摘要:
An apparatus and method to streamline updates to an index in a shared-nothing architecture. The apparatus may include an association module, a consolidation module, and a processing module. The association module may associate index update requests with an index. The index update requests may be configured to update the index in response to a redistribution of data in a data table. The consolidation module may compile the index update requests prior to processing and consolidate the index update requests into a smaller number of simplified index update requests to achieve substantially the same result. The processing module may then independently process the simplified index update requests and update the index accordingly.
摘要:
An apparatus and method to to decouple large object (“LOB”) data processing from main-line data processing in a shared-nothing architecture. The method may include relocating rows in a database table from a source partition to a target partition, where each row stores a source descriptor identifying a LOB associated with the row. The source descriptors may be read, and space sufficient to store each LOB in a target repository may be allocated accordingly. Source descriptors may be extracted from the rows, and sorted according to the location of the LOBs in the source repository to provide an ordered retrieval sequence. Each LOB may be retrieved from the source repository according to the retrieval sequence, and stored in its allocated space. The source descriptor stored in each row in the target partition may then be replaced with a target descriptor to identify the location of the respective LOB in the target repository.
摘要:
An apparatus and method to streamline updates to an index in a shared-nothing architecture. The apparatus may include an association module, a consolidation module, and a processing module. The association module may associate index update requests with an index. The index update requests may be configured to update the index in response to a redistribution of data in a data table. The consolidation module may compile the index update requests prior to processing and consolidate the index update requests into a smaller number of simplified index update requests to achieve substantially the same result. The processing module may then independently process the simplified index update requests and update the index accordingly.
摘要:
The invention relates to a method and apparatus of distributing data in portioned databases operating on a shared-nothing architecture. The invention provides two independent streams of data, where a first stream sends table data, and the second stream sends LOB data, to a receiving partition. The receiving partition re-couples the LOB data with the table data. Using two independent streams to distribute LOB data can increase data redistribution since large amounts of input/output is not performed to handle the sending of LOB data in a coupled fashion.
摘要:
The invention relates to a method and apparatus of distributing data in portioned databases operating on a shared-nothing architecture. The invention provides two independent streams of data, where a first stream sends table data, and the second stream sends LOB data, to a receiving partition. The receiving partition re-couples the LOB data with the table data. Using two independent streams to distribute LOB data can increase data redistribution since large amounts of input/output is not performed to handle the sending of LOB data in a coupled fashion.
摘要:
Techniques are described for data redistribution in a database system. Embodiments store a first set of receiver location identifiers mapped to a document identifier, the first set of receiver location identifiers assigned to a first set of receiver locations storing relational data referencing the document identifier. Embodiments further store a second set of receiver location identifiers mapped to the document identifier, the second set of receiver location identifiers assigned to a second set of receiver locations storing document data corresponding to the document identifier referenced by the relational data. Sender index keys are received separately from the relational data and the document data, the sender index keys referencing the document identifier and containing sender location identifiers. Embodiments determine, from the first set of receiver location identifiers and the second set of receiver location identifiers, which receiver location identifiers are mapped to the document identifier referenced by the sender index keys. The sender location identifiers are then replaced with the determined receiver location identifiers.
摘要:
Techniques are described for data redistribution in a database system. Embodiments store a first set of receiver location identifiers mapped to a document identifier, the first set of receiver location identifiers assigned to a first set of receiver locations storing relational data referencing the document identifier. Embodiments further store a second set of receiver location identifiers mapped to the document identifier, the second set of receiver location identifiers assigned to a second set of receiver locations storing document data corresponding to the document identifier referenced by the relational data. Sender index keys are received separately from the relational data and the document data, the sender index keys referencing the document identifier and containing sender location identifiers. Embodiments determine, from the first set of receiver location identifiers and the second set of receiver location identifiers, which receiver location identifiers are mapped to the document identifier referenced by the sender index keys. The sender location identifiers are then replaced with the determined receiver location identifiers.
摘要:
In one implementation, a method for sustaining database processing parallelism of one or more parallelized processes to build one or more index pages using pseudo record identifiers (PRIDs) is disclosed. The method comprises generating one or more PRIDs for one or more respective internal record formats (IRFs), inserting said generated PRIDs into a respective index key, generating a relational index of the PRID and a predetermined record identifier (RID) respectively determined in relation to said IRF, and building one or more index pages in relation to the relational index by comparing said RID and said PRID for their respective IRF in said relational index, and determining which one of said RID or said PRID to insert into said index page.
摘要:
The invention herein provides method and apparatus for managing attribute-tagged index entries contained in an index by assigning or tagging an index entry attribute indicator to a newly inserted index entry (that is, at the time when a key data value is inserted into the index, the key data value becomes a newly inserted index entry and an attribute indicator is tagged to the newly inserted index entry). The index entry can be placed anywhere within the index. Once a group of attribute-tagged index entries have been inserted into the index, the attribute identifiers can be switched off instantaneously (that is, the attribute associated with the attribute indicators is disabled).
摘要:
A method and system for optimizing data redistribution in a database. In one embodiment, the method includes moving, during a first scan, outgoing records from a sending partition to one or more receiving partitions, where free space is created in the sending partition due to the outgoing records leaving the sending partition. The method also includes filling, during the first scan, some of the free space with remaining records that do not leave the sending partition.