Traffic differentiation with CSMA/ECA [Part 3]

In Part 2, we saw that our SmartBackoff strategy does not contribute to a reduction in the average time between successful transmissions. Here, we are going to look at another metric: average number of dropped packets.

A little bit of background

In WiFi, a packet is dropped once the number of attempts to successfully transmit it reaches a threshold. This threshold is known as the retransmission limit.

Because with CSMA/ECA we are able to reach a collision-free schedule in saturation, we expect a reduced number of dropped packets when compared with EDCA.

Figure 1 shows the average number of dropped packets in a network with 16 nodes for EDCA, CSMA/ECA+Hysteresis (ECA_h), CSMA/ECA (ECA) and CSMA/ECA with SmartBackoff (ECA_s). All Access Categories (AC) are in saturation.

Screen Shot 2015-01-29 at 16.47.50

Figure 1: average number of dropped packets. Sixteen nodes and four ACs in saturation.

Again, given that with 16 nodes and four access categories CSMA/ECA is capable of achieving a collision-free schedule, all CSMA/ECA bars drop very few packets when compared with EDCA. Figure 2 shows the same exercise but with a network composed of 32 nodes.

Screen Shot 2015-01-29 at 16.47.22

Figure 2: average number of dropped packets. Thirty two nodes and four ACs in saturation.

Here we can actually see an increase in the average number of packets CSMA/ECA drops due to reaching the retransmission limit. This increase is expected given that we could not construct a collision-free schedule with 32 nodes, as shown in the first part.

What is the effect of SmartBackoff?

Figure 3 zooms into Figure 2 removing EDCA.

Figure 3: average number of dropped packets. Thirty two nodes and four ACs in saturation. CSMA/ECA only.

Figure 3: average number of dropped packets. Thirty two nodes and four ACs in saturation. CSMA/ECA only.

The gap between ECA_s and ECA in Video, Background and Best-effort is due to Internal Collisions (IC). Without SmartBackoff ICs are possible. Each IC is treated as a collision, which in turn contribute to the dropping of packets. With long simulations, Voice for ECA and ECA_s should be similar, given that they only deal with real collisions. I drew the standard deviation over the Voice bars for ECA and ECA_s. The reduced number of dropped packet in ECA_h is due to not performing any packet aggregation.

TL;DR;

CSMA/ECA drops much less packets than EDCA.

CSMA/ECA with SmartBackoff reduces the number of dropped packets because it eliminates Internal Collisions (ICs).

We will continue this series in future posts :).

L.

 

 

Posted under: CSMA/CA, CSMA/ECA, EDCA, Fair Share, Hysteresis, MAC, QoS, WiFi

Leave a Reply