Monet And Its Geographic Extensions: A Novel Approach to High Performance GIS Processing

Overview: Monet is an extensible (type- and algebra-extensible) main memory database management system. A set of GIS extension modules (Point, Box and Polygon) are described to demonstrate their approach and its applicability to GIS databases. The GIS modules can be "loaded at runtime [...] to obtain a functional complete GIS server."

Background: It is important to note that Monet is part of a bigger project, the MAGNUM project, whose goal is to build a "high performance parallel GIS database system with ODMG compliant O-O technology, employing a spatial reasoning system for query optimization."

Focus:

1. Ability to deal with large data volumes typical of GIS databases
   • Monet's design focuses on memory management
     • All operations done in main memory
     • Large tables are mapped into virtual memory (which allows the system to access it as if it were in main memory) circumventing the need to provide additional code to handle memory/buffer management
   • It incorporates shared memory parallelism
     • Units for parallel execution = algebraic operators (not tuple- or segment-pipelining)
     • Implication: results completely materialized at each phase in the query plan
2. Extensibility to spatial data types
   • Monet employs an extensible tool-kit approach (resembling Gral) in which an extension is encapsulated in a module (defined by an ADT interface)
   • Monet provides support for user-defined data types, search accelerators and primitives
   • GIS data types implemented:
     • Vector data model - only point and polygon datatypes implemented (what about lines?)
     • Raster data model - implemented as 'unary tables' and not as distinct data types (how are pixel sizes recorded in this format?)

• Note: spatial query optimization not a main concern

Evaluation:

• It is evaluated against the Sequoia Benchmark (i.e., the Sequoia 2000 storage benchmark for GIS)

Concerns:

• Main memory:
  • Can even large raster databases fit into main memory as required by Monet?
  • For example, the authors state that they "did not have sufficient disk space to generate all 130 rasters for the National benchmark" (Section 5)

Questions:

• What's a database hot-set?
• What are typical sizes for GIS dbs? Raster files? Vector files?
• What's a 'Box' wrt GIS data?

References: (1)_Peter A. Boncz, Wilko Quak, Martin L. Kersten: Monet And Its Geographic_ Extensions: A Novel Approach to High Performance GIS Processing. EDBT 1996: 147-163 (2) http://monetdb.cwi.nl/

-- AprilWebster - 21 Feb 2007