Storage grids potentially represent a massively available hardware infrastructure that can be placed beneath extremely large virtualized storage environments, but if they are to provide the greatest effectiveness their assets must also be capable of being reassigned (provisioned) dynamically as needs change. As part of this series on storage grids, we have often mentioned this need for dynamic provisioning, but perhaps it is time to look at some of what this might involve.
Even the earliest grids, if they are to be effective, will require that provisioning be driven by automated policies rather than "hand tuning." Such early grids will consist of what are essentially identical (or at least very similar) cells, and although management systems will handle these individual cells within the grid as discrete entities, management also will mostly treat them monolithically, adding and subtracting cells as a complete unit rather than bit by bit. Additional storage will thus be aggregated to processes on a cell-by-cell basis in an action that is purely quantitative.
With more sophisticated grids however, provisioning cells will go beyond just adding more devices and will begin to take on a qualitative dimension as well.
As storage grids grow in sophistication and as large enterprises come to rely on them to increasing degrees, many of the cells are likely to become optimized in favor of certain functions and will begin taking on "personalities" of their own. An implication of this is that, as cells are provisioned to one specialized process over another, the personalities that define the cells may also come to require personality adjustments. In such cases for example, a cell that had previously been performing a file serving function for video-on-demand (VOD) might be reallocated to perform near line storage duties. In addition to swapping out high performance disks for cheaper ones (or perhaps just reusing the original - but now obsolescing - disks), the large read ahead buffers used for VOD would no longer be necessary and would either be physically removed or perhaps dynamically reassigned.
When this occurs management functionality would be dynamically loaded (again, controlled by policies) and needed adjustments would be implemented in a theoretically straightforward manner through software downloads.
What is pretty clear is that storage grids, if they have any hope of living up to their potential, will have to support all the storage tiers needed to manage the information lifecycle, and will need the ability to change dynamically (but fortunately, not yet organically) to changing needs and changing hardware assets. Grids will evolve to become the infrastructure of the largest on-demand systems.