Re: Threads not waking on pthread_cond_signal()

[Ian Wienand <ianw gelato unsw edu au>]

On Mon, Jun 23, 2003 at 03:26:43AM -0400, Jakub Jelinek wrote:
> Is this 2.5.72 on i386 as well or some other kernel?

Yes, that result is from i386 on 2.5.72. 

The effect also seems to appear on IA64.  I had some 0.36 binaries
lying around, running the same program attached to my previous mail 5
times with them on 2.5.72 IA64 gives :

275337 wakes ups in 4.99957 sec = 55072 per second
277239 wakes ups in 4.99911 sec = 55458 per second
275627 wakes ups in 4.99922 sec = 55134 per second
254447 wakes ups in 4.99989 sec = 50891 per second
277556 wakes ups in 4.99938 sec = 55518 per second

whilst everything the same but using 0.48 gives

205466 wakes ups in 4.99915 sec = 41100 per second
177290 wakes ups in 4.99971 sec = 35460 per second
231181 wakes ups in 4.99925 sec = 46243 per second
235892 wakes ups in 4.99909 sec = 47187 per second
229463 wakes ups in 4.99982 sec = 45894 per second

Possibly the difference is that with requeue there are two futex calls
(one to move the queue, one to wake up a thread), where as the old
codepath only has one?

-i