Grid storage seems like a neat idea. Plug your servers to a socket in the wall and have a sea of data bits come through from an infinitely scalable storage utility. Sun is even offering a grid storage service - store your data on Sun-hosted disks for some many dollars/Terrabyte. Both IBM and IBM are visibly active in the grid space and HP has its concept of smart cells.

I'm dubious about this idea. The grid idea is based on the electricity grid with amps of current flowing down the wires between the socket on your wall and the scalable electricity generating and distributing grid. So too will data bits flow down the wire. But I think there is a huge difference between amps and bits, a vast difference.

The significance of an amp of electricity does not depend on the previous amp or the next amp or the organisation of amps in the grid. An amp is an amp is an amp. But, in our storage world, a bit is not a bit like other bits. Yes, of course, it is either a one and zero and all bits with one in them are exactly identical to all other bits with one in them. The basic bit can be either one or zero and we use this information to build patterns. We don't want identical bits. We want bits in a relationship to other bits. We want bits with different significance to each other because of the information that sequences of bits hold.

Sequences of amps are just that, sequences of amps. Sequences of bits are not meaningless. Bit sequences are the key to making storage work at the many levels it has to work.

Sequences of bit values enable us to detect and delineate packets of information, blocks on a disk, file headers, database records, parity bits, check digits, SCSI commands, NAS data streams, SMIS commands, etc, etc.

We can't do this unless both ends of a storage 'grid' know what to expect. But then it wouldn't be a grid in the electrical utility sense. It would be a SAN or a NAS or a VTL or whatever, partly depending upon the level at which the outlet in the wall transferred bits - layer 1, layer 2 layer 3, etc. But even within layers we need to know what type of information is being transmitted. A SAN 'grid' infrastructure can't be a NAS 'grid' infrastructure.

This leads me to think that the term 'grid storage' is too vague to be useful. HP's grid seems to be a general purpose storage structure with smart cells providing storage functions such as SAN, NAS, archiving and so on. Sun's storage grid is a basic 'store your bits, any bits' on our disks for a fee. The storage grid IBM and CERN is developing could be loosely described as a vast wide-area file system. It can be argued that these three entities are so different as to defy description under one term.