ZFS high cpu load with low io and cpu usage

Hi,

We are using zfs on one of our servers. Both deduplication and compression features of zfs are also enabled in it. We were trying to migrate a vps disk from an ext4 server using 'DD', but the migration process seems to be resulted in an increased load on zfs server and hence the migration couldn't able to complete. The load issue is fixed by rebooting the zfs server. We have a powerful CPU with 32 cores on zfs server. Didn't find anything helpful from 'iotop' and 'top' commands outputs. It seems like all the 32 cores are not functioning during the migration process. Please help us.

Thank You !

You could try dtrace and find out what calls are going out and maybe where about in the pipeline the system seems to having trouble...

Might also be able to go the round about way and boot that vps up to clonezilla, create an image then move the image and turn around and deploy it.

two things stand out:

1: Dedup and compression can both eat up quite a bit of cpu, dependant on compression settings.
2: using dd to migrate something to zfs? how? zfs doesn't work like that as there's no central block device for it. Migrating something to zfs is usually done via send/recieve or directly copying files over.

ZFS deduplication should never be used as it is broken, no matter how much RAM or CPU you have. I suggest you rebuild your zpool, with not deduplication. Most likely, the issue is the dedup. You can not just disable it, you need to rebuild your zpool. But, always enable compression, as it often will increase performance (it is faster to read 100 blocks and uncompress them in RAM, than reading 200 blocks from disk)