Wi-Fi Overhead, Part 1: Sources of OverheadBy CWNP On 04/27/2011 - 30 Comments
Radio communication requires overhead. Network protocols require overhead. Unfortunately, wireless network protocols, like Wi-Fi, are loaded with overhead. Some amount of overhead is necessary for effective communications and interoperability; however, there are also times when overhead is unnecessary. Proper network design and deployment can minimize this overhead and improve network performance. This article kicks off a two-part post that will identify the sources of overhead (part 1) on WLANs and then provide some recommendations for reducing it (part 2).
Interference — In a very generic sense, all sources of interference (non-802.11 and 802.11) create overhead. 802.11 devices must perform a clear channel assessment (CCA) to determine whether the wireless medium is busy or idle. Whenever the medium is busy, Wi-Fi stations twiddle their thumbs and wait. Non-802.11 sources of RF (above a certain threshold) can be thought of as a source of overhead.
Interframe Spaces — Before every Wi-Fi transmission, there is an idle period on the medium. This idle period is called an interframe space (IFS). There are multiple different interframe space lengths; the purpose of an IFS is to regulate conversation flow and provide priority for certain types of transmissions.
Random Backoff —When 802.11 devices are contending for access to the wireless medium, they use a backoff algorithm that randomizes access to the medium. This process ultimately reduces collisions and is one way to achieve QoS prioritization. The random backoff time represents a number of “slots” (or, periods of time) that the wireless medium must be idle.
PHY Signaling — Radio communication peers must perform some type of synchronization for reliable reception of frames. The PHY Preamble is a series of bits used to perform this function. The PLCP Header follows the preamble and communicates the attributes of the following frame to the receiver so that the receiver knows how to process the data. In other words, the preamble and PLCP header are collectively the same as if a person were to say: “OK, I’m getting ready to say something. It’s going to take me 20 seconds to say it and I’m going to say it pretty quickly, so listen up. Here goes.”
MAC Header -- A complex protocol like Wi-Fi requires that stations can coordinate their operation. The MAC header is used to coordinate supported (or unsupported) features and functions. It is necessary, but it is still overhead. In data frames, the MAC header is so small (and usually transmitted at a high data rate) that it barely qualifies as overhead. In other frames, such as 802.11n beacons, the “overhead” designation is much more applicable.
Guard Intervals — Between each 802.11 symbol, there must be a quiet period, called a guard interval, on the medium to allow previous symbols to “settle.” Without this quiet period, a symbol may interfere with the previous symbol (inter-symbol interference). The normal guard interval setting for Wi-Fi is 800 ns.
Acknowledgements — Since wireless communication is inherently unreliable and lossy, many frame types require acknowledgement. Acks are overhead in and of themselves, and they also require an additional interframe space (i.e. SIFS).
Fragmentation (and small payloads) — While frames with small payloads are not an actual source of overhead, they are often an inefficient use of the medium, making the overhead problem more apparent. A similar problem exists when organizations enable frame fragmentation to attempt to reduce collisions (smaller frames are less likely to experience interference) in a noisy environment. Fragmentation is not used often in today’s networks because it rarely produces a benefit. Instead, it usually adds overhead.
Protection Mechanisms — Incompatible PHY formats (such as 802.11b and 802.11g) require protection for proper coexistence. Protection is usually achieved via an RTS/CTS exchange or a CTS-to-Self frame prior to the transmission of data. There are other types of protection (for 802.11n) that may be used as well. In addition to using these frame exchanges for protection, there are other times (such as when there are hidden nodes) when they may be enabled to attempt to improve overall network health.
Retransmissions — When a transmitted frame is not received properly by the intended recipient (or not acknowledged), a retransmission may be required. Transmitting a frame more than once is an obvious, and significant, source of overhead. When the frame is queued for retransmission, other sources of overhead (such as random backoff, IFS, etc.) are duplicated as well. Retransmissions represent one of the most problematic sources of overhead on our networks because they are one type of overhead that we can influence (with proper design).
Shall we go on? — If we wanted to extend the article, we could also break down the upper layers of the protocol stack as well, looking at the overhead inherent in each protocol. However, from the perspective of Wi-Fi (a Layer 1 and 2 protocol), the Layer 3-7 data is considered to be the payload. Complete reductionists might say that the application data is the only real payload that is not “overhead,” and I’d tend to agree. Since we’re focused only on Wi-Fi here, I won’t take it to that extreme.
Final Comments and Suggestions (FCS)
The Wi-Fi protocol is bloated with overhead. I don’t fault the engineers who designed the protocol for that. Any radio communication protocol will require some overhead. Broad WLAN adoption and use cases have made the 802.11 protocols very successful, but that success requires a lot of engineering complexity. Complexity requires coordination, and that coordination usually shows up as overhead. Network engineers should understand the sources of overhead on their networks and, within the limitations of their use case, seek to reduce overhead when possible.
A follow-up article will address the common network planning, design, and configuration strategies that will reduce unnecessary overhead.Tagged with: Wi-Fi, half-duplex, overhead, contention, PHY rates, signaling rates, protection mechanisms, retransmissions, fragmentation, interframe spaces, random backoff