On Thu, Apr 28, 2011 at 09:42:49AM -0500, Christoph Lameter wrote: > > > Can you show in some tests how the chance of deviations is increased? If > > > at all then in some special sitations. Maybe others get better? > > > > It's kinda obvious, isn't it? Do relatively low freq (say, every > > 10ms) +1's and continuously do _sum(). Before, _sum() would never > > deviate much from the real count. After, there will be @batch jumps. > > If you still need proof code, I would write it but please note that > > I'm pretty backed up. > > "Obvious" could mean that you are drawing conclusions without a proper > reasoning chain. Here you assume certain things about the users of the > counters. The same assumptions were made when we had the vm counter > issues. The behavior of counter increments is typically not a regular > stream but occurs in spurts. Eh? Are you saying the above can't happen or the above doesn't matter? -- tejun -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxxx For more info on Linux MM, see: http://www.linux-mm.org/ . Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/ Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>