- 06 October, 2004 11:40
Continuing our discussion on grids, today, we talk about what to expect from them and which offer the potential of adding both capacity and throughput with, ideally, limited additional management requirement.
Last time, we discussed what storage grids would probably look like. Their components would include storage, capability for both communications and processing, and enough memory to handle whatever demands are put on each particular cell within the grid. But how many of these components will be generic? Or to get right to the point - how much of this will lock you in to dealing with a single vendor, whose underlying technology is key to the environment and perhaps allows it to function as a "grid provider"?
Keep in mind two things. First, what comes next is only likely to apply to storage grids, and may have no relationship at all to compute grids. And second, because we are talking about the future, all of this comes from my spending time gazing into the crystal fish bowl - far more accurate than a standard crystal ball at the county fair it's true, but still noticeably short of being infallible.
A natural assumption is for us to expect that grid cells (and indeed, the concept of grid as well) will, in the course of time, evolve through several phases, but I think it is pretty clear that they will begin as a series of essentially identical nodes from a single vendor. This of course means the earliest appearances of storage grids will likely be completely proprietary systems. As we have discussed in an earlier column, vendor lock-in of this sort may not be such a bad thing if the proprietary environment provides a level of service (via a high performance file system or some other advantage, for example) that would be otherwise unavailable.
Eventually, grid cells are likely to become quite flexible, resembling an interconnected network of standardized, modular building blocks. The contents of these modules will be able to be swapped out as the cells are repurposed. In such cases, for example, masses amounts of memory might be added to support specific application requirements or a serial-attached SCSI (SAS) storage array might be demoted to serial ATA (SATA).
Ultimately - if the vendors do it properly - very little will be proprietary within each cell; storage will be able to come from any vendor, and IT managers will have the opportunity to continue playing their vendors against one another just as many of the more clever ones do today. I fully expect however, that the part of the puzzle that provides the fundamental grid management - mostly software but perhaps, including some proprietary hardware as well - will always come from a single vendor.
Grids then will represent a danger of sorts to some managers, and because of their proprietary aspects (or simply because you don't need the value-add they have to offer), storage grids may well prove to be inappropriate for many of you. For the rest, grids are likely to provide a significant part of the infrastructure of larger IT environments and will likely find good use both for large sites getting into information lifecycle management and most particularly for those of you who will need to provide on-demand ("utility") storage services in a cost-efficient manner.