802.11n Line Rate vs. Data Rate - Part 1: Frame Aggregation

802.11n Line Rate vs. Data Rate - Part 1: Frame Aggregation

By CWNP On 10/08/2007 - 6 Comments

Although the PHY layer enhancements (such as MIMO, 40 MHz wide channels and short guard intervals) introduced in 802.11n improve the maximum line rate¹ by more than 10x, the MAC layer enhancements (such as frame aggregation) are absolutely essential to achieving data rates that are anything close to the line rate. In this first article, we will briefly examine how data rate (802.11 throughput) is increased by MAC layer enhancements, in particular, we will look at frame aggregation. Without this 802.11n feature, it would not matter what line rate the physical layer achieves – the maximum data rate would be capped by overhead.

For later comparison, let us first examine an 802.11b data transfer. For simplicity, we will look at 802.11 data throughput only, ignoring TCP throughput and throughput at layers above the wireless medium. Say we give our 802.11b STA an ideal world; it is transmitting the maximum payload size (2304 bytes) with every packet, at 11 Mbps (line rate), with short preamble enabled, no encryption and no contention for the medium by other nodes – as fast as an 802.11b node possibly can. As shown in the following figure, the total transaction period for every 2304 bytes is 1938 µsec, or a maximum of 516 transactions per second. Multiplication gets us a raw 802.11 data rate maximum of 9.51 Mbps². 

 

Looking at the figure above, it is clear that most of the time is spent transmitting the payload data. In fact, 86% of the air time is spent sending data; the remaining 14% is spent in PHY and MAC layer management overhead, such as interframe space, PLCP preamble and ACK.

 

Now, let’s look at transmitting the same 2304 bytes in the best case 802.11n scenario: Let’s also give our example 802.11n STA an almost ideal world; it is transmitting at 600 Mbps (4x4 MIMO, 40 MHz wide channel, short guard interval enabled), and no other contention for the medium. As can be seen in the following figure the total transaction period of every 2304 bytes is 176.91 µsec, or a maximum of 5652 transactions per second. This gets us a raw data rate maximum of 104.19 Mbps.

 

 

Although 104 Mbps is a significant improvement over 9.5 Mbps, we are achieving only 17% of the line rate when expressed as raw data rate. Compare that to 86% of line rate for our 802.11b STA. Looking at the figure above, one may quickly observe that most of the time is spent on overhead, not transferring data. Now, let’s introduce one level of frame aggregation into the picture. MSDU aggregation increases the maximum size of frames from 2304 octets (bytes) to 7935, by aggregating (or combining) frames transported between MAC and LLC layers. As illustrated in the following figure, the transaction period of every 7935 byte packet is just over 250 µsec, giving us a raw maximum data rate of 251.92 Mbps³, or about 42% of our line rate.

 

 

One may casually observe that there is still nearly 60% of overhead. Introducing another level of aggregation, MPDU aggregation solves this. MPDU aggregation increases the maximum frame size from 2304 octets to 65535. As can be noted in the following figure, the 802.11 transaction time for 65535 bytes is just over 1 msec. However, we’re moving a lot more data in that 1 msec, giving us a maximum data rate of 514.01 Mbps, or 86% of line rate – what we were getting with our 802.11b STA.

  

 

This demonstrates that not only is frame aggregation an important enhancement to the 802.11 MAC layer, it is absolutely critical if an 802.11n station is to take any real advantage of the High Throughput (HT) PHY layer line rates. It is probably for this reason that frame aggregation is required of all 802.11n devices, both by the IEEE and the Wi-Fi alliance.  

¹Line rate: the speed at which data bits are physically transferred over the communications medium.

² Mega bits per second: Multiply the number of transactions per second, by the number of bytes per transaction, by the number of bits per byte. (2304 x 516.04 x 8 = 9511649.28) 

³We are ignoring data rate loss due to encapsulation. Also, a single ACK is shown, as opposed to block ACK.

Warren Blackwell is a Sr. Software Engineer at AirMagnet, Inc.

wblackwell@airmagnet.com

http://www.airmagnet.com

"Wireless Network Assurance"

Tagged with: 802.11n

Blog Disclaimer: The opinions expressed within these blog posts are solely the author’s and do not reflect the opinions and beliefs of the Certitrek, CWNP or its affiliates.