ZFS ZFS mirror performance vs. single disk

[jax]

Hello all,

I am looking for some general pointers on how to test I/O performance on a ZFS mirror vs. a single disk. I've been expecting to see higher read performance from a mirror but I am unable to verify this on my test machine.

Specifically I've been creating pools using:

# zpool create foo /dev/ada1

and

# zpool create foo mirror /dev/ada1 /dev/ada2

as well as

# zpool create foo mirror /dev/ada1 /dev/ada2 /dev/ada3

On these I've created a dataset on which I've been running some different benchmarking tools: fio, randomio and also plain dd reads/writes but I don't see any difference in performance whatsoever. All the disks in questions are "300.000MB/s transfers (SATA 2.x, UDMA6, PIO 8192bytes)". The server in question has 8GB of memory.

I am far from being an expert on BSD or ZFS and I am hoping someone could give me some pointers on a testcase that would produce higher read performance on the mirrors compared to the single disk. With fio and randomio I haven't been able to create a test that gives me different results on the mirrors compared to the single disks.

Any pointers or help would be greatly appreciated.

Thanks

 

[_CiPHER_]

Show us your benchmarks!

With performance, do you mean IOps performance or do you mean sequential I/O performance?

ZFS should yield higher-than-single-disk performance for both random read IOps and sequential read throughput. In fact, i have seen ZFS come close to RAID0 in terms of sequential read on a mirror, which is pretty awesome as RAID1 with load balancing is the only one to come close to this kind of performance result.