1. 802.11n MAC layer simulation Submitted by: Niv Tokman Aya Mire Oren Gur-Arie

2. Background In the past several years, computer communication has turned toward the wireless medium. 802.11n is proposition for a new and improved protocol, that aims to deal with higher demands – video streams, VoIP conversations, file sharing and more. One of the main enhancements introduced to the MAC layer is packet aggregation

3. Packet aggregation The idea – squeeze packets into one transmission. Reduces the overhead of control packets and timeouts related to the 802.11 protocol May be destined to single or multiple receivers

4. Project goals Examine the influence of aggregated communication on periodic information – with focus on VoIP Examine the influence of aggregation size (number of packets per aggregation) on network’s behavior

5. Standard protocol transaction 1. Data arrives at Tx FIFO of STA A from VoIP data generator. IAC is sent. 2. AP receives IAC, wait SIFS and respond with RAC. 3. RAC received. STA A sends non-aggregated VoIP data 4. After DIFS, the operation is repeated by STA B 5. Aggregation size is reached. AP send IAC followed by the aggregated data 6. Data is received by STA C

6. Simulation structure and flow

7. Simulation assumptions There is no random noise in the simulation and error rate is zero. All stations are 802.11n (no legacy devices) Each station can be involved in one conversation at a time All conversations are between a local station and remote network (Through the AP) Only AP uses aggregated packages The destinations of conversations are distributed equally. The length of conversations is distributed exponentially. No Ack packages in MAC layer - the simulated network layer

8. Simulation parameters All simulations runs had these parameters: Number of stations: 250 Average conversation length: 1 sec Number of conversations: 1000 to 20,000 in steps of 500. For each number of conversations, the following aggregation max size were defined: 100-700 (bytes) in steps of 100.

9. Simulation constants Time slot = 20 u sec SIFS = 10 u sec DIFS = SIFS + 2*TimeSlot = 50 u sec TTL for aggregation: 8000 u sec

10. What are we checking ? X axis is the number of conversations simulated. The estimated network load is a measure of how much DATA is injected into the network. AvgNumOfBytesConversation*NumberOfConversation/SimulationTime Best effort is the free percentage of the network for other TCP communication. We want to check the effect of aggregation size on the packets delay and best effort, in various network loads. Packets delay measured is average delay and max delay, as well as max delay of successful short VoIP and long VoIP packets.

11. Average delay of VoIP packets

12. Average delay is under 10 ms (except for aggregation size of 100), which indicates that not many packets are lost Delay rises rapidly with small aggregation size like 100 bytes. This is because the AP sends at a high rate, and he has precedence over other stations, which causes the delay.

13. Max delay of VoIP packets

14. We can see that Max delay can be quite high. This means that packets are lost – they don’t arrive on time. Max delay tends to decrease with larger aggregation sizes, since LESS AP transmission occur. Instead, whenever the AP transmits, the transmission time increases. This increase, however, is negligible in respect to more AP transmissions

15. Max delay of successful long VoIP packets

16. Max delay of successful short VoIP packets

17. Max delay of successful VoIP packets As can be seen, in simulations with a smaller number of conversations, the max delay of successful packets is relatively far from the upper boundary. For relatively low network capacity, the max delay is derived from the TTL timing of the AP (8 ms). At higher network capacity, the max delay is almost equal to the upper boundary, which indicates that the max delay is derived from the regular stations.

18. Best effort

19. It is difficult to determine the effect of aggregation size on best effort. The difference is most noticeable on aggregation size of 100 bytes, which is in effect no aggregation. Higher aggregation sizes show higher values of best effort.

20. Aggregation size

21. Analysis 100 bytes aggregation is like no aggregation at all, which causes the AP to take over the medium and delay regular stations. Best average results are for 300 bytes aggregation size, but max delay is still high. Larger aggregations are better as network load is increased, but average gets worse as more packets wait in the aggregation.

22. Implementation issues Network capacity is represented in number of conversations. Actual network capacity is much lower and cannot exceed a certain value. Analytic calculation shows that only 125 conversations can occur concurrently in time frame of 10 ms. calculation calculation Using short time slot is very similar, therefore we focused on long time slot. Example for the similarities can be seen here. here

23. Implementation issues – cont’ Due to the capacity limit, there is little sense in running the simulation with a very large number of stations and conversations. This generates higher capacity, yet meaningless results – the unreasonable high number of stations creates a tough competition over the medium and results in horribly high delays, for example. Example: Example: 10,000 stations, 40,000-60,000 conversations. Example:

24. Thank you Niv, Oren & Aya

25. Analytic calculation of maximal number of conversation in 10ms frame time: Average transmission time: DIFS+IAC+SIFS+RAC+SIFS+DATA = 50+2+10+2+10+6 = 80 u sec Maximum number of transmissions: 10,000/80 = 125

26. Network capacity As can be seen, the actual network capacity in our simulation does not exceed 1.1 Mbps. This is due to the 125 conversation limit. However, there is a correlation between the number of conversations and the “actual” capacity over the different aggregation size, therefore number of conversations is adequate rating system.

27. Short VS long time slot There is high correlation between long time slots and short time slots, hence the analysis of long time slots simulations can be applied to short time slots simulations as well.

28. High number of stations and conversations There is no clear orientation, and average delay is unreasonably high. That many stations create a tough competition and starvation of stations. This is unrealistic and provide no insights.