10 Gb/s =~ 1.2 GB/s Each 980 Pro can do sequential reads at 7 GB/s. ZFS with record size 128 KiB will do IOs at a size that will perform about as well as seque

10 Gb/s =~ 1.2 GB/s Each 980 Pro can do sequential reads at 7 GB/s. ZFS with record size 128 KiB will do IOs at a size that will perform about as well as sequential regardless of the actual pattern. AFAICT ZFS has no need to push your drives beyond 10% of their overall capability assuming the clients are performing sequential reads or the reads are at least 128k. A more stressful read test would be a 4k random read where the working set doesn’t fit in the ARC. This would trigger a lot of read inflation. The write side is a bit more complicated as the drives may be able to write at over 5 GB/s if they have 30+% of their NAND erased, else the write rate will be closer to 1.4 GB/s. With a mirror this is pretty close to the network bandwidth. With some reads mixed in for COW the drive could become a bottleneck before the network or ZFS.