He refers to it as a storage waterfall because of the way available capacity "cascades" as it is processed from left to right in the chart. Notice on the far right where the disk utilization (the amount of data written by an application as a percentage of total raw storage capacity) for databases is estimated at 12.5%.
Now 12.5% might seem unrealistically low, but Chris is not pulling this number out of thin air. In our pre-sales work with customers, we have seen disk utilization levels like this for databases running on competitor's products.
Chris' chart identifies several process steps involved in preparing storage to be used by applications. As these processes were not identified explicitly, I thought I'd give labels to them here.
- Raw (I think everybody understands this)
- RAIDed (RAID and or mirroring are applied, including possible unusable "leftover" space)
- LUNned (creation/configuration of exportable storage LUNs, including thin provisioned LUNs)
- Designated (configured and identified for use as a storage volume by a host system, including capacity reserved for snapshots and other "recovery copies")
- Allocated (prepared for use by a file system, database or some other sw that manages space allocation)
- Written (amount of data created in a file system or database, including reservationless snapshots)
- Reduced (Data reduction such as deduplication, which reduces the amount of written data)
To increase the efficiency of storage, you want the ratio of written data to raw capacity (often referred to as storage utilization) to be larger. Depending on the storage product, filing systems, applications, best practices and performance expectations involved, there will be different profiles for storage utilization. Those utilization profiles would likely be interesting to storage customers - not just for making purchase decisions, but for understanding how well they are exercising their current storage plants.
One of the best ways to increase the utilization profile of storage is to use thin provisioning, which is part of the LUN creation process. In Chris' chart, there is a step-down of capacity, but with thin provisioning, there is actually a capacity step-up. For instance a LUN could be created from 100 GB of physical storage and presented to a host system as a thinly provisioned 400GB LUN.
Another good way to increase the storage utilization profile is to use wide RAID striping which reduces the overhead of RAID parity as well as eliminating "trapped" storage that can't be easily RAIDed.
3PAR clustered storage systems have the most advanced thin provisioning technology in the industry and achieve high storage utilization and excellent performance for mixed computing workloads.