A Database Language More Suitable for the Earth System Sciences - Towards an Interdisciplinary Approach in Earth System Science - page 216

combined with a variety of other, non-array types of data in the attempt to answer
complex questions about Earth s past, present and future. Harnessing the potential

of all this data is still not so trivial for Earth System Scientists. This is largely
because the use of databases

'

— a central component of Big Data systems — has been

con ned to metadata: small, structured, and queryable data describing the actual

data, which remained un-queryable. This data is nowadays maintained in ad hoc
solutions crafted by data centers, with functionality often constrained to le

download.
Hence, database support for massive multi-dimensional arrays is getting into
focus. Array models speci cally for databases have been published since a while

(Cornacchia et al. 2008 ; Lerner and Shasha 2003 ; Libkin et al. 1996 ; Marathe and
Salem 2002 ; van Ballegooij 2004 ), however, finding general attention by the

database community mainly since the appearance of the NewSQL movement. The
currently most in

uential models are (in historical order) rasdaman (Baumann
1994 ), SciQL (Zhang et al. 2011 ), and SciDB (Stonebraker et al. 2011 ). Notably,
ISO SQL ( 1999 ) already supports arrays, albeit only very rudimentarily. Arrays are
con

fl

ned to 1-D, and without any implicit nor explicit loops there is no practically
useful operational support. Nonetheless, the standard offers a suitable hook for
injecting array semantics into SQL that would coexist with its set semantics.
In this contribution, we present an overview of the array model ASQL (Sects. 2
and 3 , detailed description in Misev and Baumann 2014 ). Based on the ISO SQL
stub, ASQL provides a fully-

edged set of structural and operational array con-
structs completely integrated and compatible with SQL and orthogonal to its set
semantics. While based mainly on the conceptualization of rasdaman, ASQL also
honors recent developments in the array databases

fl

field, notably SciDB and SciQL.
The theoretical underpinnings of ASQL have been implemented and exempli

ed in
practice in ASQLDB, an open-source mediator system based on HSQLDB
(T. H. D. Group 2013 ) and rasdaman 1 (Sect. 4 ).

2 Data Model

Roughly, array models can be classi ed according to their relational embedding.

ISO SQL ( 2003 ), Array Algebra (Baumann 1994 , 1999 ), PostGIS Raster (Obe and
Hsu 2011 ), SciSPARQL (Andrejev and Risch 2012 ), and other models introduce
arrays as a column type which we call

. This makes arrays a
plug-into SQL with the overall set-oriented model unchanged. SciQL and SciDB,
on the other hand, follow a model where arrays are emulated by tables. Here, arrays
are at the same level as tables, and array cells correspond to tuples. We call this
“

“ array-as-attribute ”

”

modeling. With ASQL, we follow the array-as-attribute approach
as given by ISO SQL.

array-as-table

1

( www.rasdaman.org , Accessed on 22 Aug, 2013)