Answering the first question.
Yes, the performance metrics is calculated on random request distribution.
About the second question.
In ZFS the performance is scaling on a vdev and not on a disk group level.
So, for example the stripe is populated on vdevs, and therefore the best performance can be achieved with the disks used as singles, but there won't be any redundancy of course.
The next performance tier would be to use mirror groups, so we have a redundancy on 1 disk in a group.
More complicated raidz's are giving a better redundancy with the cost of a performance.
Thank You, T-Ku, thus, the performance index reflects scaling by the number of disk groups (vdevs, https://www.freebsd.org/doc/handbook/zfs-term.html). Still the disk group performance remains unclear. The calculator help states "The disks in a data group work in unison, which means that their effective performance is the same as one disk, no matter how many there are", which is very ugly in comparison with traditional approach. May be, after more statistics will be available, the calculator will be much optimistic, closer to "Complex RAID Performance" chapter at https://www.storagecraft.com/blog/raid-performance/
Reply With Quote