Dead reckoning

In the R15 postmortem, it was mentioned that R16 improves dead reckoning. The subject is very interesting, as most assume that the client is predicting things and that is often the reason why we see characters jump in the game as it gets corrected.

It would be very interesting to learn what was changed in R16. The basic algorithm is pretty well described in this article http://www.gamasutra.com/view/feature/131638/dead_reckoning_latency_hiding_for_.php

So in a tight combat situation where trajectories are extremely short, I wonder what adjustments have to be made to optimize it. Perhaps it takes into account non-linear trajectories?

I had thought that one benefit of engaging in a combat might be that throughput between those specific clients in increased by spending less cycles on everything else. I wonder if that is done as well.

This vaguely relates to CPU branch prediction and pre-fetching. In CPU's, the slow part by 100 times is the memory read, so the cores try to predict where the code is going to go and start the memory read transfers well before the reads instructions occur. When I write fast code, I often consider how those branches and memory accesses are likely to occur in the core and optimize around that. Cool stuff.