Paper review: MACAW: A Media Access Protocol for WireLess LAN's
Reviewer: Kenneth Chin
The main idea of this paper is to introduce a new media access protocol for wireless LAN. Basically, the new protocol MACAW is based on and evolved from the MACA first proposed by Karn and later refined by Biba. MACAW stands for Mutliple Access, Collision Avoidance for Wireless. By recognizing the characteristics of wireless LAN and by pinpointing the drawbacks of the MACA protocol, MACAW is proved to be able to achieve its goal: 1) High performance and 2) Bandwidth fairness.
Note that token approach is an alternative to media access control, but it is believed that multiple access scheme is more robust to the wireless enivornment where devices (technically called pads) are enter and leaving cells frequently.
First, the author presented 4 important observations about wireless LAN which directly drive the design of MACAW.
1) The relevant contention is at the receiver, not the sender.
2) Congestion is location dependent.
3) The media access protocol should propagate congestion information explicitly rather than having each device learn about congestion independently.
4) The media access protocol should propagate synchronization information about contention periods, so that all devices can contend effectively.
Immediately after the observations, the author then proposed what corresponding things can be involved to apt those obersvations in designing MACAW.
1) Carrier sense capability is not necessary.
2) In each station, separate queues for each stream which runs its own backoff algorithm.
3) Propagate the backoff counter of the sender to all hearing stations.
4) Use Data-Send packet (DS) and Request-for-Request-to-Send packet (RRTS). However, this doesn't solved all the problems and leave out some open questions.
In this paper, there are two major open questions that need attention.
1) Scenario: B1 and P1, P1 and P2, P2 and B2 are respectively in range. B1 is sending an RTS to P1 while P2 is transmitting data to B2.
Problem: P1 cannot hear the RTS from B1 because P2 is transmitting data to B2. The only time B1 can successfully initiate a transfer is when its RTS happens to arrive during those very short gaps in between a completed data transmission and the completion of P2's next RTS. Even RRTS scheme is used, it doesn't work out because RRTS has the same probability as RTS to get through to P1.
2) Scenerio: Some stations are in range with the sender but some are not.
Problem: Only those stations within range of the sender will defer, and those that are within range of a receiver but not the sender will not be given any signal to defer. Therefore, it is very likely that there exists CTS collisions. Yet we have to figure out how to make receiver generated control messages like CTS work in the case of multicast without involving several rounds of contention.
My impression to this paper is that it is a revolutionary paper in the sense that it introduces a new protocol which is more promising. In such a respect, I would grade this paper 5. However, one has to have a very solid background on how wireless lan works and what MACA and Binary Exponential Backoff (BEB) algorithm is before they can understand fully the paper. Basically, it took me almost a day to understand all those arguments in the paper.
There are several comments on the MACAW protocol. MACAW is complicated in implemenation and the more functions that are going to incorporated in the protocol, the more complicated and inefficient it gets. Secondly, one major issue of wireless LAN is that stations that are within range are actually overhearing each other, even though they are in different cells. My view is that it might be possible to do some hardware signal transform or encoding so that only stations within the same cell can hear. In this case, the protocol would be a lot less complicated, but at the same time we are still able to solve the inappropriate propagation of backoff values. One more suggestion is that there should not be peer-to-peer communication between pads. The only communication going on is only pad-to-host or vice versa.