Subj : Re: Optimizing for latency
To   : comp.programming.threads
From : scream29125
Date : Wed Jun 08 2005 06:58 am

> How does the process wait for an interrupt?

I don't know. This is part of a low-level API provided to me. I call a
function, similar to:

    receive(device_ID, request_buffer, buffer_size, BLOCK);

This function will block until the device responds with some request.
It used to sit in a loop, which consumed 100% CPU time. This has been
changed to use interrupts instead. Is there a "good" and "bad" way to
wait for an interrupt?

> What are the threads doing when they're not actively
> servicing a device?

They are blocked waiting for a request, as shown above.

I have done tests with jobs that require minimal CPU interaction, and
other jobs that require lots of CPU interaction. Those jobs requiring
more CPU interaction tend to perform worse than those require little
interaction. In all cases, "top" and "ps" report plenty of idle CPU
cycles. This is why I guessed that the problem is due to multiple
devices requiring attention simultaneously. I'm open to the possibility
that I'm wrong with this guess... if you have any suggestions on how I
can dig deeper into this.

.