We already have a 2 separate Win2012 Mirror-2 Storage Clusters (in production) and had noted that performance was very poor.
We were looking to upgrade to Win2012R2 and storage tiering, we wanted to understand why the storage systems were performing so poorly. So, we have not built out a new storage cluster, with 2 nodes, 3 JBODs and 24 x2TB Seagate ES2 SAS drives.
We have run IOMeter to measure IOPs and the numbers are very poor, regardless of the 'geometries' we choose. Our Write IOPs results for 16KB block (this our typical DB page size) are:
- 2 columns, 64KB Interleave = 1043
- 2 columns, 256KB Interleave = 1176
- 4 columns, 64KB Interleave = 1424
- 4 columns, 256KB Interleave = 1688
- 8 columns, 256KB Interleave = 1677
We only tested Write IO because this should be the slowest/worse case IOPs values.
We also tested each HDD individually and they range between 270 and 330 IOPs.
By our math, the IOPs values should be in the range of 3000-3600 == 24 HDDs x IOPs/HDD div by 2 [for mirror-2], but the tested are showing values which are barely 50% of this.
Any assistance/tips would be appreciated.