[SOLVED] How full a zpool can be in % ?

[fireon]

Hello,

how full a zpool can be, before you get performance problems? Is right to say 80% or is this depending on the storagedevice, like HDD or SSD. My SSD pool have 89% this time.

Thanks a lot

 

[mir]

The general rule of thumb is that performance degrade when pool is crossing the 80% full mark.

 

[spirit]

in past I had big degradation at 70%. (big fragmentation and new write take a lot of time to find free blocks)

 

[mir]

in past I had big degradation at 70%. (big fragmentation and new write take a lot of time to find free blocks)

Maybe the limit can be pushed by using log and cache. Of course hardware for log and cache should be of the right sort and quality.

 

[spirit]

Maybe the limit can be pushed by using log and cache. Of course hardware for log and cache should be of the right sort and quality.

Well , this was a nexenta san, with stec zeus ram as log device, and intel s3500 for l2arc.
24 15k drives.
I was around the limit of iops, but when reaching 70%, the perf degrage x5. (simply remove 1 or 2%,and performance was good again).

This was 4years ago, so maybe zfs has improve since.

 

[mir]

These new feature flags could have improved your situation:
1) async_destroy, especially with clones
2) hole_birth, especially for zvol
3) empty_bpobj, optimize snapshot handling
4) embedded_data, greatly improves performance for highly compress data

If I remember correct Nexanta 4 years ago did not use lz4 compression algorithm which typically increases compressed pools performance by 50 - 150%.

 

[fireon]

Thank you for the answers. I was able to delete some file to CAP 69% The ZFS version is in no subscriptionrepo 0.6.5.8. Maybe there is something better for CAP.

All my pools are lz4 compressed.