In the early days applications talked to files on disk. The application processed the data in the files and the storage subsystem stored the files and the operating system provided access to them. Then along came databases. A database management system (DBMS) took over some of the responsibility of an application - finding data and sorting it and reporting on it - and some of the responsibility of the operating system - organising data within files - to produce a storage-focused application. A DBMS on its own was useless. You had to have applications written that used the DBMS to store data and find it for you.
A backup application is another storage-focused application and it delivered specialised backup container files to tape. Tape media is generally invisible to an operating system's filesystem. The O/S has tape driver software but this is low-level stuff like starting and stopping. Only the backup application knows about the actual data stored on the tape and how it is laid out. Unlike a DBMS backup software does not need another application to drive it. End users can do that.
When EMC introduced Centera a new class of storage-focused application software was introduced. What a DBMS does for structured data Centera does for unstructured, fixed-content data and it needs an application written to drive it and use it. The Centera's main software isn't in the application servers though. It is in Centera itself which has its own embedded server. Specialised processing was added to the drive array and software written for it and this provided the Centera silo. EMC partners wrote the application server-based applications to use Centera.
Now Sun is set to introduce Honeycomb, a kind of super-NAS, with even more processing power added to the disk arrays. It opens an era in which the storage facility does even more of an application's work in locating and processing data. Honeycomb will be able to search for data amongst its unstructured data stores, using a variety of metadata to do so.
Okay. How does Honeycomb know what metadata to produce? How does Honeycomb know what metadata to use to find subsets of the data it stores? It's going to need application server-based software to do this. In other words it looks as if, like Centera, Honeycomb is going to need specialised application software written to drive it.
My feeling is though that Honeycomb, unlike Centera, doesn't have a natural class of data for which it is very well-focused. Centera has reference data, like HP's RISS. DBMS' have transaction data. Backup software has backup data. What class of data does Honeycomb have?
I think part of the problem Sun has in bringing Honeycomb to market is understanding how to target it? This targeting issue could well be making the job of producing the application server interface to Honeycomb difficult. It is going to be interesting to see how Sun pitches Honeycomb capabilities when it does come to market.