It's interesting. Did anybody try rebuilding the kernel with HZ=1000? I'm also curious about it.
-- Fengguang > Martin, > > Change the following line of the code to sleep for 20 sec instead of the > counter(variable i). You will find the penalty to be 10 msec. > > BTW, does the fix proposed to change the kernel with HZ=1000 worked?? > > Reg > Ved > > -----Original Message----- > From: Martin, Stephen [mailto:[EMAIL PROTECTED]] > Sent: Wednesday, February 06, 2002 12:55 PM > To: '[EMAIL PROTECTED]' > Subject: Why do threads take so long to wake up underlinux > > > I've been doing some benchmarks on my machine (a dual 733 running redhat > 7.2) > and there seems to be a 20 millisecond penalty for doing a sleep or a wait. > Here > is code that illustrates it: > > try { > for (int i = 0; i < 30; i++) { > long start = System.currentTimeMillis(); > Thread.currentThread().sleep(i); > long end = System.currentTimeMillis(); > System.out.println("total time = " + (end - start) + " for > sleep(" + i + ")"); > } > } catch (Exception ex) {} > > The output i get is > > total time = 2 for sleep(0) > total time = 21 for sleep(1) > total time = 20 for sleep(2) > total time = 16 for sleep(3) > total time = 23 for sleep(4) > total time = 19 for sleep(5) > total time = 20 for sleep(6) > total time = 19 for sleep(7) > total time = 20 for sleep(8) > total time = 20 for sleep(9) > total time = 20 for sleep(10) > total time = 29 for sleep(11) > total time = 29 for sleep(12) > total time = 29 for sleep(13) > total time = 29 for sleep(14) > total time = 29 for sleep(15) > total time = 29 for sleep(16) > total time = 26 for sleep(17) > total time = 33 for sleep(18) > total time = 29 for sleep(19) > total time = 29 for sleep(20) > total time = 39 for sleep(21) > total time = 39 for sleep(22) > total time = 39 for sleep(23) > total time = 39 for sleep(24) > total time = 40 for sleep(25) > total time = 39 for sleep(26) > total time = 39 for sleep(27) > total time = 39 for sleep(28) > total time = 39 for sleep(29) > > This is pretty much consistent on all jvms that i've tried blackdown, sun > (1.3.x and 1.4) and IBM > and would seem to therefore be an OS issue. Can anyone provide me with more > info on this and is > there a workaround? > > > ---------------------------------------------------------------------- > To UNSUBSCRIBE, email to [EMAIL PROTECTED] > with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED] > > > ---------------------------------------------------------------------- > To UNSUBSCRIBE, email to [EMAIL PROTECTED] > with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED] > ---------------------------------------------------------------------- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
