Choose a vault optimization
When creating a Concentrated Dispersal vault, the first and most notable difference from creating non-Concentrated Dispersal vaults is that the page no longer prompts for a particular IDA Width or Threshold. Instead, one is (in some cases) presented with an option between a Storage Efficiency or a Performance optimization. In some cases, for example, when there is an insufficient number of disk drives available, only a Storage Efficiency option may be provided. Making this choice depends on organizational goals, but some guidance is provided below.
- When maximizing usable capacity, or minimizing cost per GB is a paramount concern
- When the use case involves writing large objects (1 MB or larger), there is less potential benefit from having the performance optimized
- When the use case is write heavy, and reads comprise a small fraction of the requests
- When the level of IO is not significant enough to stress the Slicestor devices or their drives
- When maximizing Read IOPS is a paramount concern
- When the use case involves writing small objects (100 KB or less)
- When the use case is read heavy, especially when the reads are random.
- When the level of IO is expected to be significant enough to stress the system
In general, the performance-optimized configurations can at most double Read IOPS in a Read IOP-constrained workload, but they sacrifice 10-20% of usable storage capacity. They achieve this trade off by using a lower IDA Width and a lower IDA Threshold, thus lowering the Read Penalty. However, the lower the IDA Width, the less storage efficient the system can be for a fixed level of fault tolerance (reliability).