Hi,
i'm observing i strange situation where i would be happy to get some
help or some hints. I hope mm can point me to the right direction.
On a KVM host i'm seeing very high i/o waits while no local disks are in
use at all. First i thought it is swap related but instead i'm able to
trigger it with echo 1 >/proc/sys/vm/compact_memory . So every time the
kernel things it needs to compact memory a lot of VMs got stalled (no
answer to ping anymore) and i see high i/o waits (40%-70%).
# free -m
total used free shared buff/cache
available
Mem: 643641 429948 212067 114 1625
209630
Swap: 7629 3383 4246
Brefore compact:
# cat /proc/buddyinfo
Node 0, zone DMA 1 1 0 0 2 1 1
0 1 1 3
Node 0, zone DMA32 109 93 64 45 37 25 5
4 1 3 402
Node 0, zone Normal 526 306 35 3 6 3561 4467
4188 3778 6159 40244
Node 1, zone Normal 192307 13290 2781 794 209 95 504
532 511 671 6743
After compact:
# cat /proc/buddyinfo
Node 0, zone DMA 1 1 0 0 2 1 1
0 1 1 3
Node 0, zone DMA32 109 93 64 45 37 25 5
4 1 3 402
Node 0, zone Normal 1019 482 117 354 79 43 28
196 226 3719 43219
Node 1, zone Normal 24093 1537 627 271 131 79 39
18 109 707 7009
After 5 min after compact:
# cat /proc/buddyinfo
Node 0, zone DMA 1 1 0 0 2 1 1
0 1 1 3
Node 0, zone DMA32 109 93 64 45 37 25 5
4 1 3 402
Node 0, zone Normal 742 319 44 89 44 14 10
29 220 3604 43302
Node 1, zone Normal 34186 5735 1363 409 170 91 42
21 110 597 7033
Not sure if it is related but i'm running ksmd as well.
Thanks a lot!
Greets,
Stefan
