Freeminded.org

This post describes a bug in MiXiM’s Mac80211 which seems to be a fundamental error: when the MAC gets a packet from the Netw and the channel is busy, it schedules a senseChannelWhileIdle(currentIFS + remainingBackoff) after the ongoing transmission ends. Unfortunately, remainingBackoff is often 0 as post-backoff is likely to have completed. The result? Many synchronised collisions one IFS after the ongoing transmission.

Fixed in v2.2

What went wrong?

Mac80211 in MiXiM uses the same mechanism for backoff and post-backoff, a senseChannelWhileIdle which schedules a timer for a duration of IFS + remainingBackoff (i.e. The inter-frame spacing and the present state of the backoff counter).

If Host A were doing post-backoff when the frame from Host B arrived, the remainingBackoff would have been > 0 and backoff would have resumed after the frame from Host B finishes.

However, Host A has already completed its post-backoff (remainingBackoff was 0) so it essentially was IDLE when the beacon was generated (actually, it was receiving Host B’s frame). So what happens now is that all nodes which have an arrival during Host B’s frame AND have completed their post-backoff, will wait one IFS and then transmit, resulting in synchronised collisions one IFS after a transmission (or an EIFS after a collision).

The correct behaviour here is for Host A’s MAC to note that post-backoff has completed (remainingBackoff has reached 0) and the medium is busy, so the MAC must draw a new backoff from the contention window.

How to fix it
Proposed change in beginNewCycle():

if (!fromUpperLayer.empty()) {
	if(!contention->isScheduled()) {
		if(channel.isIdle()) {
			senseChannelWhileIdle(currentIFS + remainingBackoff);
		}else{
			//channel is busy
			if(remainingBackoff==0){
				remainingBackoff = backoff();
			}
		}
	}else{
		//busy in contention, frame is enqueued.
	}
}else{
	// we do PBO
}

Upon completion of the ongoing transmission by Host B, beginNewCycle() will be called. But now the channel has turned idle (we are in the IFS after the transmission) and we have just choosen our IFS (Host B’s frame was correctly received, so we set IFS = AIFS and not EIFS). Then, in beginNewCycle(), it will perform a senseChannelWhileIdle(currentIFS + remainingBackoff). The remainingBackoff was set previously (during the previous iteration of beginNewCycle()) to unif(0,cwMin) and now contains a valid backoff.

The results show an improvement in reception probability in lightly loaded conditions: here it is most probable post-backoff has completed and we get an arrival from the upper layer while the medium is not idle. The collision probability is decreased because the synchronisation after a transmission (e.g. Host B’s transmission) is broken by means of the backoff procedure.