Traffic differentiation with CSMA/ECA [Part 1.5]

Obviously, this is a follow-up. We can actually do better.

In the followings we propose a method for eliminating Internal Collisions (IC) among Access Categories (AC) of a node. Further, we show that by eliminating ICs the length of the transitory state towards a collision-free schedule is reduced.

Taking advantage of more information

Our solution is based on the fact the ICs happen inside a node, therefore the backoff counters of each AC may be known. Based on this, we provide a solution assuming knowledge of the counters.

We can draw a random number, B, that is not equal to any of the other ACs’s counters. Figure 1 shows an example. (In the figure the deterministic backoff after a successul transmission is Bd[4] = {4, 8 ,16, 16} for each AC, from 1 to 4 respectively.)

Figure1: The new counter must not be equal to any other AC's counter.

Figure 1: The new counter must not be equal to any other AC’s counter.

Even though this counter seems to be OK, it may actually collide with AC 1’s sixth transmission (if all previous ones were successful). We can do even better by considering that ACs in CSMA/ECA use a deterministic backoff after a successful transmission.

To avoid ICs with future transmissions of the successful AC, we must perform an additional check. The difference between the newly randomly drawn backoff, B, and B(i) (where i = 1 in the example), should not be a multiple of min(Bd(3),Bd(i)); where Bd(3) is the deterministic backoff after a successful transmission of the AC which now is using a random backoff. This way B will not cause another internal collision in future cycles. Algorithm 1 shows the process of selecting an adequate backoff for the next transmission.

Algorithm 1: Choosing a suitable backoff after an IC or collision.

Algorithm 1: Choosing a suitable backoff after an IC or collision.

 

Following Algorithm 1, we can see that B = 24 in Figure 1 is not a suitable counter. This is derived from Line 10 of Algorithm 1. Where: Bi = 24, Bdi = 16, Bdj = 4; the result would be zero.

We call the process described in Algorithm 1 as Smart Backoff (SM) generation.

What is the benefit of eliminating ICs?

We are just scratching the surface of what implies an effective elimination of ICs. Nevertheless, the following figure shows how the transitory state’s duration is drastically reduced when using SM after an IC or a collision.

Figure 2: Accumulated collisions in the last 1000 slots. Simulation performed with 16 nodes.

Figure 2: Accumulated collisions in the last 1000 slots. Simulation performed with 16 nodes.

A reduction in the transitory state means that stations are able to reach collision-free operation faster, therefore achieving the maximum throughput quicker (in saturation).

Any other things?

Well, I didn’t want to start Part 2 of these series of posts without introducing SM.

In real networks supporting traffic differentiation, the highest priority AC is not always saturated. Instead, when there is a packet to transmit this AC will always win in a IC. Forcing other ACs to delay their transmission. A random backoff after an IC may disrupt an existing collision-free schedule. Can SM avoid such disruption?

In Part 2 we will analyse the time between successful transmissions when using traffic differentiation. Furthermore, we will also take a deep look at the effect of SM, as well as its possibilities.

Regards,

L.

 

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