Method and apparatus for achieving uniform data distribution in a parallel database system

Claim 9

A computer product for use on a computer system for distributing data of a table partitioned across a parallel database system having a number of nodes comprising: program code means recorded on said medium for instructing said computer system to perform the steps of; determining the data volume for nodes of said system associated with said table; comparing said data volume stored among said nodes; identifying groups of data in nodes having higher data volumes which may be distributed to nodes having lower data volumes to obtain a more uniform data distribution with minimum required data movement activity; moving said identified data to said nodes having lower data volumes to balance the data volumes of said nodes across which said data is partitioned; wherein the smallest group of data identified for redistribution from one node to another comprises a subpartition of said data in said node; obtaining file size information for table partitions of said nodes by reading file attributes for said files and obtaining database statistics on data group volume and volume of data group usage; generating a distribution listing file depicting current data distribution; selecting one or more data groups for redistribution among selected nodes to which data is to be redistributed; generating a partitioning map for redistribution of said groups of data in which a best fit method has been applied to select data groups and a redistribution plan for redistribution according to descending weight of said data groups; a redistribution plan in which data will be uniformly distributed among said nodes; and redistributing said data groups among said selected nodes in accordance with said partition map.

Claim 10

A computer product for use on a computer system for distributing data of a table partitioned across a parallel database system having a number of nodes comprising: program code means recorded on said medium for instructing said computer system to perform the steps of; determining the data volume for nodes of said system associated with said table; comparing said data volume stored among said nodes; identifying groups of data in nodes having higher data volumes which may be distributed to nodes having lower data volumes to obtain a more uniform data distribution with minimum required data movement activity; moving said identified data to said nodes having lower data volumes to balance the data volumes of said nodes across which said data is partitioned; wherein the smallest group of data identified for redistribution from one node to another comprises a subpartition of said data in said node; obtaining file size information for table partitions of said nodes by reading file attributes for said files and obtaining database statistics on data group volume and volume of data group usage; generating a distribution listing file depicting current data distribution; selecting one or more data groups for redistribution among selected nodes to which data is to be redistributed; generating a partitioning map for redistribution of said groups of data in which a best fit method has been applied to select data groups and a redistribution plan for redistribution according to descending weight of said data groups; a redistribution plan in which data will be uniformly distributed among said nodes; and redistributing said data groups among said selected nodes in accordance with said partition map.

Claim 11

A computer program product for use on a computer system for distributing data of a table partitioned across a parallel database system having a number of nodes comprising; a recording medium; means recorded on said medium for instructing said computer system to perform the steps of; determining resource loading at nodes of the system associated with said table; comparing resource loading among said nodes; if said resource loading is distributed in a significantly unbalanced manner; (a) selecting subpartitions contained within partitions of said table at said nodes having heavy loading for movement to nodes having lower resource loading; (b) selecting subpartitions for retention at said nodes having heavy loading; moving said subpartitions selected for movement to said nodes having lower resource loading to balance the resource loading among said nodes containing partitions of said table; wherein transaction activity information is obtained for all table partitions in said nodes by reading transaction logs of said database system; generating a current workload distribution file; selecting data groups from nodes having excessive workload distribution for redistribution among selected nodes to which data is to be distributed; generating a partitioning map describing a plan of redistribution of said groups to achieve uniformity of workload while minimizing the amount of data transferred between said nodes to achieve said redistribution; and redistributing said selected data groups.

Claim 12

A computer program product for use on a computer system for distributing data of a table partitioned across a parallel database system having a number of nodes comprising; a recording medium; means recorded on said medium for instructing said computer system to perform the steps of; determining resource loading at nodes of the system associated with said table; comparing resource loading among said nodes; if said resource loading is distributed in a significantly unbalanced manner; (a) selecting subpartitions contained within partitions of said table at said nodes having heavy loading for movement to nodes having lower resource loading; (b) selecting subpartitions for retention at said nodes having heavy loading; moving said subpartitions selected for movement to said nodes having lower resource loading to balance the resource loading among said nodes containing partitions of said table; wherein transaction activity information is obtained for all table partitions in said nodes by reading transaction logs of said database system; generating a current workload distribution file; selecting data groups from nodes having excessive workload distribution for redistribution among selected nodes to which data is to be distributed; generating a partitioning map describing a plan of redistribution of said groups to achieve uniformity of workload while minimizing the amount of data transferred between said nodes to achieve said redistribution; and redistributing said selected data groups.