Information Technology Reference
In-Depth Information
go to the buffer of the node nearest to the root until the buffer fills up, after which the
buffer is emptied and the tuples from it are distributed to the buffers of the children,
eventually reaching the leaves, if not deleted earlier.
The description of flash-memory-based solid-disk drives (
SSD
s) in Sect.
15.6
is
mainly from an article by Chen et al. [
2009
], who have conducted intensive exper-
iments and measurements on different types of state-of-the-art
SSD
s. Athanassoulis
et al. [
2010
] describe techniques for making effective use of flash memory in three
contexts: (1) as a log device for transaction processing on memory-resident data,
(2) as the main data store for transaction processing, and (3) as an update cache for
hard-disk-resident data warehouses.
B-tree variants designed to work efficiently on flash-memory-based
SSD
s include
the B-tree flash-translation layer or
BFTL
of Wu et al. [
2007
] and the Bw-tree of
Levandoski et al. [
2013
]. The performance of the write-optimized B-tree of Graefe
[
2004
] on flash-memory-based
SSD
s is analyzed by Jørgensen et al. [
2011
]. Buffer-
tree-like index structures optimized for
SSD
s include the
FD
-tree of Li et al. [
2009
,
2010
] and the lazy-adaptive tree (
LA
-tree) of Agrawal et al. [
2009
].
Recent log-structured database systems that use the log as the sole data repository
include LogBase [Vo et al.
2012
] and Hyder [Bernstein et al.
2011
]. LogBase aims
to provide scalable distributed storage service for database-centric applications in
the cloud. Main-memory multiversion indexes are maintained for efficient access to
the relational data in the log. Hyder is designed for a shared-flash system in which
each transaction executes at one server of the system accessing the shared database
(cf. Sect.
14.10
), logs its updates in one log record, and broadcasts the log record
to all servers. Each server rolls forward the log against its locally cached partial
copy of the last committed state. The database in Hyder is stored in the log as a
multiversion binary-search-tree index in which updates on the leaves propagate up
to the root (copy-on-write).
