Sun has announced its long-awaited StorageTek 5800, code-named Honeycomb as its fixed-content repository. Sun claimed the product would provide faster and better access to terabytes of unstructured information through metadata processing.

Honeycomb is a clustered set of nodes. Each 2U rack shelf node is, in effect, a SunFire 2100 server with four 500GB serial ATA (SATA) drives, an Opteron 64-bit processor and 3GB RAM. Up to 16 nodes, termed a cell, can be clustered using Gigabit Ethernet to provide a 32TB system. There are two load-balancing switches and a service processor per cell.

Data, ranging from complex media files to e-mails, is stored as objects and striped across all nodes. Faulty data on one node can be automatically detected and corrected. The disk drives have RAID 6 protection as well, meaning a 2-disk failure can be tolerated. Data is backed up to tape using Bakbone NetVault software and the NDMP protocol.

The nodes in a cell 'elect' a master node at start-up time. If it fails a new one is elected and, if any node goes down, the data is reconstructed and the other nodes pick up the workload.

The software stack is layered on Solaris and is being open-sourced, meaning zero licence fees. It contains extensive meta data creation and analysis features to aid searching. Applications can be written to use the software stack's capabilities through an openly-available API.

There is no file/folder system structure. Sun director of product management, Dave Kenyon, said: "Look at the Apple Mac's Spotlight feature. Who needs a file system?" He thinks that object data storage could become general purpose. A presentational file/folder system can be constructed using user roles and virtual views if desired.

A Sun spokesperson said that the 5800 is a third generation object store. Generation one is represented by EMC's Centera in which you: "store one copy of the data and farm it out on request through added application code."

Generation two adds a little intelligence to that. Generation three has the intelligence co-located with the storage to provide faster access and better object data preservation.

Honeycomb is not a large data store, front-ended by a fast application server head holding an index and managing the store. The 'head' functionality is distributed throughout the nodes and open source software removes the proprietary storage application licence costs.

Graham Lovell, a senior director at Sun, said: "The StorageTek 5800 system, a.k.a. 'Honeycomb,' is the first in a new line of third-generation object storage systems. (It) has raised the technology innovation bar -- allowing enterprises to create petabyte-scale, safeguarded, intelligent digital repositories."

Sun has already shipped more than 400TB of 5800 capacity to partners and early access customers, including Oxford University, Purdue University, Southampton University, Stanford University, and the University of Michigan.

A 16-node cell has a US$245,000 list price.

Also Sun has added to its June, 2007 Constellation high-performance computing announcement. There are three aspects to Constellation: the Sun Blade 6048 with up to 48 processing blades; the X4500 hybrid storage/server system (Thumper); and Magnum or the Sun Datacenter Switch 3456, a dense Infiniband switch with 3,456 ports. A Constellation system is being built at the University of Texas.