com.bigdata.btree.data.package.html Maven / Gradle / Ivy
B+-Tree data record
A package for the B+Tree data records. This is a binary image which can be
processed directly in RAM. The on disk record is normally compressed using
the compression provider for the {@link com.bigdata.store.IRawStore}. The
in-memory image is a variable length immutable record encoding the persistent
data for a node or leaf of a B+Tree. The binary image supports random access
to all data fields, to the keys, and to the values. This means that the image
can not use "packed" or nibble coded representations of integer values and can
not use bit encoded representations of compression codes. However, the entire
record is typically compressed by the backing store using GZip or the like, so
the representation of the record on disk is quite compact.
Key compression and value compression are applied when the recorded is encoded.
The representation of the compressed keys or values generally consists of a
dictionary followed the code strings for the keys or values. The code strings
are byte aligned so they can be accessed easily in RAM without having to scan
through the image. Hu-Tucker coding is preferred to prefix coding for keys
because prefix coding requires a sequential scan to process the coded keys.
Huffman coding is preferred for values because it is fast and efficient and
the assigned codes do not need to preserve the lexical order of the values,
which is not true for the keys.