Hi, this should be a simple question:
Are the read operations against a RAID1 mirrored volume performed in a striped manner to increase read throughput? Or are the reads operations performed only form the primary disk (with the secondary laying dormant / discarded / spare)?
I see the "I/O request" counter increasing on both primary & secondary, but it's not clear to me whether RAID1 offers a read speed benefit.
I ask because I'm evaluating whether a 3x disk RAID5 volume would generate better read performance vs. a 2x disk RAID1 mirror -- or whether they should produce -- effectively -- the same read speed. Note: I'm putting aside differences in write performance (I understand that difference between RAID1 & 5).
Thanks!
Its not a simple answer, however.
Yes SoftRAID performs "striped reads" on Mirror volumes.
the problem is because of how disk drives work, it does not always generate additional performance. On Flash media, you will definitely get reads close to 2X writes.
but disks are designed to "prefetch" data, and when doing striped reads, unless you hit the correct boundaries, data has to be thrown away on reads. So the net result is no gain in performance.
We actually modified the SoftRAID driver at some point in version 3 to optimize for this. However it turned out every brand/model of drive behaved differently, so our optimization only worked on a single series of drives, which was useless.
So the complicated answer is there is no net gain to striped reads.
aha -- I understand, thanks. So, should I expect to see a 3x disk RAID5 volume (with 2x data; 1x parity) perform faster than a 2x mirror volume? Based upon your response above, I would think the answer is 'yes', 3x disk raid5 should outperform raid1 on reads.

