Sloan Digital Sky Survey - Object Oriented Databases vs Relational Databases
Object Oriented Database vs Relational Database
Sloan Digital Sky Survey
(Geographic Information System)
The Sloan Digital Sky Survey (SDSS) is a $32M project run by a consortium, led by Johns Hopkins Space Telescope Science Institute and Chicago's FermiLab, to produce a new, digital survey of all the objects (stars, galaxies , quasars, etc.) in the sky. The previous survey, now over 50 years old, consists of analog glass photographic plates, which dramatically limit distribution, amount of information, and accuracy of information. SDSS has constructed a new 2.5m telescope, based directly on digital (CCD, 4Kx2K chips) imaging, which will gather information including luminosity, spectra and spectral intensities in various bands, and variability, for all objects in space. It is expected to take 5 years to map the northern celestial sphere. Raw information will go into the database, as well as analysis results that collect that information into identified objects, resulting in a catalogue of 100M objects. The database will be accessible to scholars, and parts of it even to schools and the public, via the Internet. Production software entered test usage in 1996, and the final database is expected to be approximately 40 TB, in Objectivity/DB.
The key to use of this enormous archive is supporting multidimensional indexing for queries not only in 3D space, but also in luminosity in several spectral ranges. To make this efficient, SDSS developed a new indexing algorithm, based on modified quad-trees, for n-dimensions. This index partitions all 100M objects into some 40K containers, a clus tering device in Objectivity/DB. Coarse-grained maps of the multidimensional space then model the contents of these containers. Queries, then, quickly index over this coarse grain map to quickly determine which of the containers are needed. This efficiently reduces the search space, and hence query execution time, dramatically. Measured results show it to be 100x faster than traditional databases.
Fermi and Johns Hopkins evaluated other DBMSs, including RDBMSs and ODBMSs, and chose Objectivity/DB for these reasons:
- Objectivity's performance in large-scale applications
- Objectivity's support for platform heterogeneity (they can spread the very large database across multiple hardware from multiple vendors)
- Objectivity provides features (e.g. caching) that Fermi would otherwise have had to code themselves in their application
For more information, see
http://www-sdss.fnal.gov:8000/intro
http://tarkus.pha.jhu.edu/scienceArchive/
