Tunable 802.11 Parameters

802.11 provides a number of knobs that can be used to wring extra performance out of a network in dire situations, although a great deal of experimentation is probably required to find the values that work best.

Radio Management

As with other types of wireless networks, radio bandwidth is a precious resource on an 802.11 network. Radio spectrum is constrained by regulatory authority and cannot be easily enlarged. Several parameters allow you to optimize your network's use of the radio resource.

Beacon interval

Beacon frames serve several fundamental purposes in an infrastructure network. At the most basic level, Beacon frames define the coverage area of a basic service set (BSS). All communication in infrastructure networks is through an access point, even if the frame is sent between two stations in the same BSS. Access points are stationary, which means that the distance a Beacon frame can travel reliably won't vary over time.[*] Stations monitor Beacon frames to determine which Extended Service Sets (ESSs) offer coverage at their physical location and use the received signal strength to monitor the signal quality.

[*] Multipath interference may cause odd time-dependent interference patterns. A particular spot may be within range of an access point at one instant in time and subject to multipath fading seconds later. However, such a spot has marginal coverage and should not be considered a part of a basic service area. 802.16 (WiMax) is also promising for mobile data connectivity.

Transmitting Beacon frames, however, eats up radio capacity. Decreasing the Beacon interval makes passive scanning more reliable and faster because Beacon frames announce the network to the radio link more frequently. Smaller Beacon intervals may also make mobility more effective by increasing the coverage information available to mobile nodes. Rapidly moving nodes benefit from more frequent Beacon frames because they can update signal strength information more often.[] Increasing the Beacon interval indirectly increases the power-saving capability of attached nodes by altering the listen interval and the DTIM interval, both of which are discussed in the section "Tuning Power Management later in this chapter. Increasing the Beacon interval may add an incremental amount of throughput by decreasing contention for the medium. Time occupied by Beacon frames is time that can't be used for transmitting data. If you use a network built on virtual APs, each network requires its own set of Beacons, and the overhead can multiply rapidly.

[images/ent/U2020.GIF border=0>] 802.11 is not designed to support high-speed mobility, though. Cellular-based, wide-area technologies are more effective.

RTS threshold

802.11 includes the RTS/CTS clearing procedure to help with large frames. Any frame larger than the RTS threshold must be cleared for departure from the antenna by transmission of an RTS and reception of a CTS from the target. RTS/CTS exists to combat interference from so-called hidden nodes. The RTS/CTS exchange minimizes interference from hidden nodes by informing all stations in the immediate area that a frame exchange is about to take place. The standard specifies that the RTS threshold should be set to 2,347 bytes. If network throughput is slow or there are high numbers of frame retransmissions, enable RTS clearing by decreasing the RTS threshold.

In Chapter 3, I said that a hidden node was a node that wasn't visible to all the stations on the network. Under what sorts of situations can you expect hidden nodes? Just about any, really. In almost any network, there are bound to be places where two nodes can reach the access point but not each other. Let's consider the simplest network imaginable: one access point in the middle of a large field with nothing to cause reflections or otherwise obstruct the signal. Take one mobile station, start at the access point, and move east until the signal degrades so that communication is just barely possible. Now take another station and move west. Both stations can communicate with the access point, but they are invisible to each other.

The previous thought experiment should convince you that invisible nodes are a fact of life, but not very common. Hidden nodes tend to occur in harsh radio environments where stations are very far from the nearest AP. Now that wireless networks have become popular, the AP density is high enough to place a practical limit on the number of hidden nodes.

Fragmentation threshold

MAC-layer fragmentation is controlled by the fragmentation threshold variable. Any frames longer than the fragmentation threshold are sliced into smaller units for transmission. The default fragmentation threshold is the smaller of 2,346 or the maximum MAC frame length permitted by the physical layer. However, the RF-based physical layers usually have a maximum MAC frame length of 4,096 bytes, so this parameter generally defaults to 2,346. The common value immediately implies that fragmentation and RTS/CTS clearing are often used in tandem.

In environments with severe interference, encouraging fragmentation by decreasing this threshold may improve the effective throughput. When single fragments are lost, only the lost fragment must be retransmitted. By definition, the lost fragment is shorter than the entire frame and thus takes a shorter amount of time to transmit. Setting this threshold is a delicate balancing act. If it is decreased too much, the effective throughput falls because of the additional time required to acknowledge each fragment. Likewise, setting this parameter too high may decrease effective throughput by allowing large frames to be corrupted, thus increasing the retransmission load on the radio channel.

Retry limits

Every station in a network has two retry limits associated with it. A retry limit is the number of times a station will attempt to retransmit a frame before discarding it. The long retry limit, which applies to frames longer than the RTS threshold, is set to four by default. A frame requiring RTS/CTS clearing is retransmitted four times before it is discarded and reported to higher-level protocols. The short retry limit, which applies to frames shorter than the RTS threshold, is set to seven by default.

Decreasing the retry limit reduces the necessary buffer space on the local system. If frames expire quicker, expired frames can be discarded, and the memory can be reclaimed quicker. Increasing the retry limits may decrease throughput due to interactions with higher-layer protocols. When TCP segments are lost, well-behaved TCP implementations perform a slow start. Longer retry limits may increase the amount of time it takes to declare a segment lost.

Tuning Power Management

From the outset, 802.11 was designed for mobile devices. To be useful, though, mobile devices cannot be constrained by a power cord, so they usually rely on an internal battery. 802.11 includes a number of parameters that allow stations to save power, although powersaving is accomplished at the expense of the throughput or latency to the station.

Listen interval

When stations associate with an access point, one of the parameters specified is the listen interval, which is the number of Beacon intervals between instances when the station wakes up to received buffered traffic. Longer listen intervals enable a station to power down the transceiver for long periods. Long power-downs save a great deal of power and can dramatically extend battery life. Each station may have its own listen interval.

Lengthening the listen interval has two drawbacks. Access points must buffer frames for sleeping stations, so a long listen interval may require more packet buffer space on the access point. Large numbers of clients with long listen intervals may overwhelm the limited buffer space in access point hardware. Second, increasing the listen interval delays frame delivery. If a station is sleeping when its access point receives a frame, the frame must be buffered until the sleeping station is awake. After powering up, the station must receive a Beacon frame advertising the buffered frame and send a PS-Poll to retrieve the frame. This buffering and retrieval process can delay the time the frame spends in transit. Whether this is acceptable depends on the traffic requirements. For asynchronous communications such as email, lengthening the listen interval isn't likely to be a problem. But in other applications that require synchronous, time-sensitive communications (such as securities market data feeds today or an IP phone with an 802.11 interface in the future), a longer interval might not be acceptable. Certain applications may also have trouble with the increased latency. Database applications, in particular, are significantly affected by increased latency. A task group is working on MAC enhancements to provide quality of service for transmissions on 802.11 networks, but no standard has emerged yet.

DTIM Period

The DTIM period is a parameter associated with an infrastructure network, shared by all nodes associated with an access point. It is configured by the access point administrator and advertised in Beacon frames. All Beacon frames include a traffic indication map (TIM) to describe any buffered frames. Unicast frames buffered for individual stations are delivered in response to a query from the station. This polled approach is not suitable for multicast and broadcast frames, though, because it takes too much capacity to transmit multicast and broadcast frames multiple times. Instead of the polled approach, broadcast and multicast frames are delivered after every Delivery TIM (DTIM).

Changing the DTIM has the same effect as changing the listen interval. (That should not be a surprise, given that the DTIM acts like the listen interval for broadcast and multicast frames.) Increasing the DTIM allows mobile stations to conserve power more effectively at the cost of buffer space in the access point and delays in the reception. Before increasing the DTIM, be sure that all applications can handle the increased delay and that broadcasts and multicasts are not used to distribute data to all stations synchronously. If the application uses broadcast or multicast frames to ensure that all mobile stations receive the same blob of data simultaneously, as would be the case with a real-time data feed, increasing the DTIM will likely have adverse effects.

ATIM window

In an infrastructure network, access points provide most of the powersaving support functions. In an independent or ad hoc 802.11 network, many of those functions move into the network interface driver. In ad hoc networks, stations are required to power up for every Beacon transmission and remain powered up for the duration of the Announcement TIM (ATIM) window, which is measured in time units (TUs).

Decreasing the ATIM window increases the power savings because the required power-on time for the mobile stations is reduced. Stations can power down quickly and are not required to be active during a large fraction of the time between Beacons. Increasing the ATIM window increases the probability a powersaving station will be awake when a second station has a frame. Service quality is increased, and the required buffer space is potentially smaller.

Decreasing or disabling the ATIM window would probably have the same effect on synchronous or real-time applications as increasing the DTIM timer on an infrastructure networkthat is, it is likely to cause problems with less reliable communications or applications that depend on real-time data. One of the most obvious examples of a real-time application of ad hoc networking is gaming, but it is far more likely that ad hoc gaming networks would be tuned for low-delay and high-throughput than for low-power operation.

Timing Operations

Timing is a key component of 802.11 network operations. Several management operations require multistep processes, and each has its own timer.

Scan timing

To determine which network to join, a station must first scan for available networks. Some products expose timers to allow customization of the scanning process. In products that expose timers, both an active scan timer and a passive scan timer may be exposed. The active timer is the amount of time, in TUs, that a station waits after sending a Probe Request frame to solicit an active response from access points in the area. Passive scanning is simply listening for Beacon frames and can take place on several radio channels; the passive scan timer specifies the amount of time the receiver spends listening on each channel before switching to the next.

Timers related to joining the network

Once a station has located an infrastructure network to join, it authenticates to an access point and associates with it. Each of these operations has a timeout associated with it. The authentication timeout is reset at each stage of the authentication process; if any step of the process exceeds the timeout, authentication fails. On busy networks, the timeout may need to be increased. The association timeout serves a similar function in the association process.

Dwell time (frequency-hopping networks only)

The amount of time that an FH PHY spends on a single hop channel is called the dwell time. It is set by local regulatory authorities and is generally not tunable, except by changing the network card driver to a different regulatory domain.

Summary of Tunable Parameters

For quick reference, Table 25-1 summarizes the contents of this chapter, including the effect of changing each of the tuning parameters.

Table 25-1. Summary of common tunable parameters

Parameter	Meaning and units	Effect when decreased	Effect when increased
Beacon	Number of TUs between transmission of Beacon frames.	Passive scans complete more quickly, and mobile stations may be able to move more rapidly while maintaining network connectivity.	Small increase in available radio capacity and throughput and increased battery life.
RTS Threshold	Frames larger than the threshold are preceded by RTS/CTS exchange.	Greater effective throughput if there are a large number of hidden node situations .	Maximum theoretical throughput is increased, but an improvement will be realized only if there is no interference.
Fragmentation Threshold	Frames larger than the threshold are transmitted using the fragmentation procedure.	Interference corrupts only fragments, not whole frames, so effective throughput may increase.	Increases throughput in noise-free areas by reducing fragmentation acknowledgment overhead.
Long Retry Limit	Number of retransmission attempts for frames longer than the RTS threshold.	Frames are discarded more quickly, so buffer space requirement is lower.	Retransmitting up to the limit takes longer and may cause TCP to throttle back on the data rate.
Short Retry Limit	Number of retransmission attempts for frames shorter than the RTS threshold.	Same as long retry limit.	Same as long retry limit.
Listen Interval	Number of Beacon intervals between awakenings of powersaving stations.	Latency of unicast frames to station is reduced. Also reduces buffer load on access points.	Power savings are increased by keeping transceiver powered off for a larger fraction of the time.
DTIM Window	Number of Beacon intervals between DTIM transmissions (applies only to infrastructure networks).	Latency of multicast and broadcast data to powersaving stations is reduced. Also reduces buffer load on access points.	Power savings are increased by keeping transceiver powered off for a larger fraction of the time.
ATIM Window	Amount of time each station remains awake after a Beacon transmission in an independent network.	Increases power savings by allowing mobile stations to power down more quickly after Beacon transmission.	Latency to powersaving stations is reduced, and the buffer load may be decreased for other stations in the network.
Active Scan Timer	Amount of time a station waits after sending a Probe Response frame to receive a response.	Station moves quickly in its scan.	Scan takes longer but is more likely to succeed.
Passive Scan Timer	Amount of time a station monitors a channel looking for a signal.	Station may not find the intended network if the scan is too short.	Scan takes longer but is more likely to succeed.
Authentication	Maximum amount of time between successive frames in authentication sequence.	Authentications must proceed faster; if the timeout is too low, there may be more retries.	No significant effect.
Association Timeout	Maximum amount of time between successive frames in association sequence.	Associations must proceed faster; if the timeout is too low, there may be more retries.	No significant effect.