Browse Prior Art Database

Defining a Hot Level for Excluding Tables from Data Redistribution

IP.com Disclosure Number: IPCOM000109708D
Original Publication Date: 1992-Sep-01
Included in the Prior Art Database: 2005-Mar-24
Document File: 2 page(s) / 105K

Publishing Venue

IBM

Related People

Li, SG: AUTHOR

Abstract

Disclosed is a method for defining a bar to exclude tables with work loads above the bar from being chosen for data redistribution. This bar value can initially be determined automatically based on the work load distribution. A user interface design that allows the user to modify this bar value manually is disclosed as well. In parallel database systems, a table-level data redistribution operation exclusively locks a table for adding new nodes or removing existing nodes. As load balancing is one of the primary goals of data redistribution, the system has to carry this goal in choosing the table for data redistribution. Intuitively and usually, choosing a table with a higher work load can achieve load balancing more efficiently. However, the chosen table has to be exclusively locked.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 52% of the total text.

Defining a Hot Level for Excluding Tables from Data Redistribution

       Disclosed is a method for defining a bar to exclude
tables with work loads above the bar from being chosen for data
redistribution.  This bar value can initially be determined
automatically based on the work load distribution.  A user interface
design that allows the user to modify this bar value manually is
disclosed as well.  In parallel database systems, a table-level data
redistribution operation exclusively locks a table for adding new
nodes or removing existing nodes.  As load balancing is one of the
primary goals of data redistribution, the system has to carry this
goal in choosing the table for data redistribution.  Intuitively and
usually, choosing a table with a higher work load can achieve load
balancing more efficiently.  However, the chosen table has to be
exclusively locked.  The locking will block any transaction accessing
this table.  It likely will hurt the database system's performance.
Therefore, it is not always wise to choose the tables with higher
work load for data redistribution.

      By calculating the total work load of each table, the work load
distribution among the tables becomes available.  In the
configuration used for table-level data redistribution, the system
administrator can select one of the following criteria for
determining the "hot" level of work load to exclude tables from
involving in the Node++ operation:
(1)  A percentage of the total system's work load. The hot level will
be (Total system's work load) * (the specified percentage).
(2)  A percentage of the total number of tables.  For the first N
tables that have higher work loads, they should be excluded.  The
value of N is determined based on the current total number of tables
and the user-specified percentage.  This does not specify the "hot"
level directly; instead, it specifies the "hot" tables based on their
work load ranks.
(3)  A specific value of work load.  Any table with work no less than
the specified value is excluded.
The above criteria are not the exclusive list; they just show the
most possible criteria for defining this "hot" level.

      When the user wants to modify the "hot" level determined from
the configuration, the user can invoke a user interface; one example
is using a dialog box to override the "hot" level.  This user
interface displays the work load distribution among tables to provide
the user with relevant information for a better decision.  The
display format of the work load distribution is not restricted to any
particular...