Tiles

arrays
foundation
tiling
A tile is a group of cells and is the atomic unit of I/O and compression in TileDB arrays.
Note

It is strongly recommended to read the following sections before you learn about tiles.

What is a tile?

A tile is a group of cells, serving as the atomic unit of I/O and compression (see the Key Concepts: Compression and Key Concepts: Tile Filters sections for more information about compression and the other tile filters supported in TileDB). TileDB differentiates between the following tile definitions:

  • Space tile: This is defined upon array creation by specifying a tile extent per dimension, which partitions the dimension domain into equal segments.
  • Data tile: This is the actual collection of data values included in the tile and materialized on storage. In dense arrays, the space and the data tiles are equivalent. In sparse arrays, they may be different; space tiles define the overall data layout that sorts the data values on storage, but may contain empty cells that are not materialized. A data tile in sparse arrays is defined after the data is sorted and is determined by an extra parameter called capacity. All data tiles in both dense and sparse arrays have the same capacity (i.e., the same number of non-empty cells). In dense arrays, the capacity is inferred by the space tile, whereas in sparse arrays, the user specifies the capacity explicitly specified upon the array creation.

Section Key Concepts: Data Layout explains space and data tiles in more detail.

An illustration of a dense array fragment and a sparse array fragment. In the dense array, the space tiles are the same as the data tiles, whereas in the sparse array, the space tiles do not always overlap with the data tiles. An illustration of a dense array fragment and a sparse array fragment. In the dense array, the space tiles are the same as the data tiles, whereas in the sparse array, the space tiles do not always overlap with the data tiles.

In addition, it is worth mentioning the following terminology:

  • Logical tile: A logical tile (which can be either a space tile or a data tile) refers to the multi-dimensional cells of the array, regardless of how many attribute values each contains or how these are laid out on storage.
  • Physical tile: A physical tile can only be a data tile and always corresponds to the stored cell values across a specific attribute. This is the actual atomic unit of compression and I/O.

The above terminology forms the basis of other concepts and tutorials across the Academy.

Fill values

A user can populate a TileDB array partially and incrementally (see the Key Concepts: Domain section for the discussion on the non-empty domain). Therefore, the following scenarios are possible for the case of dense arrays:

  • A tile may be partially written.
  • A tile may be partially outside the array domain.
  • An empty tile may be read.

In all these cases, TileDB may need to write special fill values to tiles to indicate “empty cells”, or similarly return fill values for a query asking for unpopulated tiles. The following figure demonstrates some examples.

An illustration of the three scenarios where fill values occur in dense arrays. An illustration of the three scenarios where fill values occur in dense arrays.

TileDB supports the following default fill values per attribute type, but these can be set by the user when defining the attributes upon array creation.

Datatype Default fill value
TILEDB_BLOB 0
TILEDB_STRING_ASCII 0
TILEDB_STRING_UTF8 0
TILEDB_STRING_UTF16 0
TILEDB_STRING_UTF32 0
TILEDB_INT8 Minimum int8 value
TILEDB_UINT8 Maximum uint8 value
TILEDB_INT16 Minimum int16 value
TILEDB_UINT16 Maximum uint16 value
TILEDB_INT32 Minimum int32 value
TILEDB_UINT32 Maximum uint32 value
TILEDB_INT64 Minimum int64 value
TILEDB_UINT64 Maximum uint64 value
TILEDB_FLOAT32 NaN
TILEDB_FLOAT64 NaN
TILEDB_DATETIME_* Minimum int64 value
Note

In the case a fixed-sized attribute stores more than one value, all the cell values will be assigned the corresponding default value shown above.