Current tiered storage architectures currently take the approach of migrating extents to the appropriate data tier based-up given I/O profile for that given extent. My solution fragments a 'hot extent' into segments, profiles each segment, and places that segment to the appropriate tier based-upon profiling metrics.
Multi-Tier Extent Striping for Migration of Hot data Segments
The advantage to this design over current solutions, is that extent segments with desirable profiling characteristics can be migrated to the tier with media to best support this from a performance and reliability perspective.
An example would be an extent with segments that contain high read-to-write ratios. This segment would be migrated to a tier such as SSD, where best performance and reliability define the metric. Another example would be migrating segments with high writes to cache tier.
A clear advantage of this solution over current solutions are:
Only writing portions of extent affected, thereby potentially reducing the size of the write. Under certain circumstances, faster data access, as segments can be ordered and gathered in parallel from multiple tiers .
Extended life of the SSD media tier if exists.
1
2
3