Topic > A comparison of mechanisms for improving TCP...

The paper focuses on the logical link control (LLC) part of the link layer without taking into account improvements that can be made to the logical link control (LLC) part of the link layer medium access (MAC) to reduce non-congestion related errors. One of the main reasons why Ethernet provides better performance is due to a robust collision detection mechanism called Carrier Sense Multiple Access (CSMA) with collision detection (CD). This layer prevents hosts from simultaneously transmitting on the channel causing corrupted frames. It therefore does not allow corrupted packets to propagate up the stack and cause unnecessary retransmissions at the LLC layer that can interfere with TCP retransmissions. We know that collision detection is more difficult in wireless networks due to the “hidden terminal” problem. But we can still use “collision avoidance” mechanisms to ensure that two hosts don't transmit at the same time. Collision avoidance mechanisms can be conservative, that is, they have backoff timers that gradually increase backoff intervals when an acknowledgment is not received. Instead of backing down, we can take a more aggressive approach, so that frames are transmitted as soon as an inactive channel is detected. If an error occurs, the sender will not hear a confirmation message within the timeout interval and will be able to retransmit. Problems with end-to-end schemes The main problem in this scheme is that the paper assumes perfect knowledge of wireless losses to generate an explicit loss notification (ELN Information). This information is then propagated to the TCP sender which takes corrective action based on the type of error. In a real implementation, it is impossible to accurately identify packets lost due to errors on a lossy network. This is mainly because the only other entity is between the source and the... center of the card... the acket. This will require the TCP layer to be improved to identify and handle different types of errors. 802.11 uses CSMA/CA to detect whether the channel is down before making a transmission. The exponential backoff strategy can have a negative impact on performance. So, in situations where there are more than a certain number ("N") of hosts accessing the same wireless link, we can have an arbiter decide which host has access to the link. In situations where there aren't enough hosts, they can resort to an aggressive version of CSMA that doesn't have large backoff timers. A hybrid approach like this will help reduce non-congestion errors on a wireless link and increase TCP performance. A robust error handling mechanism will ensure that the size of the TCP congestion window does not fluctuate unnecessarily and will also avoid duplicate retransmissions and concurrent retransmissions.