Keith Bostic wrote:

On Oct 9, 2007, at 12:02 AM, Howard Chu wrote:

...

With a CPU-hog running in the background, the test with BDB 4.2.52 takes only 37 seconds, while BDB 4.6.21 takes 1:42. Watching with top you can see that BDB 4.6.21 gets a lot less CPU than BDB 4.2.52. This is the problem with using yield() on an NPTL system - whereas on most POSIX systems yield() only yields control to some other thread in the current process, on NPTL yield() gives up the CPU for the entire process.

Is there a call that would be preferable on NPTL?

What call on NPTL will only yield control from the thread, not the entire process?

When we first encountered this problem in the OpenLDAP source we used select() with a zero timeout. But I recall that that performed poorly as well, and nowadays we just avoid doing any type of yield at all. I think the real answer here is that on Linux 2.6 you really just want to use the native mutex (pthread, futex, whatever it's called) and not do anything else.