Bandwidth versus Latency
Joel Gustafson / Thoughts
There's a rat in a maze who wants to escape.
There's another rat outside the maze, who has a map to the maze, but doesn't know where inside the maze the first rat is.
The rats can exchange messages with latency
l and bandwidth
b, constrained that
b/l is some constant
Consider the effect that the choice of
l have on the optimal communication strategy between the rats:
Case 1: high
l: If bandwidth and latency are both high, the communication strategy for the rats is for the second rat to just send the first rat the entire maze and do all of the processing client-side.
Let the rat figure it out for himself; it's too expensive to interact or tell him what to do.
Case 2: low
l: If bandwidth and latency are both low, the optimal communication strategy for the rats is for the first rat to act as a thin client and be "remote controlled" by the second rat - doing all the processing server-side.
When newspapers came out once a week, stocks were tracked daily, and everyone was working off of a common (small) set of information, it was more efficient for the network to generate opinions client-side. But now latency is lightning-fast, attention spans are a precious commodity, and sound bites and tweets are the atomic unit of our media. So naturally the media has converged on generating opinions server-side.