Hi!
For quite a while we have been observing that during raid "data-check"
(initiated by cron every saturday at 0:00) the free swap spaces drops to
0. It seems as if the data-check generates many I/O's that "fill" the
buffer cache, which pushes out memory to swap. Is this a reasonable
assumption? If so, wouldn't it be better if data-check related I/O
bypasses the buffer cache?
I know that these are just assumptions, here are some facts:
What happens is illustrated by this graph:
http://www.rolf-fokkens.nl/kvm01-swap.png
- It covers 7 days
- In weekends you see the drop in free swap
- everfy night swap drops too, this is because of backups
And the related syslog messages:
Mar 8 00:00:01 kvm01 kernel: md: data-check of RAID array md0
Mar 8 00:00:01 kvm01 kernel: md: using maximum available idle IO
bandwidth (but not more than 200000 KB/sec) for data-check.
Mar 9 12:21:22 kvm01 kernel: md: md0: data-check done
Mar 15 00:00:02 kvm01 kernel: md: data-check of RAID array md0
Mar 15 00:00:02 kvm01 kernel: md: using maximum available idle IO
bandwidth (but not more than 200000 KB/sec) for data-check.
Mar 16 10:42:04 kvm01 kernel: md: md0: data-check done.
Some additional information:
- The machine has a total of 32GB RAM and 32GB SSD SWAP
- The machine is used to host 12 KVM VM's which in total have 47GB RAM
- I know this is "living on the edge"..
- but it works fine.
- in general performance is OK
- based on SMART measurements the SSD will last another 4 years.
- we reduced swappiness to 10, but that didn't make really big difference.
I'm looking forward to any information that can make us understand
what's happening.
Thanks,
Rolf
--
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
