ISM trash band).

With much excitement in the air, we finally had an opportunity to try out Netgear's offering to the 802.11a market: the HE102 Access point and HA501 client card. As avid users of 802.11b (aka Wi-Fi) gear, the specs had us salivating. 54Mbps, and up to
72Mbps in "turbo mode"! Little or no chance of noise in the band! But what was that about range?

These are the rated abilities of the HE102 access point and HA501 client card, straight from the Netgear website:

Closed Environment:

@ 54 Mbps 60 ft. (18.3m)

@ 6 Mbps 300 ft. (91.4m)

Open Environment:

@ 54 Mbps 100 ft. (30.5m)

@ 6 Mbps 1200 ft. (366m)

Fascinating. 300 feet isn't much to play with indoors (even Agere's gear claims at least 375ft indoors, albeit at 1Mbps). And 60 feet really isn't far at all in a corporate setting. Still, we thought it worth trying out firsthand.

Unfortunately, at least in our informal testing, even Netgear's meager listed specs seem particularly generous. Our test machine was a Thinkpad 770Z running Win2K (as Linux drivers aren't available for their card yet.) We took signal readings at a few set distances from the AP, and tried some throughput testing using ttcp and wsttcp.

We started with the best possible case: with the AP about 6 feet away, sitting on the same table. With nothing but air between the two, we saw signal strength in the high 90%s (this is comparable to most radio ethernet gear; you will almost never see a signal of 100% on the dumbed-down client meters.) Interestingly, the client monitor claimed a transmit rate of only 48Mbps, and a receive of 24Mbps (if 99% isn't enough signal to get the full 54Mbps, then exactly what is?)

Here are the results of our first throughput test:

buflen=8192, nbuf=2048, align=16384/0, port=5001 tcp ->
16777216 bytes in 6.46 real sec = 2536.62 KB/sec

buflen=8192, nbuf=2048, align=16384/0, port=5001 tcp <-
16777216 bytes in 6.39 real sec = 2564.41 KB/sec

This shows about 20.2Mbps in each direction (despite the driver claiming 48Mbps upload.) Interesting.

By way of comparison, here's a throughput test in the same conditions using our 802.11b network (Lucent wireless card to Lucent AP1000, no WEP):

buflen=8192, nbuf=2048, align=16384/0, port=5001 tcp ->
16777216 bytes in 38.80 real seconds = 422.23 KB/sec

buflen=8192, nbuf=2048, align=16384/0, port=5001 tcp <-
16777216 bytes in 39.46 real seconds = 415.19 KB/sec

So in the same conditions (only a few feet from the access point), the
802.11a gear roundly trounced 802.11b (20.2Mbps vs. 3.4Mbps.)

Moving to a distance of about 12 feet, behind a sheetrock-and-window wall (the next office over), we observed a signal strength of 88-89%. The client now reported 54Mbps transmit, 24Mbps receive-- actually higher than when the meter was at 99%. A couple of runs of ttcp still showed roughly 20Mbps upload and 20Mbps download.

Next stop, about 50 feet away, with two walls between the client and access point. The signal strength now read somewhere between 58% and 62%, with transmit and receive both rated at 24Mbps. Running ttcp showed the expected results, with uploads and downloads both holding at about 20Mbps.

The final stop was just about 10 feet further, but around a corner. The
signal fell off sharply (only about 4% to 10%) with transmit and receive
both claiming 12Mbps. But this time, ttcp only showed about 39Kbps upload,
and 5Kbps down! It was obvious that we had hit the edge of usable range at
only 60 feet away, through about four walls made of sheetrock and glass. Our
802.11b network, in the same setting (60 feet and a couple of intervening
walls), held at a steady 3.4Mbps, just as it did when very close to the AP.

By way of comparison, we use 802.11b equipment at O'Reilly, and are managing to just about cover two very large 3-story buildings with only three access points (all Orinoco AP-1000s, using small range-extender antennas and two radios per AP). To cover the same area with HE102s would easily require at least four or five access points per floor, per building, according to our initial tests. And speaking of range extenders, there was something obviously missing from Netgear's client card and AP-- external antenna connectors! The two omnidirectional antennas on the HA102 are permanently attached, and the HE501 client card has a moulded plastic end cap, with no connector at all. This leaves little hope for being able to create a more usable network with this gear.

The overall impression of our experience was probably best described as
bouncy. There seemed to be very little room for middle ground, both in
signal strength and throughput. Either we had terrific signal and very good
throughput, or no signal at all, and dismal performace (even lower than
802.11b at 1Mbps). People walking between the client and the AP would make
the signal drop off completely for a moment, only to return a moment later
at a solid reading. And the client definitely liked having clean
line-of-sight with the AP (turning a corner meant the difference between a
usable link and something slower than dialup.)

We will definitely take a look at more 802.11a gear as we can, but this one
is going back to the store. I still have high hopes for outdoor 802.11a, as
long distance line-of-sight shots will be very interesting at 54+ Mbps.
Hopefully other vendors have come up with ways of dealing with the limits of
line of sight at 5.8GHz. Unless they have, (and according to href="http://www.nwc.com/1313/1313f3.html">some reports, they haven't
quite yet), 802.11a will have a tough time competing with Wi-Fi for indoor use.