Traffic differentiation with CSMA/ECA [Part 4]

There are other ways EDCA can provide priority at the MAC level [1, 1.5, 2, 3]. The Arbitration Inter-frame Space (AIFS) is a new waiting period for every Access Category (AC). ACs in a station should have a period equal to AIFS[AC] = SIFS+(AIFSN[AC])*slot_time of free channel before starting to decrement their backoff counters.

Fig 1: AIFSN

As you can see in Figure 1, AIFSN are defined by the standard. That is, AIFSN[ACs] = {3,3,4,8} slots. This means that each AC has to have a period equal to AIFS[AC] of free channel before starting to decrement its backoff counter. This method ensures that higher priority ACs access the channel quicker than background traffic.

As a way to get an insight into this feature, we can try simulate real traffic differentiation in non-saturated conditions. That is, we can generate sporadic high-priority traffic that should be served as quickly as possible.

NOTE 1: we are simulating CSMA/ECA and EDCA in a customised version of the COST simulator.  Our version is available for you to test, comment and hopefully help you on your research.

NOTE 2: The settings used in the simulator to derive the following figures are:

  • Each node produces 10Mbps of data to send.
  • 40% of the traffic is assigned to AC BK, 30% to AC BE, 15% AC VI and 15% to AC VO.
  • CWmin[AC] = {8, 16, 32, 32}. From VO to BK.
  • CWmax[AC] = {16, 32, 1023, 1023}. From VO to BK.

Figure 2 shows the average time between successful transmission in EDCA without AIFS, while Figure 3 shows the same metric but using AIFS in EDCA.

Fig. 2: Average time between successful transmission in EDCA without AIFS

 

Fig. 3: Average time between successful transmission in EDCA with AIFS

 

We can see in Figure 2 that the average time between successful transmissions seems to follow the AC priorities. That is, the highest priority AC transmits more often than the lower priority ones. Nevertheless, the strong traffic differentiation seen with AIFS (Figure 3) shows that the lowest priority traffic is effectively “delayed” to make way for the higher priority ACs. This is clearly shown in the Background curve in Figure 3, which suffers very large time between successful transmissions just when all other ACs are approaching saturation.

The average time between successful transmission for the lowest priority AC in Figure 3 is equal to the simulation time when the higher priority ACs saturate. This is because the average number of packets sent by this category is drastically reduced, or in this case eliminated, as shown by the Figure 4.

Fig. 4: Average throughput in non-saturated conditions. EDCA with AIFS

 

Figure 4 shows the average throughput in non-saturated conditions when using AIFS. As is can be seen AIFS ensures that high priority traffic is always served first, which results in a sort of service deprivation for lower priority ACs (Background and Best effort curves). As the Video and Voice AC saturate they are prioritised for transmission, leaving other ACs with little chance of accessing the channel.

AIFS and CSMA/ECA

Our CSMA/ECA is not able to implement AIFS the way EDCA does. The principle that makes CSMA/ECA capable of building collision-free schedules relies on the fact that the chosen deterministic backoff used after the successful transmission of an AC must be a multiple of the deterministic backoff used on the other ACs, e.g.: Bd[AC] = {4, 8, 16, 16}.

AIFS would disrupt any collision-free schedule in CSMA/ECA, given that the additional AIFSN[AC] waiting period would prevent the chosen deterministic backoffs from being multiple of each other. Moreover, EDCA+AIFS reset the AIFS waiting period everytime the channel is found busy, which would desynchronise the count-down of the backoff counter, leading to internal collisions.

CSMA/ECA uses the minimum contention window as its tool for differentiating traffic and building a collision-free schedule (if possible). As can be seen in Figure 5, CSMA/ECA is not able to provide greater throughput than EDCA+AIFS, this is because the CWmin and CWmax values are too small and do not allow the construction of collision-free schedules with CSMA/ECA. Furthermore, the aggregation performed through the use of Fair Share increases the average time between successful transmissions, as shown in Figure 6.

 

Fig 5: Average throughput. CSMA/ECA in non-saturation using SmartBackoff

Fig 6: Average time between successful transmissions. CSMA/ECA in non-saturation using SmartBackoff

Ongoing work

The case of a EDCA / CSMA/ECA mixed network is of great interest, given that CSMA/ECA is thought to be backwards compatible and a possible replacement of EDCA. We are still battling with some ideas on how this can be done while preserving fairness. (Some of the ideas we have in order to fairly compare EDCA with AIFS and CSMA/ECA is to assign an AIFS[BK] to each of CSMA/ECA AC’s transmissions; making them longer.)

We are looking at ways to increase the throughput and reduce the average time between successful transmissions with CSMA/ECA. This can be achieved by setting the CWmin and CWmax values to appropriate levels.

More on this in future posts.

Regards,

L.

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

Leave a Reply