Re: Has pthreads become antiquated?

Chris Friesen wrote:
Joe Seigh wrote:
Maxim Yegorushkin wrote:

If you care about that kind of latency a user-mode process is probably
a no go. A kernel module is what you need in this case.


It's why signals probably won't go away. And since they won't go away,
interoperability with threads is important.

One thing that hasn't been discussed yet is that signals are generally not *designed* as a high-performance element. They're generally used to indicate an exceptional event, so there was no need to make them a fast path.

Like async i/o?


On Linux, if you look at the costs of sending a signal to another process vs sending them a message via a pipe, the pipe wins hands-down.

Latency. Overhead is not really an issue if you want low latency.

--
Joe Seigh

When you get lemons, you make lemonade.
When you get hardware, you make software.
.