HP has invented a big idea for storage. There are three main parts to the idea: smart cells; re-programmable cells; and a single management pane. Have a grid of smart cells interconnected by Ethernet and Fibre Channel. Each cell has a processor (controller), network interface, storage media and application software. The cell can be a file serving or NAS cell; it can be a block-serving or SAN cell. It could be a backup cell, an archive cell, or a content-addressable cell. HP has a list of some forty storage-related functions cells could provide.
The company has gone further. Cell software - its 'personality' - can be changed so that its storage function can be changed. A NAS cell could become a SAN cell. The grid could be dynamically reconfigured so that, if you needed more SAN capability and less NAS functionality then you dynamically 're-program' your grid.
You can scale up this grid by adding more smart cells. You can add capacity and functionality separately or combined. Off to one side there is a single management facility so that this storage grid can be managed as an entity, as a single image.
These are the basic conceptual building blocks of HP's storage grid. We have a couple of basic questions. Why would you want to do this? How is it done? There are lots of detailed inquiries to be made as well; about interconnects and bandwidth and virtualisation and data protection and so forth. Is there a single file system is another query. (IBM's Storage Tank concept envisages a single file system shared by all accessing servers.) But let's keep a focus on the main ones as we explore the HP storage grid a little further.
Smart cells are built from commodity hardware and standards are used for interconnects, management, etc. They possess a common software base on which the 'personality' software is loaded.
The collection of smart cells constitutes a storage grid; the cell controllers are federated together via the storage grid fabric infrastrucure, and it can be connected to a existing infrastructure by linking it in to a port on a switch or router. Like a storage utility the grid delivers a requested storage service. The services are represented by an Open Grid Services Architecture (OGSA), meaning grid/web services APIs.
The services carry out functions such as storing and accessing data, manipulating it in the grid, protecting it, indexing and scanning it and presenting it to applications.
Cells with similar functions are grouped together into domains, such as a file serving or NAS domain. New cells added to the grid are automatically discovered and added to the relevant domain depending on their functionality (personality). Domains are provisioned automatically and dynamically; or manually via an administator. The grid is, as it were, an adaptive structure in line with HP's concept of the adaptive enterpise.
One aspect of this is that data can be placed and moved automatically between different tiers of storage.
Smart cell controllers have software to enable them to work together. The data paths are virtualised and any controller can handle any I/O. The smart cell software enables consistency and redundancy.
A new kind of smart cell will be developed to enable existing storage to be represented as a grid component to storage grid cells. This linking to current storage infrastructures and current storage aggregating cell or smart cell head provide a means to bring in storage grids to existing storage such that both can work together.
An Enterprise Storage Group report on the HP grid idea says that data is chunked into unique pieces which are only stored once so as to minimise capacity needs. This presumably uses some kind of hash addressing scheme, much like that used in EMC's Centera and other content-addrssable stores such as the one from Archivas.
Why have a storage grid?
HP says that its storage grid will be easier to manage than scaling out today's mixed and separate sets of storage components. A storage grid will hold more data and make it more available, wherever it is needed. Its dynamism and self-managing features weill enable administrators to better cope with the complexity within, and it is undeniably complex within, because it is presented as a single logical image and because its components are built from the ground up to federate together and contribute to the management functions.
The HP storage grid information says nothing about a single file system operating across the grid. In this respect it is quite different from IBM's Storage Tank. It also doesn't provide much detail on virtualisation, apart from saying the data paths are virtualised and that data is stored in chunks. This chunking of data means that a data item, such as a mortgage record, database transaction, Word document or image, is represented virtually. It consists of pointers (metadata) to its component chunks. It's reconstructed from the stored chunks whenever it's needed.
There is also little detail on information lifecycle management and compliance. But, no doubt, these will be filled in with ILM smart cells and compliance cell personalities such as e-mail archiving.
HP's RISS or Reference Information Storage System is a shipping example of a smart cell and is, HP says, an actual storage grid in its own right. There is a roadmap to the storage grid future with much emphasis on co-operation with the installed storage infrastructure and investment protection and migration and so forth.
But there is no way you can decide now that the HP storage grid is the way to go. There is no proof that it will be easier to manage and there is no real evidence that it will perform as fast as and better than existing storage architectures.
However it is a brave attempt at synthesising an identifiable storage vision distinct from that of IBM and, perceptually at least, well in advance of those from EMC, Veritas and StorageTek. It is well worth paying attention too. Check out HP's web site and have a look at a technical white paper here.