Migration on Tiered Clusters

  • If you've ever imported a large amount of data to a HC3 Tiered node cluster (a cluster with nodes containing SSDs), you've likely noticed an odd behavior. During data import the SSDs gradually fill up much faster than any HDDs, but then their utilization goes back down seemingly on its own over time. Why?

    On a tiered cluster all new writes to virtual disks by default go through the SSD tier in order to improve performance. As blocks are determined to be hot or cold (highly active or relatively dormant) they are tiered accordingly and will either remain on SSD or will be moved down to the HDDs.

    This default behavior of prioritizing all writes to SSD may not be desirable for large data migrations to the HC3 cluster. It is possible to circumvent the behavior by setting the SSD priority level (the HEAT Priority in the HC3 web interface) for the virtual disk to 0 during the data migration. When a HC3 virtual disk's HEAT Priority is set to 0 all new writes on the virtual disk by default will be written to the spinning disks, bypassing the SSDs. Once you are finished with the migration change the SSD priority to the desired level and the cluster will automatically detect the hot or cold blocks and begin tiering them appropriately.