Messages in this thread | | | Date | Sun, 18 Dec 2011 19:16:44 +0200 | Subject | Kernel migration thread cpu usage is crazy in 3.1.5 | From | "Darius D." <> |
| |
Hi,
there has already been a report about migration thread CPU usage craziness in 3.2 ( https://lkml.org/lkml/2011/11/17/375 ), but it is in 3.1.5 as well.
problem manifests itself like this: after some time top/ps starts reporting huge time for migration/X threads, adding them 1 at time with larger and larger values (sometimes already existing entry is "incremented" ), so after a while it looks like this:
47 root RT 0 0 0 0 S 0 0.0 1100:35 [migration/14] 14 root RT 0 0 0 0 S 0 0.0 1081:39 [migration/3] 50 root RT 0 0 0 0 S 0 0.0 978:44.59 [migration/15] 53 root RT 0 0 0 0 S 0 0.0 840:53.83 [migration/16] 74 root RT 0 0 0 0 S 0 0.0 779:58.12 [migration/23] 44 root RT 0 0 0 0 S 0 0.0 773:43.01 [migration/13] 41 root RT 0 0 0 0 S 0 0.0 539:50.44 [migration/12] 59 root RT 0 0 0 0 S 0 0.0 380:05.40 [migration/18] 56 root RT 0 0 0 0 S 0 0.0 293:07.90 [migration/17] 68 root RT 0 0 0 0 S 0 0.0 150:10.87 [migration/21] 26 root RT 0 0 0 0 S 0 0.0 145:13.37 [migration/7] 11 root RT 0 0 0 0 S 0 0.0 130:11.83 [migration/2]
sar and other tools are not showing any increased cpu activity ( no misreporting, as box is under very low load and power usage would shoot sky high if there was some well hidden cpu usage )
System is dual Xeon X5650 NUMA box. Last kernel it was running with same kernel config was 2.6.39.2. The only outside visible "change" i have noticed is this: In munin graphs during low activity periods there were ~1k Local timer interrupts, after booting 3.1.5 graphs are completely different, now there are just ~500 Local timer interrupts, but Rescheduling interrupts are ~400 ( used to be none on low load ). Is this related to crazy cpu usage problem ?
Darius.
| |