Calibration of logical cost formulae for queries in a heterogeneous DBMS using synthetic database

• US 5,412,806 A
• Filed: 08/20/1992
• Issued: 05/02/1995
• Est. Priority Date: 08/20/1992
• Status: Expired due to Term

• Current Assignee: Hewlett-Packard Company (HP Inc.)
• Original Assignee: Hewlett-Packard Company (HP Inc.)
• Inventors: Krishnamurthy, Ravi, Shan, Ming-Chien, Du, Weimin
• Primary Examiner(s): Black, Thomas G.
• Assistant Examiner(s): HOMERE, JEAN RAYMOND
• Application Number:
  US07/932,426
• Time in Patent Office: 985 Days
• Field of Search: 395/600, 395/800, 395/275, 395/100, 395/155, 395/200
• US Class Current: 1/1
• CPC Class Codes: G06F 16/2471 Distributed queries
• CPC Class Codes: Y10S 707/99932 Access augmentation or optimizing
• CPC Class Codes: Y10S 707/99939 Privileged access

Abstract

A programmable machine system and method for managing electronic data access among multiple different relational databases in a network distributed database environment. The machine is programmed so that it can construct cost-effective access strategies for any of the participating databases absent any DBMS-specific cost models. The system provides query optimization across different database management systems in a network distributed database environment based on a calibrating database relying only on typical relational database statistics and cost data is developed by running queries in the various databases against the calibrating database. A logical cost model is constructed using the resulting cost data and is used to estimate the cost of a given query based on logical characteristics of the DBMS, the relations, and the query itself. The cost of a complex query is estimated using primitive queries. Optimal query access strategies are thereby designed and used to control execution of the queries across relational databases controlled by two or more different database management systems.

Claims

1. A database access system for optimizing database queries in a heterogeneous distributed database system, the system comprising:

a first database machine incorporating a first relational database management system and accompanying first database;

a second database machine incorporating a relational database management system and accompanying second database;

the first and second relational database management systems being different but conforming at least to a predetermined structured query language (SQL);

communication means for electronic bidirectional communications between the different database machines;

means coupled to the communication means for sending and receiving an electronic message to and from any of the database machines, the message containing data defining a database query;

a data access logical cost model comprising logical cost formulae for optimizing queries in each database in the system;

a synthetic database for use in calibrating the data access logical cost model for each relational database management system in the distributed database system;

means for querying the synthetic data base on each database machine to determine cost coefficients for use in said logical costs formula to calibrate the data access logical cost model; and

means responsive to a database query for accessing each of the first and second databases of said first and second database machines in accordance with a least cost index obtained from said data access logical cost model.

2. A system according to claim 1 wherein the means for querying the synthetic database for calibrating a data access logical cost model includes:

a test suite of SQL queries for querying the synthetic database;

means for transmitting the test suite of SQL queries from the first database machine to the second database machine and determining cost model data for responses to the SQL queries; and

a system wide catalog for storing resulting cost model data.

3. A system according to claim 1 wherein the synthetic database for calibrating a data access cost model includes seven columns for any number (n) of relations containing 2ⁿ tuples, said columns, 1 through 7, having the attributes including:

Column₁ ;

integer {0, n}, indexed and clustered;

Column₂ ;

integer {0, 2ⁿ -1}, indexed and de facto clustered, but not specified to the DBMS as such;

Column₃ ;

integer {0, n}, indexed and unclustered;

Column₄ ;

integer {0, n}, having no index;

Column₅ ;

integer {0, 2ⁿ -1}, indexed and unclustered;

Column₆ ;

integer {0,2ⁿ -1}, having no index; and

Column₇ ;

a long character string to meet the size of the tuple requirement.

4. A system according to claim 1 including means defining a space of executions for a database query including a remote join across the first and second database management systems.

5. A system according to claim 1 wherein the data access logical cost model includes means for determining data access cost information based on logical information of the databases including coefficients defined as functions having values comprising a composite cost of CPU utility and input/output overhead.

6. In a system for accessing data in a plurality of relational computer databases on distributed network of database machines, a method of structuring access strategies based on derived cost models for at least two participating database management systems (DBMS), each DBMS having a structured query language (SQL), but differing associated cost models, the method comprising:

constructing a model database wherein all rows, columns and relational structures are known and controlled;

conducting a series of access tests on each participating DBMS using said model database;

deriving access cost data for each participating DBMS according to said access tests;

storing the access cost data as a logical cost model in a datadictionary/catalog;

determining an optimum application plan for subsequent distributed database queries relying on the logical cost model stored in said data-dictionary/catalog; and

executing subsequent queries to return data from the distributed databases in accordance with said optimum application plan.

7. A method according to claim 6 wherein the model database construction step further comprises:

creating a database wherein for an integer "n," R_n is a relation of seven columns containing 2ⁿ tuples,

each of said seven columns (C_n, where n={1 . .7}) having the following attributes;

C₁ ;

integer {0, n}, indexed and clustered;

C₂ ;

integer {0, 2ⁿ -1}, indexed, defacto clustered, but not specified to the DBMS as such;

C₃ ;

integer {0, n}, indexed and unclustered;

C₄ ;

integer {0, n}, having no index;

C₅ ;

integer {0, 2ⁿ -1}, indexed and unclustered;

C₆ ;

integer {0,2ⁿ -1}, having no index; and

C₇ ;

a long character string to meet the size of the tuple requirement;

setting the value for the seventh attribute to be a padding field; and

setting the multicolumn key for the relation to be {C₁, C₂ } such that the relation is indexed on this key, in ascending order, with C₁ being the major key and C₂ being the minor key.

8. A method according to claim 6 wherein the step of conducting a series of access tests further comprises:

applying a minimum of twelve individual queries to each participating DBMS; and

posing each query a minimum of thirty times to derive a statistically significant sampling.

9. A method according to claim 6 wherein the step of deriving access cost data further comprises employing a Least Square Fitting algorithm to reduce errors and estimate cost coefficients.

10. A method according to claim 6 including:

structuring the cost model in accordance with a logical execution of the database queries; and

estimating the cost of a given query based on logical characteristics of each participating DBMS, logical relations between data stored in the databases, and a logical structure of the query.

11. A method according to claim 10 including estimating the cost of a complex query using a set of primitive queries.

12. A computer system for querying a plurality of databases on a network of database computers comprising:

a data storage means in each of the computers for holding database datadictionaries/catalogs and database access mechanisms;

input and display means connected to each of the computers for inputting database queries to the system and displaying results in human readable format;

communications means for transmitting and receiving queries and results by and between the plurality of database computers;

each computer including a database management system (DBMS) having a DBMS query mechanism for accessing a database in response to a database query;

a cost model including a data-dictionary/catalog containing cost data based on coefficients of composite costs including at least CPU and I/O operations relative to a database query, derived by running each individual database computer'"'"'s DBMS query mechanism against a model database for each of the tested DBMSs in the network; and

means for building query access strategies for each database computer'"'"'s DBMS based on the cost data stored in said data-dictionary/catalog.

13. A system according to claim 12 in which the cost model includes means defining data access Cost information about communication time between the database computers via the communication means.

14. A system according to claim 12 in which the cost model is structured in accordance with a logical execution of the database queries, including means for estimating the cost of a given query based on logical characteristics of a DBMS, logical relations between data stored in a database, and a logical structure of a query.

15. A system according to claim 12 including:

means for constructing a model database to derive the cost data;

means for running each individual database computer'"'"'s DBMS query mechanisms against the model database; and

means for deriving cost data for each of the tested DBMSs in the network and loading the resulting cost data into the data-dictionary/catalog.

16. A system according to claim 12 in which the means for building query access strategies is operative to construct a query access strategy including a tuple of variable size, and the cost model includes means for estimating a cost of the strategy as a linear function of the size of the tuple.

17. A system according to claim 12 in which the means for building query access strategies is operative to construct a query access strategy including a number of selection and projection clauses, and the cost model includes means for estimating a cost of the strategy as linear functions of the selection and projection clauses.

18. A method of using a programmable system to perform electronic data management among a plurality of electronic relational databases and corresponding DBMSs in a network distributed environment, said programmable system having a plurality of machine components each including a data storage device, a display device, and a communications means for interconnecting the machine components for bidirectional data communications therebetween, the method comprising:

constructing a model database in which all relational structures and components are known and controlled;

running a series of controlled access tests against the model database by each of the plurality of DBMSs in the network;

tracking and recording resulting cost data for each access test;

storing said cost data as a logical cost model in a network datadictionary/catalog file in the data storage device;

determining an optimum application plan for subsequent distributed database queries relying on the logical cost model stored in said data-dictionary/catalog; and

executing subsequent queries to return data from the distributed databases in accordance with said optimum application plan.

19. A method according to claim 18 wherein the constructing a model database step further comprises:

setting up a relation of seven columns containing 2ⁿ tuples for any integer n such that

Column 1 contains integers {0, n}, indexed and clustered;

Column 2 contains integers {0, 2ⁿ -1}, indexed and de facto clustered, but not specified to the DBMS as such;

Column 3 contains integers {0, n}, indexed and unclustered;

Column 4 contains integers {0, n}, having no index;

Column 5 contains integers {0, 2ⁿ -1}, indexed and unclustered;

Column 6 contains integers {0,2ⁿ -1}, having no index; and

Column 7 contains a long character string to meet the size of the tuple requirement.

20. A method according to claim 18 wherein the tracking and recording step further comprises employing a Least Square Fitting algorithm to reduce errors and estimate cost coefficients.