Wireless networking is one of those “agony and ecstasy” kind of technologies.
When it works well, you think you’ll never need another network cable in your house again. But then you try playing a video using wireless, and it works well enough until someone else decides to go web surfing. Then you try to copy a large file and realize how much slower it is than 100Base-T, never mind gigabit Ethernet. Finally you begin to discover the “dead zones” in your house where radio signals go in, but they don’t come out. The fundamental hurdles for wireless—throughput, quality of service (QoS), and coverage—are numerous. As we reported last fall, two large industry “Pre-N” groups, the
We got one of the very first review units LinkSys made available to the tech press and put it through its paces. What we found was very encouraging: 30+Mbits/sec throughput at a distance of 25 feet through two walls. But we were also able to get 30+Mbits/sec of throughput and very smooth playback of streamed video content…at the same time. Does it sound like we’ve been smoking something other than cigars? Well read on, because MIMO-based wireless can actually do this, and it’s going to change the face of wireless home networking as we know it.
Wireless pioneers dating all the way back to Marconi have had to deal with an issue called multi-path, which occurs when a transmitter’s signal bounces off multiple objects, with multiple reflected signals arriving at the receiver at slightly different times, out of phase. The challenge for the receiver is to figure out which signal is the “real one,” and which are effectively noise. If the different signal arrivals get far enough out of phase with one another, they can effectively cancel each other out, causing signal loss. In the wireless networking world, this means dropped packets, or even a dropped connection.
In the 1990’s, two Stanford researchers, Greg Raleigh and VK Jones (both of whom co-founded Airgo Networks) developed a radical notion: Make multipath your friend. Using multiple transmitters and receivers, and antenna arrays for each, the researchers developed statistically-based algorithms that can treat the multipath signals effectively as different signals, and multiplex data over them to increase both range and throughput. But here’s the kicker: Unlike channel-bonding techniques, MIMO can do its multiplexing within the bandwidth of a single 802.11 channel.
This approach is friendlier to non-proprietary solutions that may enter the network as visiting clients. What’s more, MIMO can extend the operating range and throughput of non-MIMO clients, though MIMO-aware clients will benefit the most. Also, in “noisy” environments where multiple APs using a 2.4GHz carrier frequency are present, you still have the “other two” channels available for other networks to use. Recall that 802.11b and 802.11g use a 2.4GHz carrier frequency. While every AP made will tell you there are 11 channels available, there are only three truly discrete channels whose bandwidths don’t overlap one another.
Linksys has coined the term SRX (speed/range extender) to describe its MIMO-based solutions. This SRX acronym is found on Linksys products using the Airgo chipset. Linksys’ new WRT54GX AP brings with it most of the usual trappings you’d expect in a home wireless router: DMZ, DHCP, port triggering, WPA security, and a four-port switch. But it also has some things you haven’t seen, including three antennas and an option to enable 802.11e Quality of Service, which gives some network traffic (audio/video traffic such as streaming video and VOIP) priority over less time-sensitive traffic such as web surfing and P2P/FTP downloads.
There has been some controversy as to what MIMO actually means. Airgo for instance has adopted a “True MIMO” marketing campaign, defining MIMO to mean multiple radios and multiple antennas. Meanwhile Atheros has labeled its channel-bonding technology “MIMO” as well, and there has been some squabbling about whose MIMO is the real deal. Semantics aside, the Airgo chipset has the advantage of using a single carrier frequency, whereas the Atheros solution requires two 802.11g channels, and multiplexes its signals across them to achieve its higher throughput. For this reason, the Atheros chipset may have a harder time playing nice with other nearby APs, as they could fight for clear air on the same carrier frequencies. On the flip side, the Airgo technology, which uses multiple radios and antennas, will raise the bill for the cost of materials and networking hardware that uses it.
Physically, the AP has mounting holes so you can hang it on a wall, although if the WAN port and all four 100Base-T LAN ports are in use, that could make for quite a wire jumble going into the device. It also includes a base that lets the unit stand vertically, or you can lay it on its side.
We hit the WRT54GX with a number of tests, concentrating primarily on throughput and video streaming performance at a distance, both on a quiet segment, and with a heavy throughput load of low-priority traffic fighting for available bandwidth as well. We connected a single machine to it via 100Base-T that acted as our server, and had the following load-out:
- Pentium 4 3.2GHz CPU
- Intel 865PERL motherboard with the Intel 865 chipset
- 1GB of system memory
- 40GB ATA hard-drive
- Windows XP SP2 (DirectX 9.0c)
Linksys isn’t yet shipping PCI- or USB-based network adapters, and it doesn’t yet have a MIMO-based repeater or bridge product on the market. As a result, we used two similarly appointed laptops as our clients running with WPC54GX CardBus network adapters in them.
| ASUS M6S00N | Dell Precision M60 | |
| CPU: | 1.6GHz Pentium-M | 1.7GHz Pentium-M |
| Chipset: | Intel 855M | Intel 855M |
| System RAM: | 512MB | 1GB |
| OS: | Windows XP Pro with SP2 (DirectX 9.0c) | Windows XP Pro with SP2 (DirectX 9.0c) |
To gauge throughput performance, we used Iperf 1.7.0, a freely available network bandwidth measurement tool developed by the Distributed Application Support Team of the National Laboratory for Applied Network Research (NLANR). It can measure throughput of both TCP and UDP traffic. We did our testing using the TCP protocol with a 128KB TCP Window Size.
To “let the ponies run,” we disabled all types of security on the AP and in the Windows Firewall settings of both the server and the clients. At each spot, we did a total of 12 ten-second tests in sets of three. After each set, we rotated the laptop 90 degrees. Some networking companies use a turntable where they place a laptop for testing to completely notch out nulls, or dead zones. While this makes sense from a pure methodology standpoint, it makes less sense from a usage model perspective, unless of course you sit on a spinning potter’s wheel when using your laptop.
We did these tests at the following distances:
- Point-Blank: The laptop was about three feet from the AP (no walls obstructing)
- 10 feet: no walls obstructing
- 25 feet: one wall obstructing
- 50 feet: two walls obstructing
We gathered a baseline set of numbers using only the MIMO-based hardware, and then gathered a second set of test data using an ASUS Centrino-based (802.11g) notebook to gauge a non-MIMO client’s performance when associated with a MIMO AP.
Next, we fired up VideoLAN, an open-source video player and streaming video server application developed at the École Centrale Paris. We configured the application to point-cast a single UDP video stream from our server to a laptop client. Our video content consisted of:
- A VOB (DVD Video Object) file ripped from the Fifth Element (bit-rate was roughly 5-7Mbits/sec)
- An 18Mbits/sec MPEG-2 transport stream video clip of an HDTV program
This second file is obviously a huge load to place on any network, but we wanted to really stress the wireless segment. If it could handle this, then streams with lower bit-rates wouldn’t be a problem. We also successfully did multiple point-casts to the same client using two different VOB files from The Fifth Element.
We also mapped a network drive using Windows File Sharing to the server system, and played back these two clips using Windows Media Player, and also played back several WMV-HD 720p and 1080i video clips whose bit rate ranged from 6-8Mbits/sec. However, we had no streaming server here to manage the client/server connection, so this was a pure “best effort” scenario, where the video playback was susceptible to stuttering and dropped frames owing to other network traffic.
We conducted our first tests on a “quiet” segment with no other traffic present. We then used one laptop system to run the Iperf throughput test in an infinite loop to generate a heavy “traffic cloud,” and repeated our video playback tests, both using the VideoLAN streaming server, and using Windows File Sharing. The goal in the VideoLAN test was to see if streamed video could maintain smooth playback, even though VideoLAN is not 802.11e-aware. The tests using WMV-HD content were worst-case scenario tests to see if the WRT54GX could maintain smooth playback of non-streamed video over our traffic cloud.
Note that we conducted these tests in our offices, rather than at home. The decision was purposeful, since we wanted to see how the MIMO-based technology would behave in a radio-hostile environment. We used NetStumbler to survey the site and find the quietest channel, which in this case was Channel 11. We primarily performed inspection tests during the day, but conducted our throughput testing after 6pm, so that traffic from adjacent offices would be at a minimum (mostly announce beacons).
At a high level, what we found is that the MIMO-based WRT54GX delivered some very impressive performance, even in our radio-hostile test environment. We did find however that its performance is asymmetrical: Upstream bandwidth is about 33% better than downstream bandwidth. This is obviously good news if you’re copying files up from your laptop to another machine, but since most traffic is coming downstream, we wish the asymmetry was the other way around, with downstream traffic moving faster.
We set the following parameters in the AP’s admin control panels:
CTS disabled: The Clear to Send (CTS) feature is to allow 802.11b clients onto an 802.11g network. Enabling it does allow 11b clients onto the segment, but will gate the performance of 11g clients. Since we weren’t testing with any 11b hardware, we disabled this feature.
Network Density Medium: This feature “shapes” the radio coverage, giving you either better coverage with lower signal strength, or a smaller coverage area with better signal strength. The default value here is Low, but after doing some initial inspection tests, we found we got our best throughput/coverage results using the Medium setting.
CORRECTION: Airgo informed us that we improperly explained what the Immediate ACKS setting does in the Linksys WRT54GX router. The description that follows is the correct description.
Immediate ACKS: Controls MAC-level ACKs used by the 802.11 MAC to acknowledge each 802.11 packet or fragment transferred between the AP/router and the Station. When ACKs are not received (i.e. a packet is lost over the air), the MAC automatically retransmits the packets. The two other settings are Burst ACK or No ACKs, which will reduce aggregate traffic on the segment, but should only be used in radio-friendly environments where there is minimal traffic from other APs. In addition, the Airgo chipset relies on ACK packet data to help characterize the multipath environment, so we used the default Immediate ACKS setting.
At 10 feet, the MIMO-to-MIMO connection is moving things along very well, with an average bandwidth of about 30Mbits/sec, whereas the MIMO-to-G connection is down around 20Mbits/sec. Moving to 25 feet, we see that the MIMO connection maintains roughly the same performance, whereas the MIMO-to-G connection is cut by more than half. At 50 feet, both have trailed off considerably, but the MIMO is still at around 12Mbits/sec.
Here we see the MIMO-to-MIMO connection pull well ahead, with its peak bandwidth of about 39Mbits/sec coming at a distance of 25 feet. Both connections drop off quite a bit going out to 50 feet, and although the MIMO connection takes a more than 50% hit, it’s still about half again as fast as the MIMO-G connection.
Next we look at the two connection types individually to gauge upstream versus downstream performance.
You can see that the MIMO connection’s upstream bandwidth clearly outpaces its downstream, by as much as 33% at the 25-foot test distance. We’d rather see the performance to be more symmetrical, or favor downstream bandwidth. But even so, both are performing well and delivering considerably more performance than 802.11g products, which typically get into the 20-22Mbits/sec range. We also note that the curves have a similar shape, so that as performance degrades at the 50-foot test distance, the upstream and downstream bandwidths become equal. At this distance, you still have about 15Mbits/sec of usable bandwidth, more than enough for most video playback.
In this test scenario, the upstream and downstream bandwidths are more comparable, although downstream bandwidth takes a bigger hit sooner (at 25 feet), and is down around 8Mbits/sec. Upstream bandwidth degrades more gracefully, settling to about 10Mbits/sec at 50 feet. This performance, while certainly less impressive than the MIMO-only connection, is still pretty good, though you’d be hard-pressed to play back video over this link at 25 feet. But for web surfing and file downloads, your downstream bandwidth is still well in excess of your broadband connection, so the wireless connection isn’t your bottleneck.
The 802.11e spec was finally ratified last fall, bringing Quality of Service (QoS) to wireless networking. But it languished in committee for quite some time, so the Wi-Fi Alliance took a subset of the IEEE 802.11e WLAN QoS draft standard, and created the Wi-Fi Multimedia (WMM) specification. Linksys’ WRT54GX supports both 802.11e and WMM QoS, and we tested with 802.11e enabled.
The slide below from the Wi-Fi Alliance illustrates the problem of traffic prioritization. The red line represents the bandwidth allocated for video, and when traffic conditions change, the video traffic will retain its allocated bandwidth, whereas non-priority traffic shares the remaining bandwidth. In a purely best-effort environment, as aggregate traffic approaches its ceiling, all streams in flight, media or otherwise, suffer the effect, resulting in stuttering playback of video and audio.
However, we have yet to find any applications that explicitly use the 802.11e QoS scheme, so we’re still left with best-effort where all traffic is given equal prioritization. It turns out that best-effort, while not ideal, does work pretty well for streaming media, provided you have sufficient headroom. It’s when traffic on your wireless segment approaches its limit that video and audio streams will get into trouble, because these applications are being starved of the data they need to arrive at a specific time. And even if they buffer, the receive buffers can wind up being starved as well.
We used VideoLAN to UDP point-cast single and multiple streams to our wireless client. We first ran our video streams over a quiet network segment (no other traffic) and saw smooth video playback of all content: the DVD clip from The Fifth Element, WMV-HD clips, and even the monster 18Mbits/sec HDTV MPEG-2 transport stream. With the exception of the huge HDTV stream, we were able to walk out to about 70 feet with three walls obstructing our client-to-AP signal path before we saw any video breakup.
This was pretty cool, but then it got better: We set up a second wireless client to run an Iperf upstream/downstream bandwidth test in an infinite loop to generate a heavy traffic load, and then we went back and tried to play our video clips. Here we still saw smooth playback at around 70 feet. Meanwhile Iperf running on our other client about 10 feet from the AP was reporting upstream/downstream bandwidth of 30Mbits/sec and 22Mbits/sec, respectively.
Media likes MIMO. Plenty of headroom with the presence of QoS for media applications means the WRT54GX will allow streaming video to peacefully co-exist with other traffic on your wireless network. Wireless being what it is however, we’re going to add the caveat that your mileage will vary, depending on your home’s construction, and the physical location of the AP and the clients.
Unfortunately, as of press time, Linksys is not yet shipping PCI- or USB-based network adapters, so client machines will have to be laptops for now. You could try to fit one of these cards into a CardBus-PCI adapter, but it’s unclear how well this would perform. According to Linksys, PCI-and USB-based products using the Airgo chipset will ship some time late this spring or in early summer. A company official added that Linksys hasn’t decided whether it will do a bridge solution that incorporates the Airgo technology.
See more of our
The AP itself, with its three antennae, is a bit unsightly, and you might want to tuck it somewhere out of sight. Aesthetics aside, this is a pretty slick AP from Linksys that delivers a lot of performance, but will cost you about another $100 over other current-generation 802.11g APs. The WRT54GX plus two network adapters will run you about $260. However, because it plays nice with 802.11g, you can upgrade only the AP now, and upgrade your network adapters when they become available. If you crave speed, coverage, and headroom for streaming media, then the WRT54GX will give you what you need.
| Product: | Linksys WRT54GX Access Point and WPC54G CardBus Network Adapter |
| Company: | Linksys |
| Price: | $170 (street) for WRT54GX access point; $45 (street) for WPC54G network adapter. |
| Pros: | Fast, good coverage, and media-friendly; works well (and can improve performance) with 802.11g; power supply is a line-lump, and its plug doesn’t eat an entire AC socket. |
| Cons: | Expensive, and not the prettiest thing in the world; no PCI or USB-based adapters available yet; no bridges available yet. |
| Summary: | A solid-performing though pricey access point that dramatically improves performance, and enables good streaming media performance, even at some distance. |
| Rating: | |