To meet the stringent SAP HANA storage Key Performance Indicators (KPIs) for throughput and I/O on Azure, it is necessary to aggregate the performance of multiple Azure Premium Storage disks. This is achieved by creating a striped volume using a software RAID, such as mdadm in Linux. RAID0 (striping) is the recommended configuration for both the /hana/data and /hana/log volumes. It combines the IOPS and throughput of all underlying disks without the performance overhead of parity calculations found in RAID5 or RAID6. The required data redundancy is provided by the underlying Azure Storage platform (Locally Redundant Storage - LRS), which maintains three copies of the data, making OS-level RAID for redundancy unnecessary.

A. RAID1: RAID1 (mirroring) provides redundancy but does not aggregate write performance, failing to meet the high throughput requirements for SAP HANA data and log volumes.

B. RAID5: RAID5 is explicitly not recommended for the write-intensive /hana/log volume due to the significant write penalty associated with parity calculations, which severely impacts performance and latency.

D. RAID6: RAID6 has an even higher write penalty than RAID5 because it calculates and writes two parity blocks, making it the least suitable option for high-performance database workloads like SAP HANA.

1. Microsoft Docs

"SAP HANA Azure virtual machine storage configurations"

Last updated: 01/29/2024.

Section: Linux LVM and mdadm: "For the SAP HANA data and log volumes

we don't recommend any RAID redundancy level. The reason is that Azure Premium Storage already provides a sufficient level of redundancy with three images of the disks... The main purpose of building a RAID volume is to stripe I/O against multiple Azure Premium disks." This statement directly supports using striping (RAID0) and advises against redundancy levels like RAID1

RAID5

and RAID6.

Section: Production recommended storage configuration: This section provides specific examples for different VM types

all of which use multiple disks striped together (RAID0) for /hana/data and /hana/log to achieve the required performance. For example

for an M32ts VM

it recommends "4 x P10" disks in a striped set for /hana/data.

Section: Considerations for the /hana/log volume: "For the /hana/log volume

the usage of RAID 5 is not supported." This explicitly disqualifies RAID5 as a correct answer.