The major failing of TCP when it comes to wireless lossy networks is that packet loss in TCP indicates congestion, and rarely in wired networks do packets actually get lost. Thus, the basic strategy is to deal with packet loss through some other mechanism, not indicating that congestion is occurring.
My major criticism is that the division of these modifications is dealt with by having each category treaded in its own section, with not as much comparison across the various strategies until the last section. I think more detail in how the different strategies compare would have been useful; this is more of a organizational problem and less a research question.
Overall, I think the most promising is the link-layer schemes which perform the best (I believe). The end-to-end protocols give good performance, but because they require changes at the ends, it seems impractical to implement. In addition, the ELN enhancement seems difficult to implement, even given their suggestions. The split-connection protocols, at least as implemented here, do not perform well.
2 comments:
Let's be sure to distinguish between error loss and congestion loss. Packets do get lost on wired networks, but hopefully not often and mostly at congested exchange points.
True--- I meant to say that loss in general in wired networks almost always occurs where there is congestion, so loss=congestion is reasonable.
Post a Comment