Disk requirements and precheck

This topic describes the disks performance requirements that must be met before you use IBM Storage Scale Erasure Code Edition.

In the IBM Storage Scale configuration, the network performance and the disk performance are the key factors for file system performance. I/O requests from file system require disks to provide service in parallel according to the erasure code applied in the declustered array of the recovery group. For example, a full block that is read with 8+3p erasure code would ask eight disks to provide data in the declustered array, a full block write with 8+3p erasure code writes data to 11 disks in the declustered array, and all data stripes are scattered across all disks between nodes in the declustered array.

IBM Storage Scale supports NVMe, SSD, and HDD type disks, and all disks in one declustered array must have the same type to provide the same rate of throughput.

Disk Key Performance Indicators are listed as follows (based on the random read performance of 128 K raw device logical block size):

  • The minimum IOPS performance on NVMe drive must be greater than 10000, and the average IOPS performance on NVMe drive must be greater than 15000.
  • The max latency of NVMe drive must be less than 20 msec, and the average latency of NVMe drive must be less than 1.5 msec.
  • The IOPS performance on SSD drive must be greater than 800, and the average IOPS performance on SSD drive must be greater than 1200.
  • The max latency of SSD drive must be less than 100 msec, and the average latency of SSD drive must be less than 20 msec.
  • The IOPS performance on HDD drive must be greater than 55, and the average IOPS performance on HDD drive must be greater than 110.
  • The max latency of HDD drive must be less than 1500 msec, and the average latency of HDD drive must be less than 150 msec.
  • The performance on drives of same type must not have more than 10% difference.
Note: Do not test write performance if you have valuable data on the disk. It overwrites the existing data on the disk.

You can use the ece_storage_readiness open-source tool to check the defined KPI. This tool is available on the IBM Storage Scale Tools GitHub (https://github.com/IBM/SpectrumScaleTools). Contact IBM® for further details.

Run the ece_os_readiness tool to ensure that the system has met the requirements before the disk performance testing. ece_storage_readiness tool tests read performance by default. Be careful to use the -w or --random-write option, which test writes performance. This option overwrites the existing data on the test drives.