Sometimes you have to wonder how much we are suffering "death by a thousand cuts" though. This is still quite noticeable on mobile and in certain applications.

You left out a really big, significant sector: finance. Minimal latency is king (at least in trading). This often means little or no heap allocation (if there is any, it's often done up-front). User-space network stacks to avoid the user to kernel context switches. Web devs think that sub 10ms response time are good. In finance, often anything over 10 nanosecs (for example when responding to an IOI - indication of interest) is trash.

Yeah, sure, in some domains latency isn't terribly important, but there certainly are a number of domains where it is.

Yes, we are in agreement (except for what I perceive as a derision toward "web devs", which by the way I am not and didn't even mention, so I'm not sure why you even mentioned it). I also left out specific kinds of some very resource constrained domains in general; a terrible omission on my part. I'm sure there are others as well that I didn't call out specifically. Not having experienced every single kind of programming problem type I rather think it would be silly and pretentious of me to try to enumerate them all.

> except for what I perceive as a derision toward "web devs", which by the way I am not and didn't even mention, so I'm not sure why you even mentioned it

I didn't read it as derision. I read it as an indication of a different focus between two areas of development. Microseconds don't matter in most cases. Milliseconds matter more often. Something optimized for microsecond-ish timing will make a tradeoff somewhere else (like code readability, generalness, etc).

He probably means 10 microseconds, even if he's talking about HFT. The state of the art for FPGA's seems to be around 700 nanoseconds. [0]

[0] http://stackoverflow.com/questions/17256040/how-fast-is-stat...