How a grid works

Last time, we discussed the major expectations of a storage grid: that storage scales in a near-limitless fashion, that throughput scales at the same pace as does capacity, that all this capacity and throughput must always be available, and that the whole system must be easily manageable. Today we will look at how this is accomplished.

Last time, we discussed the major expectations of a storage grid:  that storage scales in a near-limitless fashion, that throughput scales at the same pace as does capacity, that all this capacity and throughput must always be available, and that the whole system must be easily manageable. Today we will look at how this is accomplished.

A picture that shows the topography of a grid is often drawn as lattice, with each intersection representing a node, or cell. Within each of these cells is the storage of course, but there is also a great deal more. 

The node, or cell, is the basic unit of the grid.  Because each node on a storage grid must support the added throughput that goes along with increased storage, each node also has a local processor devoted to I/O operations, cache (when necessary) to support the processor’s operations, and a communications link to the other nodes in the system. 

Each cell within the grid is connected to the other cells by multiple pathways (picture the lattice), and is part of a single virtual image that the user sees.  The virtualization software that looks across the grid will be capable of balancing loads across cells in addition to other management functions.

The flexibility of such nodes will most likely vary substantially from one vendor to another.   Some early implementations undoubtedly will be quite simple, while the more sophisticated ones will probably turn out to be extremely flexible in the kind of value-add they give to stored data. 

For example, HP’s implementation of the storage grid is based on a network of what it calls “smart cells,” each of which can use either proprietary or third-party software to be tuned to perform specific tasks (hence, presumably, the “smart” in smart cell).  Some specialized smart cells might, for example, be configured as NAS file servers, while others might be dedicated to providing block storage.  Yet other cells might be content sensitive, providing WORM storage or another type of service that addresses issues of regulatory compliance.  When the cells have to be repurposed, the grid management system will be flexible enough to change the “personalities” of these cells on the fly.

The management software within each smart cell can be dynamically loaded, and changed easily with a software download.

Repurposing cells is a way to maximize their value.  For example, a high performance cell might contain management software with data copy, access path management and other management software appropriate to its need.  As the cell’s hardware ages, the cell may be repurposed to serve an archiving function.  In such a case for instance, some of the existing software might be exchanged for data migration software.  As long as there is enough intelligence within the cell, cells can be allocated and reallocated to whatever tier of storage best suits the lifecycle management needs of the site.

How much of all this is likely to be proprietary and how much commodity product?  Good question.  Next time we’ll look at how proprietary these grids are likely to be.