This paper develops a new link-layer access protocol for mobile wireless devices and evaluates it's performance in a number of different configurations. The devices in this case are pads and they communicate with base stations. Because the pads are mobile, and the range of each base station is quite small, this is a complex problem.
The authors first show why a carrier-sense approach fails. Specifically, if two pads are in range of a base station, but not of each other, they will both sense, find nothing, and send, but there will be interference at the base-station. This indicates that in wireless networks conflicts occur at the receiver, not the sender.
The MACA protocol attempts to deal with this problem by using a Request To Send (RTS) packet, which a receiver replies to with a Clear To Send (CTS) packet if there is no congestion at the receiver. By snooping on this traffic nodes can co-ordinate and not step on each other, and congestion control is performed at the receiver, as desired.
MACAW improves on MACA in a number of key ways:
- They improve the backoff algorithm when there is a loss to not be so aggressive. In addition, to improve fairness, they share backoff information. In MACA this was a problem as one host could end up dominating a link while the others all backed off.
- They introduce a more fair contention model that tries to maintain equal flows over each stream of data, rather than each sender (so a base-station sending to two pads will get half the available bandwidth instead of one third)
- ACKs are added for faster retransmit since TCP's timeout is more geared to the wide area and therefore has timeouts that are too long.
- They add a Data Sending (DS) packet that gets sent after a CTS message. This prevents the case where a host hears the RTS but not the CTS. This is a bit counter-intuitive as it would seem safe for a host to send in this case. However, with the bi-directional RST-CTS-DATA exchange congestion can occur at both ends of the link so the DS packet is necessary.
- Finally, as a fix for the situation in which two pads are communicating with different base stations, both want to receive data, and they are close enough to interfere with each other. In this case one pad will likely dominate the channel, as the other can only send a CTS if it gets a RTS right after the other pad finishes a data transmission, which is a small window. To fix this the authors add a Request-for-Request-to-Send (RRTS) packet which can prompt a sender to re-send it's RTS. This doesn't deal with this case where one pad is totally blocking the channel however, and the second host will never even hear the initial RTS packet.
Comments
This was an interesting paper, and they propose a number of interesting a clever solutions to deal with the problems one encounters in a wireless environment. The paper does feel a bit like they thought up some situations and picked specific solutions for them, however, and since their results are all simulated versions of those situations, it's obvious they will do better. Their approach is slower than standard MACA on a non-contended link (due to the extra RTS/CTS/ACK traffic) so it's not clear MACAW is always better.
I would like to have seen more results from a general deployment, or an attempt to generalize their solutions and show why other methods wouldn't work, and/or why the situations they consider in the paper cover the majority of situations mobile clients will run into in the real-world.
An interesting paper none-the-less, and one I would keep on the reading list.
No comments:
Post a Comment