Editor's note: To view a PDF version of this article, click here.
Final approval of the IEEE 802.11g wireless LAN (WLAN) draft standard is expected by the middle of this year, though the technical aspects are essentially stabilized at this time. As often occurs, manufacturers are bringing product to market in anticipation of final approval of the standard. However, early evaluations have shown that while 802.11g brings much higher data rates to the 2.4-GHz band, data throughput rates drop significantly when a legacy 802.11b client is introduced into an 802.11g network. As a result, controversy is arising.
But the controversy might be somewhat overblown, given the fact that 802.11g incorporates protection mechanisms to mitigate issues that might arise in a mixed-mode environment. Further, the drop in throughput observed is common to any network in which high-rate and low-rate devices share the same medium. To gain accurate insight into the mechanics that underlie 802.11g WLANs, it is necessary to examine its operation under homogeneous (802.11g-only) as well as mixed-mode operation in which an 802.11g client shares an access point (AP) with a legacy 802.11b client.
Lessons from the past
The wireless medium is inherently more challenging than wired, with the effective range falling as the data rate increases. For this reason, 802.11b clients located close to an AP should have no problem connecting at 11 Mbits/second. However, these same devices will fall back to data rates as low as 1 Mbit/s at extreme ranges.
To understand this fall-back process, consider two homogeneous 802.11b-only scenarios. In the first, scenario Ab, an 802.11b AP is streaming large files downstream to two clients, each of which is proximal to the AP and operating at 11 Mbits/s. The second, scenario Bb, is similar to the first except one client has now roamed to the edge of the network and has fallen back to a data rate of 1 Mbit/s to preserve the network connection.
Some words on terminology: An AP and all of the associated clients are collectively referred to as a basic service set (BSS). Within any 802.11 BSS, data can flow downstream (AP to clients), upstream (clients to AP) or in a bidirectional manner (combination of upstream and downstream traffic). Although every data packet is immediately followed by an acknowledgement frame (ACK) in the opposite direction (unless the packet is received in error), the direction of data flow refers to the direction of the data packet. Thus, downstream data flow refers to a packet exchange consisting of a downstream data packet followed by an upstream ACK. When considering the case of downstream traffic flows, it should be kept in mind that the network throughput is determined by the speed at which the AP can access the medium.
When two clients have queued traffic at the AP, packets are on average sent to each client on an alternating basis. Recall that IEEE 802.11 networks use a carrier sense-multiple-access/collision avoidance (CSMA/CA) channel-sharing scheme. As a result, before the AP can transmit it must contend for access to the medium. Figure 1a depicts Scenario Ab in which a series of 1,500-byte packets is being transmitted downstream from the AP to two clients at 11 Mbits/s.
Note that each packet is followed by an ACK and a contention, or "back-off " interval. Importantly, when the AP has queued traffic, data packets are sent to Client No. 1 and Client No. 2 on an alternating basis. Under the conditions depicted in Figure 1a, actual network throughput is about 7.2 Mbits/s. This does not include TCP/IP or application-layer overhead. Therefore, if network throughput is measured by means of a large file transfer, results would be somewhat lower (more about this later).
Now consider Scenario Bb, which is an equally plausible situation (Figure 1b). Assume that Client No. 1 roams out to the edge of network coverage. In this condition, the AP would require a staggering 12,794 microseconds to transmit the same 1,500-byte packet and receive the ensuing ACK from Client No. 1. The time required for the same exchange between the AP and Client No. 2 would remain unchanged at 1,323 microseconds.
It is important to remember that for downstream traffic the AP will alternate transmissions between Client No. 1 and Client No. 2. From Figure 1b it should be clear that Client No. 1 dominates air time due to operation at 1 Mbit/s. As a result, network throughput falls from 7.2 Mbits/s in scenario Ab to just 1.6 Mbits/s in scenario Bb (Figure 2b). This represents a drop in network throughput of 77 percent. In this situation, network throughput is shared equally among clients. On an individual basis, each client will realize 800-kbit/s effective throughput. The drop in throughput is not a result of a network problem. Rather, it is a completely predictable consequence of network operation in a mixed-rate environment. This situation is analogous to the conditions under which most 802.11g network tests are being conducted (downstream transfer of large packets). It is worth emphasizing that the 77 percent drop in network throughput is not due to any interoperability issue, nor has Client No. 2 reduced its data rate.
IEEE 802.11a networks will exhibit similar behavior, giving a 70 percent drop in total throughput (see Fig. 2b).
Downstream 802.11g
Up to this point, we've been discussing some relatively simple scenarios consisting of only two clients receiving downstream traffic from the AP. But network throughput depends on several factors, including the relative amount of time devoted to operating at peak data rates as compared with the amount of time spent operating at lower rates. For situations involving many clients, network throughput depends on the number of high-speed clients relative to the number of slower clients.
Due to the fact that 802.11g network operation involves devices using different waveforms, the situation is slightly more complicated. However, the same basic dynamics are applicable. Let's consider 802.11g network behavior under conditions analogous to those described above, again using two scenarios. In scenario Ag, two 802.11g clients are operating at 54 Mbits/s (Figure. 3a). In scenario Bg, one 802.11g client is operating at 54 Mbits/s while one legacy 802.11b client is operating at 11 Mbits/s (Figure 3b). In both scenarios all clients are performing large downstream file transfers.
For scenario Ag, note that for downstream traffic, packets are again shared equally among the clients. However, due to much higher data rates and a smaller contention period between packet transmissions, network throughput is much higher (30 Mbits/s).
Now consider what occurs if a legacy 802.11b device is substituted for one of the high-speed 802.11g clients. To ensure backward compatibility with legacy 802.11b equipment, a number of changes occur. The most obvious is the use of so-called protection mechanisms. IEEE 802.11g devices can recognize both legacy complementary code-keying (CCK) 802.11b packets as well as orthogonal frequency-division-multiplexed (OFDM) packets. However, legacy 802.11b devices are incapable of recognizing OFDM transmissions. Because CSMA/CA is a listen-before-talk algorithm, this would present a problem unless other measures were employed.
In mixed-mode operation, 802.11g OFDM packets are preceded by a short clear-to-send (CTS) packet transmitted using legacy 802.11b modulation (CCK). The CTS packet conveys the length of time required for the ensuing high-speed OFDM packet and ACK. Legacy clients receiving this information will remain idle for the specified period of time, thereby avoiding collisions with the OFDM packet exchange. Although the CTS protection mechanism results in a marginal increase in network overhead, this effect is dwarfed by the fact that even while operating at 11 Mbits/s, legacy 802.11b clients are much slower on a relative basis.
For scenario Bg depicted in Fig. 3b, overall network throughput is reduced to 11.2 Mbits/s, representing a drop of about 64 percent. The resulting drop is significant, but is actually less severe than what already occurs in 802.11a and 802.11b networks under similar conditions as described previously.
Because the AP alternates transmissions to clients when sending queued traffic downstream, each client receives the same effective throughput regardless of the fact that they are operating at much different data rates. This may explain why some early evaluations of 802.11g equipment indicated that data rates fell back to 802.11b levels while in mixed-mode operation.
Multiple 802.11g clients
Measuring 802.11g network throughput via downstream file transfer is not an unrealistic approach, though it does represent only a single scenario out of the thousands that may be encountered in actual use. In practice, it is possible for dozens of clients to be associated with the AP at any point in time and throughput realized depends on several factors.
One of the most important issues is the number of high-speed 802.11g clients relative to the number of 802.11b clients. A network consisting of a majority of 802.11g devices will operate a larger fraction of time at higher rates and a lower fraction of time at lower rates. The opposite is true if a majority of the devices are slower legacy 802.11b clients. The estimated 802.11g throughput with various mixes of client types is shown (Figure 4). This includes TCP/IP overhead and a 15 percent correction factor for network efficiencies. Note that for downstream traffic, throughput is shared equally among clients.
These figures do not include application-layer overhead, so throughput tests based on large downstream file transfers will render marginally lower results. Bear in mind that these figures represent total network throughput, which is shared among all active clients. The highlighted values along the diagonal axis all represent scenarios involving ten clients. Moving from upper left to lower right, the number of 802.11g clients increases and the number of legacy 802.11b clients decreases. Note also that network throughput increases substantially as the ratio of 802.11g clients increases.
Generalized traffic flow
This discussion has thus far focused entirely on downstream transfer of large files. This is entirely appropriate given that to date, evaluations of 802.11g equipment have largely been conducted in this manner. However, there are some interesting points that come into play if one considers more generalized traffic flow in a mixed-mode 802.11g network.
As described above, when data flow is exclusively in the downstream direction, the AP is the only node that may be contending for medium access. This is not true in general. Under typical operating conditions, multiple nodes may well be attempting to access the network at any given moment.
The issue of slower nodes consuming a disproportionate share of air time was a point of active discussion within the IEEE's Task Group G as the standard was being developed. If every node has a statistically equal chance of accessing the medium, slower nodes will dominate traffic flow. As we have seen, this is precisely what occurs when traffic is exclusively downstream. Since the AP is the only node contending for medium access in this situation, every node having queued traffic will receive an equal number of packets.
Under more typical conditions, traffic flow in a WLAN is bidirectional and in the case of upstream traffic flow multiple nodes are contending for medium access at the same time. The CSMA/CA medium access mechanism is based on the fact that every node monitors the network prior to transmitting. Once the medium becomes idle, each node waits a specified time (50 microseconds for 802.11g mixed-mode networks) and then begins decrementing an internal timer, referred to as the back-off counter.
If another node begins transmitting before the back-off counter reaches zero, the decremented value is retained. When the medium becomes idle again and the client has waited 50 μs, the back-off counter can begin counting down again. Once the back-off counter reaches zero, the client can begin transmission.
The back-off counter is initialized by selecting a random number of slot times from within a predetermined window. For 802.11g mixed-mode operation, a slot time is 20 microseconds in duration. Legacy 802.11b devices normally initialize the back-off counter by selecting a random variable from 0 to 31 corresponding to the number of slot times.
To combat the problem of slower nodes dominating air time, faster 802.11g devices are given a statistical edge to help them access the network more frequently. For 802.11g devices, the normal selection window for the number of slot times is 0 to 15 slot times.
With this in mind, let's re-examine 802.11g mixed-mode operation with two clients in the BSS. This time, however, the direction of data flow is reversed. With data flowing upstream, both the legacy 802.11b client and the higher-speed 802.11g client will end up contending for the medium. However, because the back-off counter is statistically set to a lower number in the 802.11g client, it will get twice as many transmit opportunities (TXOPs) on average as the slower, legacy 802.11b client.
Referring to Figure 5, it can be seen that with approximately a 2:1 advantage in terms of TXOPs, the network does a better overall job of balancing air time between 802.11g and legacy 802.11b clients. As a result, 802.11g mixed-mode throughput in this situation is 12.4 Mbits/s upstream, as compared with only 11.2 Mbits/s downstream (excluding TCP/IP overhead).
Protection imperative
In mixed-mode operation, it is absolutely essential that a CTS packet transmitted using legacy 802.11b modulation precede each high-speed OFDM transmission. In the absence of the CTS packet, legacy 802.11b devices will transmit during an OFDM transmission, thereby colliding with 802.11g traffic. Use of the CTS packet in this manner is referred to as a protection mechanism. Responsible vendors will provide 802.11g products that properly support the use of protection mechanisms during mixed-mode operation.
Some simple tests can be conducted to determine whether protection mechanisms have been properly implemented. One result of poorly implemented mechanisms would be that all 802.11g clients fall back to legacy 802.11b operation in a mixed environment. If all nodes fall back to 802.11b modulation and data rates, total network throughput will be the same regardless of the relative number of 802.11g devices in the network.
A simple test could be conducted by noting total throughput with an 802.11g AP that is streaming downstream traffic to a single 802.11b legacy client to establish a benchmark. If 802.11g devices are falling back to 802.11b rates in mixed-mode operation, overall network throughput will remain essentially constant even as more 802.11g clients are added (recall that overall throughput is shared by each client).
On the other hand, if OFDM packets are transmitted without the use of protection mechanisms in mixed mode, the 802.11b clients will frequently begin transmission during 802.11g packet transmissions. This will almost always result in the loss of both packets. Recall that the 802.11g client would still defer to the 802.11b transmission. A simple test involves upstream transmission of data with several legacy 802.11b clients and an equal number of 802.11g clients on the same AP. If protection mechanisms are not in use and the 802.11g devices continue to transmit at higher rates using the OFDM waveform, throughput for the 802.11b clients will actually be higher than for the 802.11g clients, in spite of the fact that the legacy equipment operates at much lower data rates. This is a strong indication that the product under test does not support protection mechanisms and will perform very poorly in a mixed environment.
Related Article
"802.11g Starts Answering Range Questions"; www.commsdesign.com/design_center/OEG20030114S0008
About the Authors
Menzo Wentick (menzo.wentick@intersil.com) is a system architect for Intersil's Wireless Networking Product Group. As a MAC expert, he participates in the IEEE standards process. He has a master's degree from Twente University of Technology in The Netherlands.
Tim Godfrey (tgodfrey@phy.mail.intersil.com) is a strategic marketing manager at Intersil and also secretary of the IEEE 802.11 standards committee. Godfrey has been involved in wireless networking for more than 10 years and has an EE degree from the University of Kansas.
Jim Zyren (jzyren@intersil.com) is a director of strategic marketing at Intersil and marketing chairman of the Wi-Fi Alliance. He has an MBA from the University of Central Florida and a MSEE degree from the University of Michigan.
