Database Reference
In-Depth Information
This may require specialized algorithms and data structures to ensure that
objects are correctly matched, taking into account that their location infor-
mation may contain some bounded error.
7.6.5 Storage Layout
As mentioned before, it is anticipated that most data will be bulk loaded into
SciDB from input sources such as scientific measurement devices or sensors.
The data will be linearized based on some ordering of the dimensions (typi-
cally, time will be the most dominant dimension) and will be initially stored
in the compute node's memory. Due to the large volume of data, it will need
to be written into disk buckets that will contain rectangular chunks of the
array. Keeping track of the location on disk and contents of these buckets will
require a data structure such as R-tree. Algorithms for determining optimal
shapes of buckets (size of stride in each dimension), bucket compression, as
well as deciding when to merge several buckets into a larger one, are still open
research issues.
Acknowledgments
The work on the MonetDB/SkyServer project was supported by the Bsik-
Bricks and MultimediaN programs. We would like to thank all members of the
CWI database team in the past decade for their joint efforts that made Mon-
etDB into a successful open-source database product. The work on the section
describing XLDB/SciDB was supported by the Director, Oce of Science, Of-
fice of Advanced Scientific Computing Research, of the U.S. Department of
Energy, under Contract No. DE-AC02-05CH11231.
References
[1] MacNicol, R., French, B.: Sybase IQ multiplex—designed for analyt-
ics. In
Proc. of the 30th Int. Conf. on Very Large Databases
, Toronto,
Canada (2004).
[2] http://www.sybase.com/products/datawarehousing/sybaseiq. Accessed
May 22, 2008.
[3] Svensson, P.:
Contributions to the design of e
cient relational data base
systems. Summary of the author's doctoral thesis
. Report TRITA-NA-
7909, Royal Institute of Technology, Stockholm (1979).
