Subj : Re: Optimizing for latency
To   : comp.programming.threads
From : Jomu
Date : Wed Jun 08 2005 10:17 am

10% CPU for single device controlled is alarming... It's a lot of CPU
cycles. That "interrupt driven" system is very busylooploking.

One way (short of inspecting that low level API) would be to improvise
- to use that API in "non-blocking mode" and there make smaller time
slices while waiting for I/O. That would be pooling mode (as it is
obviously now, but with higher expectations :).

So it will be:

   while (!receive(..., FALSE)... ) { sleep (0.005) };

Of course, YMWV, you will know speed of your device better, and time in
which whole package can arrive - that is time I would use for delaying,
or half of it.

dd

.