Rumors that Hitachi Ltd was working on its own object database (CI No 2,812) turn out to be true: at the Commercial Parallel Processing Conference in Chicago this week, Hitachi’s Computer Products America division has introduced and demonstrated Thor, a new object/relational database system aimed at the enterprise data warehousing market, and described by Hitachi as the only true parallel object-relational system that is intrinsically object-oriented. Thor, which rather awkwardly stands for The Hitachi Object Relational database technology, is currently in beta and runs only on specially designed hardware: it’s unlikely to emerge as a generally available product until the third quarter of next year. At its heart is the Thor ObjectManager, supporting such features as row-level concurrency control, user-defined data types and business rules, stored procedures and triggers, transaction logging and recovery, and database backup and recovery. It is fully object-oriented, says Hitachi, so that tables are classes and rows are instances of that class, although an optimizing relational engine presenting a single, logical image of the database is also provided. Dataflow techniques within the software break down complex data queries and optimize them for efficient parallel processing. ObjectManager itself could easily be ported to other platforms, says Hitachi, but at the moment the hardware component, called the Thor MPP Data Server, is a turnkey shared nothing system with up to 288 nodes using the PowerPC chip, PCI bus and SCSI II disk drives. It has a toroidal matrix of processor modules that’s said to be more efficient than the switched architecture of Hitachi’s SR4300, which is an IBM SP/2 re-badged. Hitachi says it was after a platform with a higher input/output bandwidth capability than the scientific and engineering biased systems out there today. The SR4300 is high on the list for future porting, however. ObjectManager is hosted on an embedded operating system, and accessed through an open gateway server such as an NT box, Sun workstation or IBM RS/6000 -anything that supports the Sybase Open Server, which Hitachi resells. There are also gateways to standard SQL and ODBC. The system is optimized for processing complex queries created by decision support applications on very large databases, typically characterized by high input/output activity and long searches with varying response times, and is scalable to handle data sizes from a few gigabytes to terabytes on hundreds of processors, says Hitachi. Thor was developed in the US under database guru Dr Robert Hartmann.
