Hi *,
for a change, I just wanted to share some positive experience with
mclock. On a customer cluster (recently upgraded to Reef), I drained
one host with 23 HDD OSDs (1,1 TB size). And to my surprise, it took
less than an hour until all PGs were backfilled. The OSDs weren't too
full, I believe something around 1,6 TB had to be moved away from this
host. But the recovery rates were quite nice, especially compared to
what other users on Reef have been reporting. I saw values between
several hundred MB to 2 or 3 GB per second. Most of the configs are
default, especially with regards to mclock there haven't been any
changes. Although the general recommendation is still to switch back
to wpq, I haven't done that yet in this cluster. I usually wait until
I encounter any issue pointing to mclock. But since this was a
positive experience, I'll keep mclock enabled until otherwise required.
Have a nice weekend!
Eugen
_______________________________________________
ceph-users mailing list -- ceph-users@xxxxxxx
To unsubscribe send an email to ceph-users-leave@xxxxxxx
