IBM Storage Scale with TRIM-supporting SSDs and NVMe SSDs

IBM Storage Scale provides features that address the performance issues of flash memory.

File systems that create and remove files and directories often reuse storage blocks by overwriting the same storage blocks with new data contents. However, an NVMe or a general solid-state drive (SSD) device cannot overwrite a page of flash storage without first erasing the entire block of storage in which the page is located. This behavior creates a performance issue for I/O writes to previously used blocks of data when compared with I/O writes to unused or erased blocks. Also, write amplification occurs when the drive must save a copy of existing pages that are unaffected by the I/O write operation, then erase the entire block, and then restore the unaffected pages to the block after the erase.

To improve performance, the file system can issue a TRIM operation to an SSD notifying it about which blocks of data are no longer in use and can therefore be erased. The SSD erases the unused blocks before the blocks are required for the next reuse, which improves the performance of future I/O writes to the device. The TRIM operation also reduces fragmentation, because unused blocks are erased.

For more information about write amplification, garbage collection, and performance issues, see the vendor specifications for the flash memory devices that you are using.

%nsd:
  nsd=gpfs1nsd
  usage=metadataOnly
  pool=system
  thinDiskType={scsi-recl | nvme}