MACAW: A Media Access Protocol for Wireless LANs
url show details
You need to log in to add tags and post comments.
| papers | collections | search | login | register | forgot password? |
|
| papers | collections | search | login | register | forgot password? |
|
This paper addresses basic design issues in wireless MAc protocol by evaluating and improving the previous work. Although some of its assumptions simplified the problem targeting an infrastructure-based indoor wireless network environment, it well focused on general design considerations applicable for various wireless networks such as wireless ad hoc networks.
MACA and MACAW use the RTS/CTS handshake for avoiding collision in a hidden terminal situation. But, there must be a threshold where RTS/CTS overhead outweights performance degradation due to hidden terminal. I found 802.11 has a threshold option (packet size) for the RTS/CTS handshake but not any specific guideline for this option. I wonder how the packet-size-based option works and what other type of options could work for the threshold of RTS/CTS performance.
MACAW solves many problems by simply adding more control packets. Although the authors claim the overhead is not significant when compared to the benefits of the protocol, it seems that we could keep adding more packets and there would still something to solve(specially synchronization problems). Eventually, the overhead could be to big to make this approach worth it.
On the other hand, the authors do state that some of this packets could be replaced by carrier-sense (DS packet, for instance).
I wonder how this protocol or 802.11 would handle multiple wireless networks in the same area. This situation seems to be really common nowadays because most people have their own wireless routers. For example, at my apartment, I found at least 20+ wireless networks in my range, but somehow I still get almost full bandwidth all the time.
Even though this paper is written a long time ago, it's a bit weird that they didn't consider any security matters. For example, how can the base station guarantee fairness if one of the senders would not conform to the backoff algorithm? And, what if people flood the base station with RTS packets?
And, as Miguel mentioned above, I think these extra control packets could introduce new problems like control overhead.
When reading this paper, one thing that kind of bothered me was that all the (pessimistic) examples assume unidirectional streams. Even with just UDP, I'd expect some sort of back-and-forth of control data to happen at the application level on the multi-minute timescale of these simulations. With this in mind, maybe fairness problems aren't really so bad.
Regarding 802.11b networks: if I recall, there are a number of channels (11-ish?), so contention on 20 networks will only be 4/channel (again, if I recall, channels overlap slightly). Microwaves and cordless phones, however, aren't quite so forgiving.
This paper describes a media access control protocol (MACAW) well-suited for wireless environments. At many points the authors seem to be guided by intuition and hence some of the suggested solutions seem overly simplifying (e.g. add RRTS before RTS to avoid having only one client consuming all of the throughput). However this intuitive approach also made the paper easier to comprehend and in many instances more convincing.
Now I will try to address one of the issues brought up by Samir (comment #1). I believe that the "symmetry" assumption ("If A can hear B, then B can hear A") is not very realistic. This is especially true when the bases (access points) are omni-directional and the pads (clients) have directional antennas. In this case the clients will most probably hear more than one bases, but only one base will be able to hear them. In this scenario, if the bases have a way to communicate to each other, then maybe it would make sense to have an explicit protocol that informs the pads (clients) about which bases (access points) can hear them. This will allow the clients to know which bases are affected by their transmissions.
Among multiple assumptions made in the paper, I think the assumptions that:
- All base stations and pads transmit at the same signal strength
- No two base stations are within range of each other
- No pad is within range of two different base stations
seem not to be practical.
Also, I have a question about the backoff copying that this pape used to achieve fairness. Wouldn't "make all stations attempting to communicate with the same receiving station use the same backoff value" cause more collision, and thus violate the purpose of random backoff time to aviod collision among several transmitters?
Thanks for the excellent review, Samir. I think the unreasonable assumptions are:
2. All base stations and pads transmit at the same signal strength
4. No two base stations are within range of each other
5. No pad is within range of two different base stations
I liked how this paper did a lot of testing to examine their configurations.
For today's wireless networks, I think the most unreasonable assumption is that interference/capture can be ignored. Unlike the small (3 meter radius) cells addressed by the paper, today's IEEE-802.11 networks have a large radius, so in urban or dense suburban residential areas interference will definitely be an issue.
I think the decision to avoid token-based protocols was a good one. In a setting where connections are unreliable and hosts are rapidly entering and leaving the network, I think way too much overhead would be spent on token management and dealing with lost or duplicate tokens.
I'm satisfied with my daily experience with Wi-fi, so I'm guessing some of these assumptions have been fixed or were deemed OK. I vote for these assumptions to be unreasonable in the context of this paper:
3. Capture and interference can be ignored
4. No two base stations are within range of each other
5. No pad is within range of two different base stations
As for assumption 2 "All base stations and pads transmit at the same signal strength" - it doesn't seem like it'd be too much of a problem for say laptops, but does it pose a problem for cellular networks?
The paper did a very good job of finding corner cases and proposing solutions to it. I think the general approach of addressing a problem case by case doesn't make me feel like the solution is complete.
Most unreasonable assumptions:
2. All base stations and pads transmit at the same signal strength
4. No two base stations are within range of each other
5. No pad is within range of two different base stations
Samir, I checked wikipedia and found that Europe has 13 channels. Are they using a 1-7-13 guideline?
I thought the way the paper analyzed the shortcomings of MACA and improved it well. Among the assumptions that Samir listed, I thought the most unreasonable assumption was the assumption of symmetry. I wonder what current systems avoid this problem since it would seem that power variation would be especially prevalent.
I liked Section 3.1 in which they argue for the MILD backoff algorithm over BEB. They argue that both algorithms perform nearly identically in experiments, but MILD is much more fair (they give a good example under BEB in which a single pad transmits at channel capacity and the other pad is at maximal backoff). The analysis of fairness vs. efficiency issues is difficult in general, but led to a clear decision in this case.
In MACAW, the exchanging involves RTS-CTS-DS-DATA-ACK. It seems RTS-CTS-DATA-ACK is adopted by 802.11. Why the DS signal is used by 802.11? (http://en.wikipedia.org/wiki/IEEE_802.11_RTS/CTS)