Great find, Douglas. This is the type of initiative on the application side that will spur the development of new processing architectures that will complement the evolution of SANs and its many-to-many connectivity. So far we have seen IBM migrate novel internal switch technology from its supercomputers to its mid-range server product line. It is also working closely with rival Intel in making its SBB (Scalable Building Block) program work with IBM (Sequent) NUMA technology. HWP is also developing new crossbar switching technology on the server side to complement the pre-Infiniband internal switched technology used by Hitachi in its latest storage systems.
Its academic pedigree and key support by powerful CIOs may make this type of application technology more palatable to the traditional database powerhouses. Overall, IBM and Oracle are neck to neck, depending on the research house. IBM is dominant in mainframe databases but ranks second to Oracle in Unix and third to Oracle (#1) and Microsoft (#2) in the Windows market. Due to strategic missteps, Informix and Sybase continue to tread 2nd tier waters, but they still retain a significant installed base. It does take a while for this type of thing to take off, but this may assume the same type of significance in the storage business as the Berkley RAID papers (Patterson, Gibson, Katz) that led to the formation of the RAID (redundant array of inexpensive disk) industry that is expected to be a $100 billion a year market in 2005, twice the size of the server market in less than 2 decades after cogent abstraction.
Here's more on DSTP:
ABSTRACT
Currently, data storage on all platforms is executed in an ad hoc fashion. Even as new applications are created, new formats for storing data associated with those applications are created. This creates an enormous challenge to other users and applications that wish to access this data, but do not wish to be constrained to a specific platform or application. As HTTP, HTML, web servers and browsers introduced a way to share documents across different platforms, DSTP, DSML, and data servers and clients introduce a platform independent way to share data over a network. DSTP relies only on data storage concepts (currently columns and rows), and is independent of the type of data storage used, whether it be files, database, or a distributed data warehouse structure. DSTP makes it possible in one location to locate, access, and analyze data from several other locations. DSTP also reduces the dependency on the data file, because it correlates data based on common keys in different data sets. DSTP allows the true conceptualization of a data 'space.'
ncdm.uic.edu |
