System and methodology for cost-based subquery optimization using a left-deep tree join enumeration algorithm

Abstract

A system providing methodology for cost-based enumeration of subqueries using a left-deep tree join enumeration algorithm is described. In one embodiment, for example, in a database system, a method of the present invention is described for optimizing a database query, the method comprises steps of: receiving a database query including at least one subquery; building a query optimization graph for each query block of the database query, the query optimization graph including plan nodes representing subqueries of each query block; generating a set of access methods and join methods for each plan node, including generating at least one access method for a subquery quantifier based on subquery type and semantic properties of the database query; determining an optimal access plan for each query block based upon selecting access methods, join methods, and join order for plan nodes of the query optimization graph having favorable execution costs; and constructing a detailed access plan for execution of the database query based upon the optimal access plan determined for each query block.

Description

COMPUTER PROGRAM LISTING APPENDIX

[0002] This application includes a transmittal under 37 C.F.R. § 1.52(e) of a Computer Program Listing Appendix comprising duplicate compact discs (2), respectively labeled “Copy 1” and “Copy 2”. The discs are IBM-PC machine formatted and Microsoft® Windows Operating System compatible, and include identical copies of the following list of files:

	File Name	Created/Last Modified	File Size (KB)
	SourceCode.txt	Apr. 26, 2004, 5:55 pm	129 KB

[0003] All of the material disclosed in the Computer Program Listing Appendix is hereby incorporated by reference into the present application.

COPYRIGHT NOTICE

[0004] A portion of the disclosure of this patent document contains material which is subject to copyright protection. The copyright owner has no objection to the facsimile reproduction by anyone of the patent document or the patent disclosure as it appears in the Patent and Trademark Office patent file or records, but otherwise reserves all copyright rights whatsoever.

BACKGROUND OF THE INVENTION

[0005] 1. Field of the Invention

[0006] The present invention relates generally to information processing environments and, more particularly, to a database management system (DBMS) having a methodology for cost-based subquery optimization using a left-deep tree join enumeration algorithm.

[0007] 2. Description of the Background Art

[0008] Computers are very powerful tools for storing and providing access to vast amounts of information. Computer databases are a common mechanism for storing information on computer systems while providing easy access to users. A typical database is an organized collection of related information stored as “records” having “fields” of information. As an example, a database of employees may have a record for each employee where each record contains fields designating specifics about the employee, such as name, home address, salary, and the like.

[0009] Between the actual physical database itself (i.e., the data actually stored on a storage device) and the users of the system, a database management system or DBMS is typically provided as a software cushion or layer. In essence, the DBMS shields the database user from knowing or even caring about underlying hardware-level details. Typically, all requests from users for access to the data are processed by the DBMS. For example, information may be added or removed from data files, information may be retrieved from, or updated in, such files, and so forth, all without user knowledge of underlying system implementation. In this manner, the DBMS provides users with a conceptual view of the database that is removed from the hardware level. The general construction and operation of a database management system is known in the art. See e.g., Date, C., “An Introduction to Database Systems, Volume I and II”, Addison Wesley, 1990; the disclosure of which is hereby incorporated by reference.

[0010] DBMS systems have long since moved from a centralized mainframe environment to a decentralized or distributed environment. One or more PC “client” systems, for instance, may be connected via a network to one or more server-based database systems (SQL database server). Commercial examples of these “client/server” systems include Powersoft® clients connected to one or more Sybase® SQL Anywhere® Studio (Adaptive Server® Anywhere) database servers. Both Powersoft and Sybase SQL Anywhere Studio (Adaptive Server Anywhere) are available from Sybase, Inc. of Dublin, Calif.

[0011] In today's computing environment, database technology can be found on virtually any device, from traditional mainframe computers to cellular phones. Sophisticated applications, whether human resources information systems or sales force automation systems, can “push” much of their complexity into the database itself. Indeed, this represents one of the main benefits of database technology. The challenge, however, is to support these applications, and the complex queries they generate, on small computing devices. At the same time, users expect the productivity and reliability advantages of using a relational DBMS.

[0012] One purpose of a database system is to answer decision support queries. A query may be defined as a logical expression over the data and the data relationships set forth in the database, and results in the identification of a subset of the database. Consider, for instance, the execution of a request for information from a relational DBMS. In operation, this request is typically issued by a client system as one or more Structured Query Language or “SQL” queries for retrieving particular data (e.g., a list of all employees earning more than the average salary of all employees) from database tables on a server. In response to this request, the database system typically returns the names of those employees earning more than the average salary, where “employees” is a table defined to include information about employees of a particular organization. The syntax of SQL is well documented, see e.g., “Information Technology—Database languages—SQL”, published by the American National Standards Institute as American National Standard ANSI/ISO/IEC 9075: 1992, the disclosure of which is hereby incorporated by reference.

[0013] SQL queries express what results are requested but do not state how the results should be obtained. In other words, the query itself does not tell how the query should be evaluated by the DBMS (i.e., the query is declarative). Rather, a component called the optimizer determines the “plan” or the best method of accessing the data to return the result required by the SQL query. The query optimizer is responsible for transforming an SQL request into an access plan composed of specific implementations of the algebraic operator selection, projection, join, and so forth. The role of a query optimizer in a relational DBMS system is to find an adequate execution plan from a search space of many semantically equivalent alternatives.

[0014] A query optimizer transforms an SQL query into an access plan by generating different join strategies and, based on cost, choosing the best strategy. The process of generating a subspace of the space of all the join strategies is called join enumeration. Since relational databases typically only provide physical operators that can join two tables at a time, a join of a number of different tables (n-way join) must be executed as a sequence of two-way joins, and there are many possible such sequences. The optimizer must typically enumerate some or all of these sequences and choose one based on estimates of their relative execution costs.

[0015] In order to optimize a query, a query optimizer must be able to concurrently solve several different problems including: choosing the access method (e.g., sequential scan, index scan) for each base table used in the query; choosing the order in which to join the tables (i.e., the join order), and choosing the join method to be used for each join operation. A complete access plan comprises a join order for joining the relations (tables), join methods for each join operation, and an access method for each base table used in the query.

[0016] The SQL language allows the usage of queries using nested subqueries. In general, a subquery is embedded in the WHERE clause of a query to further restrict data returned by the query. A database optimizer, in general, optimizes a subquery block separately from the rest of the query block the subquery is used in. The evaluation of the subquery is then done when it is needed in the context in which it is used. For example, if the subquery appears in a filter predicate on a table, the subquery is evaluated for each row of the table while evaluating the predicate.

[0017] Subquery evaluation techniques and cost-based subquery optimization have been proposed in the relational database literature as a means to evaluate queries containing nested subqueries. Some of the most important work in this domain, the disclosure of which are hereby incorporated by reference, are:

[0018] Dayal, U., “Of Nests and Trees: A Unified Approach to Processing Queries that Contains Nested Subqueries, Aggregates, and Quantifiers”, in Proceedings of the 13th International Conference on Very Large Data Bases, Brighton, England, pp. 197-208, 1987.

[0019] Galindo-Legaria, C. A. et al, “Ortogonal Optimization of Subqueries and Aggregation”, Proceedings of ACM SIGMOD Conference of Management of Data, Santa Barbara, Calif., pp. 571-581, May 2001.

[0020] Muralikrishna, M., “Improved Unnesting Algorithms for Join Aggregate SQL Queries”, Proceedings of the 18th International Conference on Very Large Data Bases, Vancouver, Canada, pp. 91-102, August 1992.

[0021] Kim, W., “On Optimizing an SQL-like Nested Query”, in ACM Transactions on Database Systems, 7(3): pp. 443-469, September 1982.

[0022] A major challenge in database systems is to significantly reduce the amount of memory required to enable a database system to be run on small computing devices. More particularly, what is required is a solution which enables queries to be effectively optimized while requiring a minimal amount of memory for generating the search space and for storing the information required for query optimization. One of the challenges in cost-based subquery optimization in database systems is to integrate the subquery optimization into the optimizer's enumeration process without significantly increasing the amount of memory required, or significantly increasing the search space the optimizer is considering.

[0023] What is needed is a query optimization solution which integrates subquery optimization with an optimizer's general enumeration methodology. The solution should not significantly increase the search space considered during query optimization as an increase in the search space typically would require a corresponding an increase in the amount of memory required by the optimizer. The present invention provides a solution for these and other needs.

GLOSSARY

[0024] The following definitions are offered for purposes of illustration, not limitation, in order to assist with understanding the discussion that follows.

[0025] Bushy tree: A processing tree is called a “bushy tree” if it has composite (i.e., not necessarily base relations) for the left and right children of the join nodes. For further description of a bushy tree search space, see e.g., Vance, B. et al “Rapid Bushy Join-Order Optimization with Cartesian Products”, in Proceedings of ACM SIGMOD Conference of Management of Data, pp. 35-46, June 1996.

[0026] Correlated subquery: A subquery which refers to a column of a table from the outside query block. Such a column is referred to as an outerreference or correlation. For example, the following query uses a scalar correlated subquery which refers to the column “O.X” of a table “0” from the FROM clause of the outer query block: “SELECT*FROM O WHERE 100>(SELECT T.Y FROM T WHERE T.X=O.X)”.

[0027] Hash Filter: A hash filter built on a relation R, on its attribute X—denoted by HFR(X)—is an array of bits built using a hash function h( ). The n-th bit of the hash filter HFR(X) is set to 1 if, and only if, there exists a value x of the attribute R.X of the relation R such that h(x)=n. Otherwise the bit is set to 0. Hash filters can be used to reduce the number of rows participating in a join. If a join predicate R.X=T.Y is used to join the tables R and T, then a row of T can be probed, before the join, against the already-built hash filter HFR(X), by testing if the h(y) bit is set to 1 in the hash filter HFR(X), where y is the value of the attribute T.Y of that row. Such a probing will be denoted as a predicate “T.Y IN HFR(X)”.

[0028] Query block: A query block refers to an atomic portion or block of a query that has more than one block because the query contains derived tables, views, and/or subqueries. A query block can be the main block of a SQL SELECT, UPDATE, INSERT, or DELETE statement. A query block can be the main block of a SELECT-FROM-WHERE block (with or without GROUP BY and HAVING clauses) defining a view or a derived table. A query block can also be the main block of a subquery. In “Access Path Selection in a Relational Database Management System”, Selinger, P. G. et al, Proceedings of ACM SIGMOD Conference of Management of Data, pp 23-34, May 1979, a query block is defined as being “represented by a SELECT list, a FROM list, and a WHERE tree containing, respectively the list of items to be retrieved, the table(s) referenced, and the boolean combination of simple predicates specified by the user”. A single SQL statement may have many query blocks because a predicate may have one operand which is itself a query.

[0029] Left-deep tree: A left-deep tree is a processing tree that has only base tables or derived tables as the right child for any join. Left-deep trees are desirable because they reduce the need to materialize intermediate results and for several types of join implementations they result in more efficient execution plans. For further description of left-deep processing trees, see e.g., Cluet, S. et al “On the Complexity of Generating Optimal Left-Deep Processing Trees with Cross Products”, in Proceedings of the Fifth International Conference on Database Theory, pp. 54-67, January 1995.

[0030] Non-scalar subquery: A subquery that may return zero or more rows (with one column). Such subqueries are used in existential predicates such as “NOT EXISTS(SELECT 1 from R WHERE . . . )”, or quantified comparison such as “T.X>ALL(SELECT R.X FROM R WHERE . . . )”.

[0031] Scalar subquery: A scalar subquery is a subquery which can return at most one row (with one column). A scalar subquery can be used in any place where a relational expression can be used in a SQL query, for example, in a comparison predicate of the form “T.X>(SELECT AVG(R.X) FROM WHERE . . . )”.

[0032] SQL: SQL stands for Structured Query Language. The original version called SEQUEL (structured English query language) was designed by IBM in the 1970's. SQL-92 (or SQL/92) is the formal standard for SQL as set out in a document published by the American National Standards Institute in 1992; see e.g., “Information Technology—Database languages—SQL”, published by the American National Standards Institute as American National Standard ANSI/ISO/IEC 9075: 1992, the disclosure of which is hereby incorporated by reference. SQL-92 was superseded by SQL-99 (or SQL3) in 1999; see e.g., “Information Technology—Database Languages—SQL, Parts 1-5” published by the American National Standards Institute as American National Standard INCITS/ISO/IEC 9075-(1-5)-1999 (formerly ANSI/ISO/IEC 9075-(1-5)-1999), the disclosure of which is hereby incorporated by reference.

[0033] Subquery: A subquery is a basic SELECT block whose result is a table with exactly one column.

SUMMARY OF THE INVENTION

[0034] A system providing methodology for cost-based enumeration of subqueries using a left-deep tree join enumeration algorithm is described. In one embodiment, for example, in a database system, a method of the present invention is described for optimizing a database query, the method comprises steps of: receiving a database query including at least one subquery; building a query optimization graph for each query block of the database query, the query optimization graph including plan nodes representing subqueries of each query block; generating a set of access methods and join methods for each plan node; determining an optimal access plan for each query block based upon selecting access methods, join methods, and join order for plan nodes of the query optimization graph having favorable execution costs; and constructing a detailed access plan for execution of the database query based upon the optimal access plan determined for each query block.

[0035] The methodology of the present invention includes defining, for each qualifying subquery, a set of access methods (or execution methods), and a set of join methods to be used to join such subqueries with the rest of the query's tables. The access methods are defined based on the subquery execution strategies which could be employed for a subquery given its type and the semantic properties of the query the subquery is used in. One execution strategy utilized in the present invention is the execution of the correlated subqueries as uncorrelated derived tables with hash filters. The enumeration process is further modified to treat subqueries in a similar way with the base or derived tables, by enumerating and costing subqueries using these special subquery access and join methods. Each subquery has assigned a property vector (containing CPU cost, I/O cost, order properties, etc.) similar to the property vector used for any other base table, which is recomputed for each partial access plan ending with that subquery. Each access method of a subquery participating in the enumeration process has a property vector associated to it computed by optimizing the subquery's Derived Table Block (DTB) specifically for that access method. Each special join method associated to an access method of a subquery implements the computation of its property vector during enumeration process (e.g., each subquery join method computes its own cost based on the current subquery access method and the prefix of the current access plan under consideration). During the enumeration process, the access methods and join methods associated to a subquery are treated in the same fashion with any other access methods (e.g., a partial index scan) or join methods associated to base or derived tables. Compared to other cost-based subquery optimization techniques, the advantages provided by the present invention include the following:

[0036] (1) The methodology of the present invention provides that the subquery participating in the enumeration process is not transformed during optimization to reflect different subquery execution strategies. Instead, after the optimization is completed, the transformation required by the best access plan is applied in order to build the Data Flow Operator (DFO) tree corresponding to the chosen access method/join method for a subquery. This approach uses far less memory than other transformation-based subquery optimization techniques where a SQL query is rewritten to reflect a certain subquery evaluation strategy, and, for each rewritten query, the optimization process is applied.

[0037] (2) The optimizer does not have to define new constructs to make it possible to enumerate subqueries in the same manner as any other base or derived tables. The modification of the enumeration process is minimal which makes it possible to easily consider new subquery execution strategies.

[0038] (3) The execution engine does not have to define new DFO operators in order to apply different subquery execution strategies. The join methods and access methods associated to a subquery correspond to DFO operators already used for base and derived tables.

[0039] (4) An execution strategy is defined for qualifying subqueries where a correlated subquery can be executed as an uncorrelated derived table with hash filters. For this type of strategy, as for any other subquery execution strategy, the qualifying subquery has defined a corresponding access method whose property vector reflects the estimated cost of the uncorrelated derived table with hash filters.

BRIEF DESCRIPTION OF THE DRAWINGS

[0040] FIG. 1 is a block diagram of a computer system in which software-implemented processes of the present invention may be embodied.

[0041] FIG. 2 is a block diagram of a software system for controlling the operation of the computer system.

[0042] FIG. 3 illustrates the general structure of a client/server database system suitable for implementing the present invention.

[0043] FIG. 4 illustrates the overall process of query optimization in the database system.

[0044] FIGS. 5A-D illustrate Query Optimization Graph (QOG) representations of several sample table expressions.

[0045] FIG. 6 is a graphical representation of the search space for a subplan which is traversed using a depth-first search procedure.

[0046] FIG. 7 is a graphical representation of a portion of the search space for a sample subplan which illustrates the optimization methodology of the present invention.

[0047] FIGS. 8A-C comprise a single flowchart illustrating at a high level the methods of operation of the present invention in generating bushy trees using a left-deep tree join enumeration algorithm.

[0048] FIGS. 9A-B comprise a graphical representation of a portion of the search space illustrating the optimization process for a query using the cost-based subquery optimization methodology of the present invention.

[0049] FIGS. 10A-B comprise a single flowchart illustrating at a high level the enumeration process including the cost-based subquery optimization methodology of the present invention.

DETAILED DESCRIPTION OF A PREFERRED EMBODIMENT

[0050] The following description will focus on the presently preferred embodiment of the present invention, which is implemented in a desktop and/or server software (e.g., application, driver, or the like) operating in an Internet-connected environment running under a desktop or server operating system, such as the Microsoft® Windows operating system running on an IBM-compatible PC. The present invention, however, is not limited to any one particular application or any particular environment. Instead, those skilled in the art will find that the system and methods of the present invention may be advantageously embodied on a variety of different platforms, including Macintosh, Linux, BeOS, Solaris, UNIX, NextStep, FreeBSD, and the like. Therefore, the description of the exemplary embodiments that follows is for purposes of illustration and not limitation. The exemplary embodiments are primarily described with reference to block diagrams or flowcharts. As to the flowcharts, each block within the flowcharts represents both a method step and an apparatus element for performing the method step. Depending upon the implementation, the corresponding apparatus element may be configured in hardware, software, firmware, or combinations thereof.

[0051] I. Computer-Based Implementation

[0052] A. Basic System Hardware (e.g., for Desktop and Server Computers)

[0053] The present invention may be implemented on a conventional or general-purpose computer system, such as an IBM-compatible personal computer (PC) or server computer. FIG. 1 is a very general block diagram of an IBM-compatible system 100. As shown, system 100 comprises a central processing unit(s) (CPU) or processor(s) 101 coupled to a random-access memory (RAM) 102, a read-only memory (ROM) 103, a keyboard 106, a printer 107, a pointing device 108, a display or video adapter 104 connected to a display device 105, a removable (mass) storage device 115 (e.g., floppy disk, CD-ROM, CD-R, CD-RW, DVD, or the like), a fixed (mass) storage device 116 (e.g., hard disk), a communication (COMM) port(s) or interface(s) 110, a modem 112, and a network interface card (NIC) or controller 111 (e.g., Ethernet). Although not shown separately, a real-time system clock is included with the system 100, in a conventional manner.

[0054] CPU 101 comprises a processor of the Intel Pentium® family of microprocessors. However, any other suitable processor may be utilized for implementing the present invention. The CPU 101 communicates with other components of the system via a bi-directional system bus (including any necessary input/output (I/O) controller circuitry and other “glue” logic). The bus, which includes address lines for addressing system memory, provides data transfer between and among the various components. Description of Pentium-class microprocessors and their instruction set, bus architecture, and control lines is available from Intel Corporation of Santa Clara, Calif. Random-access memory 102 serves as the working memory for the CPU 101. In a typical configuration, RAM of sixty-four megabytes or more is employed. More or less memory may be used without departing from the scope of the present invention. The read-only memory (ROM) 103 contains the basic input/output system code (BIOS)—a set of low-level routines in the ROM that application programs and the operating systems can use to interact with the hardware, including reading characters from the keyboard, outputting characters to printers, and so forth.

[0055] Mass storage devices 115, 116 provide persistent storage on fixed and removable media, such as magnetic, optical or magnetic-optical storage systems, flash memory, or any other available mass storage technology. The mass storage may be shared on a network, or it may be a dedicated mass storage. As shown in FIG. 1, fixed storage 116 stores a body of program and data for directing operation of the computer system, including an operating system, user application programs, driver and other support files, as well as other data files of all sorts. Typically, the fixed storage 116 serves as the main hard disk for the system.

[0056] In basic operation, program logic (including that which implements methodology of the present invention described below) is loaded from the removable storage 115 or fixed storage 116 into the main (RAM) memory 102, for execution by the CPU 101. During operation of the program logic, the system 100 accepts user input from a keyboard 106 and pointing device 108, as well as speech-based input from a voice recognition system (not shown). The keyboard 106 permits selection of application programs, entry of keyboard-based input or data, and selection and manipulation of individual data objects displayed on the screen or display device 105. Likewise, the pointing device 108, such as a mouse, track ball, pen device, or the like, permits selection and manipulation of objects on the display device. In this manner, these input devices support manual user input for any process running on the system.

[0057] The computer system 100 displays text and/or graphic images and other data on the display device 105. The video adapter 104, which is interposed between the display device 105 and the system's bus, drives the display device 105. The video adapter 104, which includes video memory accessible to the CPU 101, provides circuitry that converts pixel data stored in the video memory to a raster signal suitable for use by a cathode ray tube (CRT) raster or liquid crystal display (LCD) monitor. A hard copy of the displayed information, or other information within the system 100, may be obtained from the printer 107, or other output device. Printer 107 may include, for instance, an HP LaserJet® printer (available from Hewlett-Packard of Palo Alto, Calif.), for creating hard copy images of output of the system.

[0058] The system itself communicates with other devices (e.g., other computers) via the network interface card (NIC) 111 connected to a network (e.g., Ethernet network, Bluetooth wireless network, or the like), and/or modem 112 (e.g., 56K baud, ISDN, DSL, or cable modem), examples of which are available from 3Com of Santa Clara, Calif. The system 100 may also communicate with local occasionally-connected devices (e.g., serial cable-linked devices) via the communication (COMM) interface 110, which may include a RS-232 serial port, a Universal Serial Bus (USB) interface, or the like. Devices that will be commonly connected locally to the interface 110 include laptop computers, handheld organizers, digital cameras, and the like.

[0059] IBM-compatible personal computers and server computers are available from a variety of vendors. Representative vendors include Dell Computers of Round Rock, Tex., Hewlett-Packard of Palo Alto, Calif., and IBM of Armonk, N.Y. Other suitable computers include Apple-compatible computers (e.g., Macintosh), which are available from Apple Computer of Cupertino, Calif., and Sun Solaris workstations, which are available from Sun Microsystems of Mountain View, Calif.

[0060] B. Basic System Software

[0061] Illustrated in FIG. 2, a computer software system 200 is provided for directing the operation of the computer system 100. Software system 200, which is stored in system memory (RAM) 102 and on fixed storage (e.g., hard disk) 116, includes a kernel or operating system (OS) 210. The OS 210 manages low-level aspects of computer operation, including managing execution of processes, memory allocation, file input and output (I/O), and device I/O. One or more application programs, such as client application software or “programs” 201 (e.g., 201a, 201b, 201c, 201d) may be “loaded” (i.e., transferred from fixed storage 116 into memory 102) for execution by the system 100. The applications or other software intended for use on the computer system 100 may also be stored as a set of downloadable computer-executable instructions, for example, for downloading and installation from an Internet location (e.g., Web server).

[0062] Software system 200 includes a graphical user interface (GUI) 215, for receiving user commands and data in a graphical (e.g., “point-and-click”) fashion. These inputs, in turn, may be acted upon by the system 100 in accordance with instructions from operating system 210, and/or client application module(s) 201. The GUI 215 also serves to display the results of operation from the OS 210 and application(s) 201, whereupon the user may supply additional inputs or terminate the session. Typically, the OS 210 operates in conjunction with device drivers 220 (e.g., “Winsock” driver—Windows' implementation of a TCP/IP stack) and the system BIOS microcode 230 (i.e., ROM-based microcode), particularly when interfacing with peripheral devices. OS 210 can be provided by a conventional operating system, such as Microsoft® Windows 9x, Microsoft® Windows NT, Microsoft® Windows 2000, or Microsoft® Windows XP, all available from Microsoft Corporation of Redmond, Wash. Alternatively, OS 210 can also be an alternative operating system, such as the previously-mentioned operating systems.

[0063] The above-described computer hardware and software are presented for purposes of illustrating the basic underlying desktop and server computer components that may be employed for implementing the present invention. For purposes of discussion, the following description will present examples in which it will be assumed that there exists a “server” (e.g., Web server) that communicates with one or more “clients” (e.g., desktop computers). The present invention, however, is not limited to any particular environment or device configuration. In particular, a client/server distinction is not necessary to the invention, but is used to provide a framework for discussion. Instead, the present invention may be implemented in any type of system architecture or processing environment capable of supporting the methodologies of the present invention presented in detail below.

[0064] C. Client/Server Database Management System

[0065] While the present invention may operate within a single (standalone) computer (e.g., system 100 of FIG. 1), the present invention is preferably embodied in a multi-user computer system, such as a client/server system. FIG. 3 illustrates the general structure of a client/server database system 300 suitable for implementing the present invention. As shown, the system 300 comprises one or more client(s) 310 connected to a server 330 via a network 320. Specifically, the client(s) 310 comprise one or more standalone terminals 311 connected to a database server system 340 using a conventional network. In an exemplary embodiment, the terminals 311 may themselves comprise a plurality of standalone workstations, dumb terminals, or the like, or comprise personal computers (PCs) such as the above-described system 100. Typically, such units would operate under a client operating system, such as a Microsoft Windows client operating system (e.g., Microsoft Windows 95/98, Windows 2000, or Windows XP).

[0066] The database server system 340, which comprises Sybase® Adaptive Server® Anywhere (available from Sybase, Inc. of Dublin, Calif.) in an exemplary embodiment, generally operates as an independent process (i.e., independently of the clients), running under a server operating system such as Microsoft Windows NT, Windows 2000, or Windows XP (all from Microsoft Corporation of Redmond, Wash.), UNIX (Novell), or Linux (Red Hat). The network 320 may be any one of a number of conventional network systems, including a Local Area Network (LAN) or Wide Area Network (WAN), as is known in the art (e.g., using Ethernet, IBM Token Ring, or the like). The network 320 includes functionality for packaging client calls in the well-known SQL (Structured Query Language) together with any parameter information into a format (of one or more packets) suitable for transmission across a cable or wire, for delivery to the database server system 340.

[0067] Client/server environments, database servers, and networks are well documented in the technical, trade, and patent literature. For a discussion of Sybase-branded database servers and client/server environments generally, see e.g., Nath, A., “The Guide to SQL Server”, Second Edition, Addison-Wesley Publishing Company, 1995. Additional documentation describing the operations of Sybase SQL Anywhere Studio version 9.0 (Adaptive Server Anywhere) is available from Sybase, Inc. as “SQL Anywhere Studio 9.0 Collection Documentation” (e.g., at sybooks.sybase.com/awg0900e.html).

[0068] In operation, the client(s) 310 store data in, or retrieve data from, one or more database tables 350, as shown at FIG. 3. Typically resident on the server 330, each table itself comprises one or more rows or “records” (tuples) (e.g., row 355), each storing information arranged by columns or “fields.” A database record includes information which is most conveniently represented as a single unit. A record for an employee, for example, may include information about the employee's ID Number, Last Name and First Initial, Position, Date Hired, Social Security Number, and Salary. Thus, a typical record includes several categories of information about an individual person, place, or thing. Each of these categories, in turn, represents a database field. In the foregoing employee table, for example, Position is one field, Date Hired is another, and so on. With this format, tables are easy for users to understand and use. Moreover, the flexibility of tables permits a user to define relationships between various items of data, as needed.

[0069] In operation, the client(s) 310 issue one or more SQL commands to the server 330. SQL commands may specify, for instance, a query for retrieving particular data (i.e., data records meeting the query condition) from the database table(s) 350. The syntax of SQL (Structured Query Language) is well documented; see e.g., the above-mentioned “An Introduction to Database Systems”. In addition to retrieving the data from Database Server tables, the client(s) also have the ability to insert new rows of data records into the table(s); client(s) can also modify and/or delete existing records in the table(s).

[0070] In operation, the SQL statements received from the client(s) 310 (via network 320) are processed by engine 360 of the database server system 340. The engine 360 itself comprises parser 361, normalizer 363, compiler 365, execution unit 369, access methods 370, and DB store (page and index manager) 375. Specifically, the SQL statements are passed to the parser 361 which converts the statements into a query tree—a binary tree data structure which represents the components of the query in a format selected for the convenience of the system. In this regard, the parser 361 employs conventional parsing methodology (e.g., recursive descent parsing).

[0071] The query tree is normalized by the normalizer 363. Normalization includes, for example, the elimination of redundant data. Additionally, the normalizer 363 performs error checking, such as confirming that table names and column names which appear in the query are valid (e.g., are available and belong together). Finally, the normalizer can also look up any referential integrity constraints which exist and add those to the query.

[0072] After normalization, the query tree is passed to the compiler 365, which includes an optimizer 366 and a code generator 367. The optimizer is responsible for optimizing the query tree. The optimizer performs a cost-based analysis for formulating a query execution plan. The optimizer will, for instance, select the join order of tables (e.g., when working with more than one table); it will select relevant indexes (e.g., when indexes are available). The optimizer, therefore, performs an analysis of the query and determines the best execution plan, which in turn results in particular access methods being invoked during query execution. The operations of the optimizer in generating a query execution plan are described in greater detail below.

[0073] For enhancing the storage, retrieval, and processing of data records, the server 330 maintains one or more database indexes 345 on database tables 350. These database indexes 345 facilitate quick access to the data records. A database index, typically maintained as a B-Tree data structure, allows the records of a table to be organized in many different ways, depending on a particular user's needs. An index may be constructed as a single disk file storing index key values together with unique record numbers. An index key value is a data quantity composed of one or more fields from a record which are used to arrange (logically) the database file records by some desired order (index expression). The record numbers are unique pointers to the actual storage location of each record in the database file. Both are referred to internally by the system for locating and displaying records in a database file.

[0074] II. Generating Bushy Trees Using Left-Deep Join Enumeration Methodology

[0075] A. General Process of Query Optimization

[0076] Before describing the invention in more detail, it is helpful to describe the overall process of query optimization in the database management system of the currently preferred embodiment. The database management system used in the currently preferred embodiment is version 9.0 of Sybase® Adaptive Server® Anywhere (ASA). Adaptive Server Anywhere is the relational database system that forms the core module of Sybase® SQL Anywhere® Studio, a suite of technologies intended for use in workgroup, mobile, and embedded applications.

[0077] FIG. 4 illustrates the overall process of query optimization in the database system of the currently preferred embodiment. As shown, the input to the plan generation phase of the query optimizer is a Query Optimization Graph (QOG), pronounced “cog”. A QOG is the internal representation of a query block or Derived Table Block (DTB) as hereinafter described. Each query block is optimized independently in a bottom-up fashion.

[0078] Unlike IBM's Starburst optimizer, which separates join enumeration from plan generation and cost estimation (see, e.g., Ono, K. et al, “Measuring the Complexity of Join Enumeration in Query Optimization”, in Proceedings of the 16th International Conference on Very Large Data Bases, pp. 314-325, August 1990), the optimizer of the currently preferred embodiment combines these processes into a single unified methodology as described in more detail below. As shown at FIG. 4, cost estimation is an integral part of the enumeration method, because it is through comparing the costs of partial access plans that the optimizer can quickly prune significant portions of the search space. The following description will focus on those aspects of the database management system that are helpful for understanding the methodology of the present invention.

[0079] B. Overview

[0080] The present invention provides a methodology for generating bushy trees using a left-deep tree join enumeration algorithm in a relational database system. The methodology efficiently optimizes high join degree queries using a novel approach for generating join trees during query optimization. The optimization methodology of the present invention starts by finding the most favorable access plans for each query block in a bottom up fashion. The generation of join trees is done in a step-by-step manner, combining selection of access methods and join methods for each base relation with join order selection and costing. The methodology of the present invention uses a deterministic branch-and-bound join enumeration method, a cost-based technique for choosing the access methods and join methods, and cost-based pruning methods to efficiently prune the search space. The methodology of the present invention is currently implemented in Sybase® SQL Anywhere® version 9.0 and requires minimal memory for generating the search space and for storing the information required to restore the best access plans.

[0081] The optimization of a query block or a “Derived Table Block” (DTB as described below) consists of four phases. In the first phase, rewrite optimizations are applied and a semantically equivalent rewritten query is generated. For example, the nested outer joins are semantically analyzed and transformed into inner joins whenever possible. The second phase includes the building of the Query Optimization Graph (QOG) for each query block. The main objects defined in the QOG and manipulated during optimization are “subplans” and “quantifiers”, called plan nodes. For example, the system models the left, right and full outer joins as subplans which correspond to null-supplying sides of an outer join. For each base relation or derived table, a quantifier object is defined in a QOG. Some other key elements built into QOG are arrays of access methods defined for each plan node (for example, relevant indexes for quantifiers ranging over base tables), a join dependency graph, and an index dependency graph. For example, each base table quantifier has its array of access methods defined as a set of relevant indexes consisting of elements modeling table indexes that, based on the sargable predicates, may be used in an Access Plan (AP). In the third phase, a left-deep join enumeration algorithm is applied, in a bottom-up fashion, to each subplan. The algorithm prunes based on the property vector of a generated access plan. The property vector includes properties such as CPU cost, input/output (I/O) cost, order properties, and pipeline characteristics of the intermediate results. In the last phase, the best plan is rebuilt and a Data Flow Operator (DFO) tree is built corresponding to this best plan. Compared with other join enumeration algorithms generating bushy trees, the system and methodology of the present invention requires the use of far less memory.

[0082] The optimization methodology of the present invention starts by finding the most favorable access plans for each query block in a bottom up fashion. The generation of join trees for each subplan of a query block is performed in a step-by-step manner, combining selection of access methods and join methods for the plan nodes with join order selection and costing. Pruning is performed by comparing the best complete plan previously found with the current access plan or partial access plan (which may (or may not) contain all the plan nodes). An access plan (also referred to herein as a “prefix”) is built by adding plan nodes (or plan segments) one at a time. In other words, at each step a plan segment (i.e., a plan node that is not yet part of the access plan together with an associate candidate access method and join method) is added to the current prefix. The current prefix has all of the join methods and access methods already chosen, and it has a property vector (e.g., cost) already computed. During the optimization process a property vector of the best complete plan found so far is retained. If the current prefix or access plan under consideration is estimated to be less favorable than the cost of the best plan previously found, the current access plan is modified by changing the segment (i.e., plan node, access method, and/or join method) that was added in the last position. The plan node most recently added (i.e., joined) to the current subtree is replaced by a new plan node which has not yet been considered at this position if all access method and join method candidates for the plan node have been considered. Otherwise, the access method and the join method of the last plan node are replaced with other methods not yet evaluated that can be selected at the current position in the current access plan.

[0083] Unlike the other deterministic algorithms, the current invention combines the costing, the pruning, the enumeration algorithm, and choice of the join methods and access methods in a single optimization process. This optimization strategy makes it possible to store, at any point in the enumeration process, only the information about the best plan that has been found and the current access plan that is being considered. Hence, the memory required by the present invention is far less than that required, for example, by the Dynamic Programming Algorithm. The “Dynamic Programming Algorithm” is a deterministic technique and it is one of the optimization techniques most frequently implemented in database management systems. It was first used in System-R (see e.g., Selinger, P. G. et al, “Access Path Selection in a Relational Database Management System”, Proceedings of ACM SIGMOD Conference of Management of Data, pp. 23-34, May 1979). Moreover, the optimization methodology of the present invention can easily be extended with new join methods, access methods, or new types of subplans. For example, to add a new join method it is enough to define how the logical properties (e.g., size of the intermediate result), and physical properties (e.g., cost) of that method are to be computed during the enumeration process.

[0084] C. Definition of Terms

[0085] Before describing the operations of the present invention in greater detail, it is useful to define certain of the terms used in the following discussion in order to assist in explaining the optimization methodology of the present invention.

[0086] Definition 1: A “left-deep tree” is a processing tree that has only base tables or derived tables as the right child for any join. A derived table is a result (often materialized) of a view or table expression containing “UNION”, “GROUP BY”, “ORDER BY” or “TOP N”. Generally, there are n! (n factorial) ways to allocate n base relations to a left-deep tree's leaves. As described by Cluet, S. et al in “On the Complexity of Generating Optimal Left-Deep Processing Trees with Cross Products”, in Proceedings of the Fifth International Conference on Database Theory, pp. 54-67, January 1995, generating an optimal left-deep processing tree with Cartesian products is an NP-complete problem. An NP-complete problem is any one of a class of computational problems for which no efficient solution has been found. A problem is called NP (nondeterministic polynomial) if its solution can be guessed and verified in polynomial time.

[0087] Definition 2: A processing tree is called a “bushy tree” if it has composite (i.e., not necessarily base relations) for the left and right children of the join nodes. This space contains the space of the left-deep trees. For n base relations, the number of bushy trees is equal to: 1$\left(\begin{array}{c}2\left(n-1\right)\\ n-1\end{array}\right)\left(n-1\right).$

[0088] The access plan for a SELECT Derived Table Block (DTB) is generated by a join enumeration operation. The internal representation used by the enumeration operation is Query Optimization hypergraph (or “QOG”). A base table's DTB is defined for each base table used in a query. A proper derived table's DTB corresponds to a query block such as unflattened subqueries, views, or derived tables. For each proper DTB, a QOG is built that is used by the query optimizer during the query optimization process.

[0089] Definition 3: A “quantifier” is an atomic object defined in a QOG. A quantifier is defined for each table (i.e., derived table, base table, view, or subquery) used in the “FROM” clause, for each subquery used in the predicates or “SELECT” list. Hence, a quantifier corresponds to a Derived Table Block (DTB), and each DTB (other than the outermost DTB) has a quantifier ranging over that DTB. A quantifier inherits properties from its underlying DTB, such as the I/O (input/output) cost to generate all of the rows, the size of the relation that a DTB generates, and so forth.

[0090] Definition 4: A QOG is a representation of a single Derived Table Block (DTB). A QOG is defined as a hypergraph Q=<V,E> where V and E are defined as follows:

[0091] 1. V is the set of vertices that can be quantifiers or subplans that are referred to herein as plan nodes. A subplan node is a representation of a subset of plan nodes together with its own edges defined on that subset of plan nodes. Each QOG Q has an outer subplan (hereinafter referred to as Subplan (Q) and usually denoted by S0), which is not nested inside any other subplan. Each quantifier can belong to one and only one subplan. Each nested subplan can belong to one and only one subplan. Subplans are defined for the following table expressions:

[0092] (a) null-supplying side of an outer join, called a “null-supplying” subplan.

[0093] (b) a right hand side or left hand side of a full outer join, called a “full-outerjoin-side” subplan.

[0094] (c) a full outer join, called a “full-outerjoin” subplan.

[0095] (d) a QOG Q, called an “outer” subplan. This subplan is hereinafter denoted as S0 or Subplan (Q) unless otherwise specified.

[0096] The subplans are built in a bottom-up fashion such that nested subplans are built before the subplans that they belong to. Each subplan S is a tuple S=(PNS, pS, ES) where PNS is the set of subplan's plan nodes, pS is a predicate and ES is a set of edges defined for the plan nodes in PNS. For each subplan, Quantifiers(S) denotes all quantifiers used by the subplan S or by any nested subplan. If q is a quantifier then Quantifiers(q) is defined as the set {q}. The following “Quantifiers” pseudocode illustrates the method for finding the set of quantifiers used by a subplan:

1: Procedure: Quantifiers(s)
2: Purpose: Find the quantifiers used in a subplan.
3: Input:
4:  - A subplan S defined in a QOG Q.
5: Output:
6:  - A set of quantifiers.
7: Notations:
8:  - S = (PNS, ps,Es).
9: begin
10:  Qs  ∅
11:  for each plan node pn ∈ PNs do
12:   if pn is a quantifier then
13:    Qs  Qs ∪ {pn}
14:   else /* pn is a subplan */
15:    Qs  Qs ∪ Quantifiers(s)
16:   fi
17:  od
18:  return Qs
19: end

[0097] 2. E is a set of directed and undirected hyperedges. There are several types of edges defined in a QOG.

[0098] Ej: An undirected edge (or join edge) among plan nodes. An undirected edge is represented as (U, p) where U is a subset of V, U⊂V and p is a predicate referring to all quantifiers of U. In other words, the schema of p and V has in common the quantifiers defined by U:

sch (p)∩V=∪{u|uεU}Quantifiers(u).

[0099] Eo: Directed edges (or outerjoin edges) are defined for each null-supplying subplan. A direct edge is represented as (U, S) where U is a subset of V, U⊂V and S is a subplan in V. For each left outer join, a directed edge exists (U, S), where U is the set of plan nodes in the preserved side and S is the null-supplying subplan defined for this outer join. U is referred to as a “preserved hypernode” (see Definition 7 below). Note that the ON condition is part of the definition of the subplan S. The set of quantifiers in Quantifiers(S) is referred to as a null-supplying hypernode (see Definition 6 below).

[0100] Ef: Bidirected edges (or full-outerjoin edges) are defined for full-outerjoin-side subplans. A bidirected edge is represented as (S1, S2, pS1,S2) where S1, S2 are subplans in V corresponding to the left and right children of a full outer join and, pS1,S2 is the ON condition defined for this full outer join (see Definition 5 below).

[0101] Esj: The sargable join edge set Esj consists of directed edges esj=({qR}, Vexpr, p) corresponding to a WHERE or ON predicate p of the form p=R.Xθexpr(R1,R2, . . . ,Rn) where Vexpr={qR1, qR2, . . . , qRn}. Vexpr is the set of quantifiers (local to this DTB) that appear in the right hand side of the predicate; θε{<, ≦, >, ≧, =)}. If the predicate is an ON conjunct then an edge in Esj is created only if R does not appear on the preserved side of the outer join (this conjunct cannot be used for an index selection if R is in the preserved side of the outerjoin). It should be noted that that the column R.X does not have to be an index column. The set Esj will be used for (1) ordering the set of candidates during enumeration (e.g., the plans without cross products are generated first); (2) index selection; and (3) choosing the join strategies (e.g., merge or hash join selection is based on the existence of some sargable equality join conjuncts).

[0102] Esl: The sargable local edge set Esl consists of undirected edges esl=({qR}, p) for which there exists a sargable local conjunct p on a column of R, p=R.Xθexpr, with expr a constant expression and θε{<, ≦, >, ≧, =, IS NULL}.

[0103] Several examples will now be discussed to explain the structure of a Query Optimization Graph (QOG).

EXAMPLE 1

[0104] For a predicate of the form p=R.X>T.y+Z.w, only one hyperedge is defined in Esj: e=({qR}, {qT, qZ}, p).

[0105] FIGS. 5A-D illustrate Query Optimization Graph (QOG) representations of the several sample table expressions that are described below.

EXAMPLE 2

[0106] 2 $$ ( T ⁢ p T , R ⁢ R ⁢ ⟶ P R , P ⁢ P ) ⁢ ⟶ S P P , S .

[0107] For the above table expression, the set of quantifiers corresponding to the base tables is {qT, qR, qP, qS}. FIG. 5A is a graphical representation of a QOG hypergraph for the table expression shown in this example.

[0108] The subplans in V are defined as follows:

S2=({qP}, PR,P, ∅), S1=({qS}, pP,S, ∅)

e1=({qT, qR}, PT,R), e2=({qR}, S2), e3=({S2}, S1).

S0=({qT, qR, S1, S2},PT,R, {e1, e2, e3}).

Hence, V={qT, qR, qP, qS, S0, S1, S2}. E={e1, e2, e3}.

[0109] Now consider the following Example 3:

EXAMPLE 3

[0110] 3 $$ T ⁢ p T , R ⁢ R ⁢ ⟶ P ⁢   ⁢ r , P ⁢ ( P ⁢ ⟶ P p , S ⁢ S ) .

[0111] For this Example 3 table expression, the set of quantifiers corresponding to the base tables is {qT, qR, qP, qS}. The subplans in V are defined as follows:

S1=({qS}, pP,S, ∅), S2=({qP},pR,P, {e3}) where e3=(qP, S2),

S0=({qT, qR, S2}, PT,R, {e1, e2}).

V={qT, qR, qP, qS, S0, S1, S2}. E={e1, e2, e3}, where e1=({qT, qR}, pT,R), e2=({qR}, S2).

[0112] The graphical representation of a QOG hypergraph Q=<V, E> for the above table expression (Example 3) is illustrated at FIG. 5B. Note that the above two table expressions (Example 2 and Example 3) are not equivalent if the ON conjunct pp s is null-tolerant (e.g., pP,S=(P.X IS NULL or P.X=S.X)). This semantic difference is captured in the two very different QOG representations of the two table expressions illustrated at FIG. 5A and FIG. 5B, respectively.

[0113] Definition 5: The “full-outerjoin hypemode” of a fill outerjoin is the set of all quantifiers that appear in the right hand side or a left hand side of a full outer join (including the quantifiers from the nested outer joins). For each full order join, there are two full-outerjoin hypemodes defined.

[0114] Definition 6: The “null-supplying hypemode” of a left outer join is the set of all quantifiers that appear in the right hand side of the left outer join (including the quantifiers from the nested outer joins).

[0115] Definition 7: The “preserved hypemode” of a left outer join is the intersection of the set of all quantifiers that are not in the null-supplying hypemode with the set of all quantifiers referenced in the ON conjunct.

[0116] Observation 1: Note that a null-supplying hypemode is defined to be exactly the null-supplying side of the outer join according to ANSI. The same is not true for the preserved hypemode; the preserved side of an outer join is defined in ANSI to be the set of all the quantifiers that are not in the null-supplying side. The simplified definition of the preserved hypemode does not affect the ability to correctly find the original table expression.

[0117] Axiom 1

[0118] For any quantifier qεV, the set Vq={Quantifiers(S)|∃S a subplan and ∃eεEo, e=(U, S) such that qεQuantifiers(S)} is totally ordered with respect to the inclusion of sets.

EXAMPLE 4

[0119] 4 $$ ( R ⁢ ⁢ S ) ⁢ ⟶ P R , A oj1 ⁢ [ ( A ⁢ ⁢ B ) ⁢ ⟶ P B , D oj2 ⁢ ( C ⁢ ⁢ D ⁢   ) ]

[0120] FIG. 5C illustrates a QOG hypergraph representation of the above Example 4 table expression. As shown at FIG. 5C, Eo has one edge corresponding to the left outerjoin oj1, eo1, and one edge to the left outer join oj2, eo2:

e 0 2 =({qB}, S2)

e 0 1 =({qR}, S1)

where

S 2 =({qC, qD}, PB,D, ∅)

S 1 =({q A , q B , S 2 },P R,A , {e 0 2 })

[0121] For example, VqD={{qC, qD}, {qC, qD, qA, qB}} is totally ordered with respect to inclusion of sets, i.e., {qC, qD}⊃{qC, qD, qA, qB}.

EXAMPLE 5

[0122] 5 $$ ( R ⁢ ⁢ S ) ⁢ ⟶ P R , A oj1 ⁢ [ ( A ⁢ ⁢ B ) ⁢ ⟶ P S , D oj2 ⁢ ( C ⁢   ⁢ ⁢ D ⁢   ) ]

[0123] FIG. 5D illustrates a QOG hypergraph representation of the above Example 5 table expression. As shown at FIG. 5D, Eo has two edges one corresponding to the left outer join oj1, eo1, and one to the left outer join oj2, eo2:

e 0 1 =({qR}, S1)

e o 2 =({qS}, S2)

S 1 =({qA, qB},PR,A, ∅)

S 2 =({qC, qD}, pS,D, ∅)

EXAMPLE 6

[0124] For a predicate of the form p=R.X>(subquery), a hyperedge is defined in Esj: e=({qR}, q, p) where q is the quantifier corresponding to the subquery.

[0125] Definition 8

[0126] A join method is said to be input independent if and only if the input streams are always generated independently from each other.

[0127] The hash join method, for example, is input independent: the building phase reads completely the inner relation, and the probing phase reads tuples from the probing relation which are generated independently from the inner relation. The sort-merge join is also input independent. The nested loops join method and the block nested loops join method are not input independent. For both join methods, a new inner stream is generated for each tuple of the outer relation. The inner stream, in general, depends of the attribute values of the current tuple of the outer relation. For example, if the join predicate is R.X=S.X, where R is the outer relation and S is the inner relation, and an index on S.X is used for the relation S, for each tuple of R the inner stream has only the corresponding matching tuples from the relation S.

[0128] The join methods that are input independent have some interesting properties that are explored by a query optimizer constructed in accordance with the present invention:

[0129] 1. The inner and outer relations may be interchanged. For example, the probing relation for a hash join may be chosen to be any of the two children.

[0130] 2. None of the input relations depends on the expressions generated by the other input relation (i.e., the right subtree of an input independent join cannot refer to expressions generated in the left subtree and vice versa). This property, for example, restricts the indexes that may be used in the right subtree's base tables.

[0131] Conversely, the join methods that are not input independent have the following properties:

[0132] a. The inner and outer relations are not interchangeable.

[0133] b. The inner relation depends on the expressions generated by the outer relation (i.e., the right subtree may refer to expressions generated in the left subtree). For example, an index scan used for the inner base table of the nested loops join may have the matching predicate referring to the outer relation's attributes.

[0134] Definition 9

[0135] A plan node object is defined for each quantifier, subplan, and unary operator (such as a sort operator, a materializing operator, or a group by operator).

[0136] During the enumeration process, a property vector is set for each plan node. This property vector describes both the properties of the individual plan node in the context of the current partial access plan and the properties of the current partial access plan ending in this plan node. The property vector is used to compare the current access plan with the best plan that has been previously found. A main component of the property vector is the cost estimate associated with a plan node or a partial access plan. Note that each plan node in a partial access plan preserves its property vector while new plan nodes are added to the suffix of the access plan. This technique allows the prefix of an access plan to be reused without computing its properties for the set of all suffixes that the enumeration procedure might generate.

[0137] Definition 10: The set of the dependents for a plan node pn defined in a QOG Q, denoted by Dependents(pn), is a set of quantifiers defined as follows:

[0138] 1. If pn is a subplan corresponding to a null-supplying side of an outerjoin, then the set of dependents Dependents(pn) is the set of quantifiers in the preserved side of the outer join (see Definition 7 above).

[0139] 2. If pn is a subplan corresponding to a side of a full outer join, then the set of dependents Dependents(pn) is the set of quantifiers in the other full outer join hypemode (see Definition 5 above).

[0140] Definition 11

[0141] The access methods for a plan node are defined as following:

[0142] 1. For a quantifier corresponding to a base table's DTB, an access method is defined for each partial index scan that can be built given the sargable predicates of the query, for each complete index scan that can be used to satisfy some interesting properties of the query, and for the sequential scan of the base table.

[0143] 2. For a quantifier corresponding to a proper DTB, an access method is defined to represent the best access plan found for that DTB. The access method has the physical and logical properties of the best plan found for that DTB.

[0144] 3. For a proper subplan S (see Definition 4), two access methods are defined to represent the two best access plans found for that subplan:

[0145] one access method that can be used with input independent join methods, and is denoted by S<am[∅]>; and

[0146] a second access method that can be used with the rest of the join methods (see the “OPTIMIZE SUBPLAN” function described below) is denoted by S<am[Dependents(S)]>.

[0147] Definition 12

[0148] An access plan of a subplan is an ordered sequence of the subplan's plan nodes, where for each plan node the join method, the access method, the filter, and the prefilter have been set. Each plan node has associated to it a property vector that contains values for logical and physical properties such as I/O and CPU cost, intermediate result sizes, and so forth. Note that because each plan node can be a proper subplan, in accordance with the methodology of the present invention an access plan generally represents a bushy join tree.

[0149] Definition 13.

[0150] A partial access plan is a prefix of an access plan: pn0<am0>jm1pn1<am1> . . . jmkpnk<amk>. The prefix is composed of a subset of the subplan's plan nodes pn0, pn1, . . . pnk. The join methods jmi, access methods ami (e.g., complete index scans, partial index scans, sequential scans, access plans for the proper subplans (see Definition 11)), filters, and prefilters have been set for each plan node in the partial access plan. Each plan node of a partial access plan also contains a property vector describing the prefix ending with such plan node.

[0151] The following operations can be applied to a partial access plan during the enumeration procedure:

[0152] 1. A new plan node can be added after the last plan node of the prefix such that the new prefix is still a partial access plan. This operation is called “Place”. Each Place operation must compute a new property vector for the partial access plan it creates given the property vector of the new plan node added to the prefix and the property vector of the previous partial access plan.

[0153] 2. The last plan node can be removed from the prefix such that the remaining prefix is still a partial access plan. This operation is called “Unplace”. Because each plan node in an access plan contains the property vector of the prefix ending in that plan node, an Unplace operation does not need to compute the property vector of the partial access plan it creates.

[0154] FIG. 6 is a graphical representation of a subspace of the search space for a given subplan. The search subspace 610 is traversed using a depth-first search procedure. At each step of the depth-first search, next candidates (i.e., plan segments that are candidates for inclusion in the access plan) are dynamically generated, such that, at any time during the enumeration process, the methodology of the present invention only requires storing the following information: (a) information about the current partial access plan; and (b) for each plan node used in the current access plan, information indicating how to generate the next candidates to be considered. The next candidates can be (1) a set of plan nodes; and (2) sets of (join method, access method) pairs as shown at FIG. 6. The next candidates include these two different types of candidates because the methodology of the present invention involves enumerating both what relations are to be added next to the prefix, and how they are added (i.e., choices of access method (e.g., index) and join method). Thus, the next candidates for a given plan node are tuples of the form (join method, access method) (i.e., (jmi, ami) pairs). The choices of access method (e.g., index) and join method are dependent on each other to a considerable extent, and for this reason pairs of candidates (jmi, ami) are generated together in the currently preferred embodiment instead of selecting them separately (e.g., choosing first the access method and then the join method). The set of candidates is dependent on the prefix of the plan (e.g., a join method requiring an equi-join predicate (for example, hash join) cannot be considered if the equi-join predicate cannot be placed given the current prefix). In the currently preferred embodiment, sets of all access methods (see Definition 11) and join methods that can be used for a plan node are built during a pre-optimization phase. During enumeration, the set of candidates (jmi, ami) is selected from these pre-computed candidate sets such that each candidate is valid in the context of the current prefix. If a plan node pnk is successfully placed in the last position of the prefix, having the join method and the access method chosen to be (jmk, amk), then a new set of plan node candidates is computed for this new prefix.

[0155] The current partial access plan 620 is shown at the right hand side of FIG. 6. Each pni is a plan node that belongs to the subplan and can represent another subplan or a quantifier. A candidate access method of the plan node pnj is denoted as amj, while jmj represents a candidate join method that can be used to join pnj<amj> with the prefix of the plan. If pnj represents a proper subplan, the partial access plan containing pnj<amj> is in fact a bushy tree.

[0156] Consider the following query Q1 shown below as Example 7:

EXAMPLE 7

[0157] SELECT *

[0158] FROM Nation

[0159] (Customer LEFT OUTER JOIN Order ON C_Custkey=O_Custkey AND O_Orderdate<‘1995-12-31’)

[0160] LEFT OUTER JOIN (Lineitem JOIN Part ON L_Partkey=P_Partkey) ON 0 OrderKey=L_Orderkey AND P_Type=‘%TIN%’

[0161] WHERE C_Nationkey=N_Nationkey AND C_Acctbal>7000 AND N_Name=‘EGYPT’

[0162] Some elements of the QOG for Q1 are defined as follows (excluding here a description of the edges):

[0163] V={qN, qC, qO, qL, qP, S0, S1, S2}.

[0164] S2=({qL, qP}, O_OrderKey=L_Orderkey AND P_Type=‘%TIN%’, . . . )

[0165] S1=({qO}, C_Custkey=O_Custkey, . . . )

[0166] S0=({qN, qC, S1, S2}, C_Nationkey=N_Nationkey AND C_Acctbal>7000 AND N_Name=‘EGYPT’, . . . ).

[0167] FIG. 7 is a graphical representation of a portion of the search space for a sample subplan S0 for the above query Q1 described above as Example 7 which illustrates the optimization methodology of the present invention. Assume that the access methods for the subplan S1 are:

[0168] S1<am[∅]>=Orders<seq>, and

[0169] S1<am[qC]>=Orders<FK_C>;

[0170] and that the access methods for subplan S2 are:

[0171] S2<am[∅]>=Part<seq>JH Lineitem<seq>, and

[0172] S2<am[qO]>=Lineitem<FK_O>JNL Part<PK>.

[0173] As shown at FIG. 7, the search space 710 includes the two highlighted paths. These two highlighted paths in the search space represent two access plans for the subplan S0. The first access plan for the subplan S0 is designated AP1 720 and the second access plan is designated AP2 730. These two access plans are as follows:

[0174] AP1=qN(NULL,<seq>)qC(JNL,<FK N>) S1 (JHO, S1<am[∅]>) S2 (JNLO, S2<am[qO]>)

[0175] AP2=qN(NULL,<seq>) qC((JNL,<FK N>)S1(JHO, S1<am[∅]>)S2 (JHO, S2<am[·]>)

[0176] Theoretically, the two access plans correspond to the join trees depicted in the two AP inserts AP1720 and AP2730 shown at FIG. 7. Note that the access methods for the subplan S2 (i.e., S2 (JNLO, S2<am[qO]>) and S2 (JHO, S2<am[Ø]>) correspond to the two best plans found for the subplan S2. The methods of operation of the present invention will now be described.

[0177] D. Methods of Operation

[0178] FIGS. 8A-C comprise a single flowchart 800 illustrating at a high level the methods of operation of the present invention in generating bushy trees using a left-deep tree join enumeration algorithm. The following description presents method steps that may be implemented using computer-executable instructions, for directing operation of a device under processor control. The computer-executable instructions may be stored on a computer-readable medium, such as CD, DVD, flash memory, or the like. The computer-executable instructions may also be stored as a set of downloadable computer-executable instructions, for example, for downloading and installation from an Internet location (e.g., Web server).

[0179] The process commences at step 801 with the receipt of a database query (e.g., a query requesting information from a plurality of database tables). At step 802, rewrite optimizations are applied and a semantically equivalent rewritten query is generated. For example, the nested outer joins are semantically analyzed and transformed into inner joins whenever possible.

[0180] At step 803, a Query Optimization Graph (QOG) is generated for each query block or Derived Table Block (DTB). The QOG contains “subplan” objects and also includes arrays of access methods (e.g., relevant indexes), arrays of join methods, a join dependency graph, and an index dependency graph as previously described. In the currently preferred embodiment, all potential join methods and access methods for each plan node are also built during this phase.

[0181] At step 804, the optimization methodology of the present invention commences the process of examining each query block in order to determine the most favorable access plans for each such query block. This process proceeds with examining each query block in a bottom-up fashion in order to build an access plan for providing data in response to the query. The process of optimizing each query block includes consideration of all the nested subplans as well as outer subplans of a query block. For each subplan in the QOG, determining the best access plans starts with the most nested plans as provided at step 805.

[0182] The generation of join trees for each subplan of a query block is performed in a step-by-step manner. A current access plan (also referred to herein as a “prefix”) is built by adding plan nodes together with their join methods and access methods to the current prefix that is being considered. In other words, at each step a plan node (e.g., a join with relations that are not yet part of the access plan) is added to the current partial access plan. The current prefix contains the join methods and access methods for all plan nodes that have already been chosen using the methodology of the present invention, and the current prefix is already costed. More particularly, the process is as described below.

[0183] At step 806 a set of plan node candidates to be placed in the next position of the current access plan under consideration is generated. Next, at step 807 a particular plan node candidate (pn) is selected from the set of candidates generated at step 806. At step 808, a set of join method, access method (jm, am) candidates to be used with the current plan node pn is generated. Next at step 809, a new access plan is created by adding the candidate plan node pn together with a candidate join method and access method (jm, am) to the current prefix (i.e., “placing” the pn(jm, am) candidate plan segment in the current access plan).

[0184] At step 810, the estimated property vector of the new prefix (which may (or may not) contain all the plan nodes) is compared to the estimated property vector of the best complete plan found so far. The estimated property vector includes pipeline characteristics as well as execution costs computed based upon such factors as CPU cost and input/output (I/O) cost. Based upon this comparison, the method proceeds to step 811 if the current prefix under consideration is less favorable than the best plan previously found. Otherwise, if it is more favorable the method proceeds to step 812.

[0185] If the current partial access plan under consideration is estimated to be less favorable than the best plan previously found, then at step 811 the current prefix is modified by pruning (or “unplacing”) the last plan node that was added in the last position (i.e., the node placed at step 809). In this case steps 809, and 810 are repeated with the examination of new join methods and access methods for a plan node while candidate join methods and access methods (jm, am) are available for the current plan node pn. If all candidate join methods and access methods (jm, am) for a given plan node have been examined, then steps 808, 809, and 810 are repeated while additional candidate plan nodes are available. This process may continue until all plan nodes and their join methods and access methods have been examined at this position. It should be noted that this comparison process may also be interrupted if it is determined that continuing the search of a particular subspace will not improve the best plan previously found or if the effort spent in the current search space exceeds an established threshold as described below.

[0186] If the current partial access plan under consideration (i.e., the one built at step 809) has an estimated property vector that is more favorable than the best plan previously found, the current prefix is retained and the method proceeds to step 812 or 813. At step 812, a determination is made as to whether the current access plan represents a complete plan for the subplan currently under consideration. If the current access plan is a complete plan, the plan is retained as the new best plan and the method proceeds with examining additional alternatives while candidates remain available. Otherwise, at step 813 if the current access plan is not a complete plan, the method proceeds to examine the next position in the subtree and returns to step 806 to examine one or more plan nodes for the next positions in the suffix of the current access plan.

[0187] This process continues until all portions of the current search subspace have been considered (or until the quota available for considering alternatives is exhausted as hereinafter described). At step 814, steps 805-813 are then repeated for additional query blocks until all query blocks have been optimized. At step 815, after all query blocks have been optimized, a detailed access plan for providing the data requested by the query is constructed based upon the most favorable plans identified for each query block as a result of the foregoing steps.

[0188] The methodology of the present invention requires little memory as at any point in the enumeration process only the information about the best plan that has been found and the current partial access plan that is being considered must be stored. Hence, the memory required by is far less than that required by other techniques. The operations of the system of the present invention will now be described in greater detail.

[0189] E. Detailed Internal Operation

[0190] The process of optimization of a query block or DTB commences with a call to the following “OPTIMIZE DTB” procedure for the root DTB of a query.

1: Procedure: OPTIMIZE DTB
2: Purpose: Optimize a Derived Table Block (DTB).
3: Create the Data Flow Operator (dfo) tree representation for
the root DTB.
4: Input:
5: - dtb0 a parse tree representation of a DTB.
6: - q0 the quantifier ranging over dtb0
7: Output:
8: - A dfo execution tree if dtb0 is the root DTB.
9:
10: begin
11:  Q  BuildQOG(dtb0)
12:  for each quantifier q ∈ Q do
13:   dtb  DTB(q)
14:   call OPTIMIZE DTB(dtb, q)
15:  od
16:  call OPTIMIZE SUBPLAN(Subplan(Q),Q)
17:  if q0 is not null then
18:   set the properties of the access plan of q0, denoted by
q0 < seq >, to be the properties of APbest (Subplan (Q)) [∅].
19:   return null
20:  else
21:   dfo  BuildDFO(Q)
22:   return dfo
23:  fi
24: end

[0191] As shown at line 11, a Query Optimization Graph (QOG) is built for each Derived Table Block (DTB). The optimization of the DTBs built for the query is performed in bottom-up fashion as illustrated by the call to OPTIMIZE DTB at line 14. After each DTB is optimized, the outer subplan of the root DTB is optimized by a call to “OPTIMIZE SUBPLAN” at line 16. (The “OPTIMIZE SUBPLAN” method is described below). It should be noted that after a nested DTB is optimized, the properties of the best access plan found for the DTB are set to be the properties of the access method for the quantifier ranging over that DTB as provided at line 18 above. For example, if a DTB is a base table DTB, the main access method for the quantifier ranging over that DTB is the sequential scan which has the physical and logical properties set to be the properties of accessing the base table sequentially. If the DTB is the root DTB (the only DTB without a quantifier), the Data Flow Operator (DFO) tree is built for the best plan found by reconstructing, in a bottom-up manner, the best plans found for each nested DTB as provided above at line 21.

[0192] The following “OPTIMIZE SUBPLAN” procedure is called by the above “OPTIMIZE DTB” function for each outer subplan of a DTB:

1: Procedure: OPTIMIZE SUBPLAN
2: Purpose: Optimize a subplan defined in a QOG.
3: Optimal access plans are found and saved in the following contexts:
4:  (1) all outer references are available; and
5:  (2) none of the outer references is available.
6: Input:
7: - A QOG Q of a DTB.
8: - A subplan S defined in Q.
9: Notations:
10: - Subplans(S) is the set of proper subplans defined for S.
11: - Dependents(S) is the set of outer reference quantifiers
12:  which are referenced by the subplan S.
13:
14: begin
15:  for each subplan node sp ∈ Subplans(S) do
16:   call OPTIMIZE SUBPLAN(sp,Q)
17:  od
18:  or_availQs  Dependents(s)
19:  call ENUMERATE(Q, S, or_availQs)
20:  set the properties of the first access method for the subplan S,
i.e., S < am[or_availQs] >, to be
21:  the properties of APbest(S) [or_availQs].
22:  if or_availQs == ∅ then
23:   return
24:  else
25:   call ENUMERATE(Q, S, ∅)
26:   set the properties of the second access method for the subplan
S, i.e., S < am[∅] >, to be the properties of APbest(S) [∅].
27:  fi
28: end

[0193] The above “OPTIMIZE SUBPLAN” method is not only called for each outer subplan of a DTB, but is also called for each nested subplan as illustrated above at line 16. A subplan which is not an outer subplan is generally optimized twice. Typically, two access methods are set as access methods for that subplan as follows: APbest(S)[∅] and APbest (S)[Dependents(S)]. Logically each intermediate relation corresponding to a proper subplan can be computed in two ways. First, it can be computed completely in isolation from the rest of the plan nodes defined in the parent DTB as shown at line 25. Second, it can be computed for each tuple generated by the outer reference dependents as provided at line 19. The access method defined by APbest(S)[∅] (denoted by S<am[∅]>) is used in conjunction with input independent join methods, while the access method defined by the APbest(S)[Dependents(S)] (denoted by S<am[Dependent(S)]>) is used in conjunction with the nested join methods for which the subplan is the inner relation and the Dependents(S) is the outer relation.

[0194] The following “ENUMERATE” procedure is called recursively to add a new plan node with its access method and join method to a partial access plan:

1: Procedure: ENUMERATE
2: Purpose: Enumerate join trees and set the best access plan for
3: a subplan S when the set of outer reference quantifiers is
4: or_availQs.
5: Input:
6: - A QOG Q of a DTB.
7: - A subplan S defined in Q.
8: - A set of outer reference quantifiers or_availQs.
9: Notations:
10: - APpartial(S) is the current partial access plan defined for S.
11: - APbest(S) [or_availQs] is the best access plan found so far for S
when the set of available outer reference quantifiers is
or_availQs.
12: Local variables:
13: - PNc is the set of next plan nodes candidates.
14: - JAMc is the set of <join method, access method> candidates.
15: - prune is set to TRUE if APpartial(S) is less optimal than
APbest(S) [or_availQs].
16: - c_availQs a set of currently available quantifiers.
17:
18: begin
19:  PNc  Iterator(FIND NEXT PLANNODES CANDIDATES(Q, S, or_availQs))
20:  for each plan node pn  Next(PNc) do
21:   c_availQs  or_availQs ∪ Quantifiers(pn) ∪ Quantifiers
(APpartial(S))
22:   JAPC
23:   Iterator(FIND NEXT JOIN ACCESSMETHOD CANDIDATES(Q, S, pn,
c_availQs))
24:   for each <join method, access method> candidate < jm, am >
Next(JAPC) do
25:    APpartial(S)  Place(pn,< jm, am >, APpartial(S))
26:    prune  Compare(APpartial(S), APbest(S) [or availQs])
27:    if !prune then
28:     if APpartial(S) is complete then
29:      APbest(S) [or_availQs]  APpartial(S)
30:     else
31:      call ENUMERATE(Q, S, or_availQs)
32:     fi
33:    fi
34:   APpartial(S)  Unplace (pn,< jm, am >, APpartial(S))
35:   od
36:  od
37:  return
38: end

[0195] One of the input parameters to the above “ENUMERATE” method is the subplan S that is currently being optimized. The information about the current partial access plan, APpartial(S), as well as the best plan found so far, APbest(S)[or_availaQs], is stored in the subplan itself. At line 19 above, plan nodes that are candidates for the next position in the current partial access plan are found through a call to the “FIND NEXT PLANNODES CANDIDATES” function described below.

[0196] For each plan node candidate, a set of potential candidates (join method, access method) is generated by calling a “FIND NEXT JOIN ACCESS METHOD CANDIDATES” function as shown at line 23. A new partial access plan is created by adding the candidate plan segment pn<jm, am> (the candidate plan segment represents a plan node pn with the <jm, am> representing a candidate join method and access method pair) to the current access plan as illustrated at line 25. The “Place( )” function at line 25 computes for the new plan node pn the property vector for the new partial access plan (e.g., number of rows in the intermediate result represented by the current access plan, the cost of the current access plan, and so forth). The cost function used for the cost component of the property vector of a partial access plan is a monotonically increasing function which uses metrics such as buffer pool availability, the number of cached data pages, the properties of the base table indexes, and the like.

[0197] The “Compare” function at the line 26 compares the current partial access plan with the best plan previously found. The essence of the branch-and-bound paradigm is the pruning strategy applied when the current partial access plan is considered more expensive than the best complete plan already found: in this case, the last candidate plan node (join method, access method) pair and the search subspace that would have been generated for this partial access plan is abandoned. The “Unplace( )” operation at line 34 restores the last prefix of the current access plan such that a new candidate can be tried. The “Compare( )” function may also be advised by the optimizer governor to abandon a current search subspace if the optimizer governor believes that the search subspace will not improve the best plan previously found or if the effort spent in the current search space exceeds the current quota. The optimizer of the currently preferred embodiment employs an optimizer governor to control the effort spent on plan enumeration. The optimizer governor controls the effort spent on each search subspace by measuring the amount of effort as the number of plan nodes visited, choosing a quota of how many plan nodes the enumeration process visits, and recursively distributing this quota over the search space during enumeration. Hence, even if the current partial access plan is not inferior to the best plan previously discovered, the enumeration process might abandon the current search subspace in favor of trying new subspaces. For further description of the operations of the optimizer governor in controlling the effort spent on join enumeration, see commonly owned co-pending U.S. application Ser. No. 10/249,791 entitled “Database System with Methodology for Distributing Query Optimization Effort over Large Search Spaces”, Docket No. SYB/0091.01, the disclosure of which is hereby incorporated by reference.

[0198] The “FIND NEXT PLANNODES CANDIDATES” function which is called by the above “ENUMERATE” function is as follows:

1: Procedure: FIND NEXT PLANNODES CANDIDATES
2: Purpose: Find a set of plan nodes to be considered next for the
3: current partial access plan APpartial(S).
4: Input:
5: - A QOG Q of a DTB.
6: - A subplan S defined in Q.
7: - A set of available quantifiers availQs.
8: Output:
9: - An ordered set of plan nodes candidates.
10: Notations:
11: - APpartial(S) is the partial access plan defined for S.
12: - PlanNodes(S) is the set of plan nodes of the subplan S
13: - PlanNodes(APpartial(S)) is the set of plan nodes placed in the
partial access plan APpartial(S).
14: - Quantifiers(APpartial(S)) is the set of quantifiers in the
15: partial access plan APpartial(S).
16: - sch(p) is a set of local and outer reference quantifiers
referenced by the predicate p.
17: - Dependents(pn) is the set of the dependents (see Definition 10).
18: Local variables:
19: - CP is a set of plan nodes not connected with APpartial(S) (i.e.,
20: the plan nodes that will introduce a Cartesian product).
21: - PN is a set of plan nodes.
22:
23: begin
24:  PN  ∅
25:  CP  ∅
26:  for each plan node pn ∈ PlanNodes(S) \ PlanNodes
(APpartial(S)) do
27:   if Dependents(pn) ⊂ availQs then
28:    if pn has a predicate (p) such that sch(p) ⊂
Quantifiers (APpartial(S)) then
29:     PN  PN ∪ {pn}
30:    else
31:     CP  CP ∪ {pn}
32:    fi
33:   fi
34:   if PN == ∅ then
35:    PN  CP
36:   fi
37:   if PN != ∅ then
38:    call RankPlanNodeCandidates(PN)
39:    call SortPlanNodeCandidates(PN)
40:   fi
41:  od
42:  return PN
43: end

[0199] Given the current partial access plan, the above “FIND NEXT PLANNODES CANDIDATES” function computes an ordered array of potential candidates for the next position in the access plan. The “RankPlanNodeCandidates” routine at line 38 uses heuristics to rank each plan node (PN) based on factors which include the size of the relations and the number of join predicates with the prefix of the partial access plan. At line 35, the Cartesian products are deferred as long as there exist candidates that can be joined with the prefix of the plan by checking at line 28 whether any join predicate can be used given the current prefix. After the candidate plan nodes are ranked they are then sorted as illustrated at line 39.

[0200] The following “FIND JOIN ACCESSMETHOD CANDIDATES” function is called to generate an array of join method and access method (jm, am) pairs to be used with a current plan node candidate pn:

1: Procedure: FIND NEXT JOIN ACCESSMETHOD CANDIDATES
2: Purpose: Find a set of <join method, access method> candidates
3: to be considered for the plan node pn to be placed next to a
4: partial access plan APpartial(S).
5: Input:
6: - A QOG Q of a DTB.
7: - A subplan S defined in Q.
8: - A plan node pn to be placed next in the current partial access
plan.
9: - A set of available quantifiers availQs.
10: Output:
11: - JMAMs an ordered set of <join method, access method> candidates.
12: Notations:
13: - APpartial(S) is the partial access plan defined for S.
14: - PlanNodes(S) is the set of plan nodes of the subplan S.
15: - PlanNodes(APpartial(S)) is the set of plan nodes placed in the
partial access plan APpartial(S).
16: - Quantifiers(APpartial(S)) is the set of quantifiers in the
17: partial access plan APpartial(S).
18: - JoinPredicates(jm) is the set of join predicates defined for the
19: join method jm. (Note: all join methods but the Cartesian product
require at least one join predicate.)
20: Local variables:
21: - JMs is a set of candidates join methods.
22: - AMs is a set of access method candidates for the plan node pn.
23: - preds is a set of predicates.
24: - iiavailQs is a set of quantifiers whose expressions may be used
25: when choosing an access method for the plan node pn for join
26: methods that are input independent.
27:
28: begin
29:  JMs  ∅
30:  iiavailQs  availQs \ Quantifiers (APpartial(S))
31:  for each join method jm ∈ JoinMethods(pn) do
32:   if jm does not require join predicates then
33:    JMs  JMs ∪ {jm}
34:   else
35:    preds  JoinPredicates(jm)
36:    for each p ∈ preds do
37:     if sch(p) ∩ availQs != ∅then
38:      JMs  JMs ∪ {jm}
39:     fi
40:    od
41:   fi
42:  od
43:  JMAMs  ∅
44:  for each join method jm ∈ JMs do
45:   if jm is input independent (Definition 8) then
46:    AMs  FIND ACCESSMETHOD CANDIDATES(QOG, S, pn, jm,
iiavailQs)
47:   else
48:    AMs  FIND ACCESSMETHOD CANDIDATES(QOG, S, pn, jm, availQs)
49:   fi
50:   JMAMs  JMAMs ∪ {(jm, am)|am ∈ AMs}
51:  od
52:  if JMAMs != ∅then
53:   call RankJoinAccessMethodCandidates (JMAMs)
54:   call SortJoinAccessMethodCandidates (JMAMs)
55:  fi
56:  return JMAMs
57: end

[0201] The join methods are valid candidates if they can be used given the current partial access plan. For example, at line 35 a hash join method may be used if and only if at least one equi-join predicate may be placed. In the optimizer of the currently preferred embodiment, all potential join methods and access methods for each plan node are built during a pre-optimization phase. In other words, the set JoinMethods(pn) is already built for the node pn as shown at line 31. For example, if a predicate of the form T.X=R.X is a WHERE predicate, a hash join method with the equi-join predicate T.X=R.X is built for the quantifier corresponding to table T, and a hash join method with the equijoin predicate T.X=R.X is built for the quantifier corresponding to table R. These prebuilt objects allow the enumeration process to quickly generate the search spaces with little effort.

[0202] The below “FIND ACCESSMETHOD CANDIDATES” pseudocode illustrates the procedure used for identifying access method candidates:

1: Procedure: FIND ACCESS METHOD CANDIDATES
2: Purpose: Find a set of access methods candidates
3: to be considered for the plan node pn to be placed next in
4: a partial access plan APpartial(S) when a join method jm is
5: to be used.
6: Input:
7: - A QOG Q of a DTB.
8: - A subplan S defined in Q.
9: - A plan node pn to be placed next in the current partial
access plan.
10: - A join method jm to be used for the plan node pn.
11: - A set of available quantifiers avail Qs.
12: Output:
13: - A set of access method candidates.
14: Notations:
15: - APpartial(S) is the partial access plan defined for S.
16: - PlanNodes(S) is the set of plan nodes of the subplan S.
17: - PlanNodes(APpartial(S)) is the set of plan nodes placed in
18: the partial access plan APpartial(S).
19: - Quantifiers(APpartial(S)) is the set of quantifiers in the
20: partial access plan APpartial(S).
21: - GetOptimalAccessMethods(pn, availQs) returns an already found
22: access method for the subplan pn when the available
quantifiers are the set availQs.
23: - HasMatchingPredicates(idx, availQs) returns TRUE if the
24: relevant index idx may be used (e.g., it has at least one
matching predicate p such that sch(p) ⊂ availQs.
25: Local variables:
26: - AMs is a set of access method candidates for the plan node pn.
27: - preds is a set of predicates.
28: begin
29:  if pn is a subplan then
30:   AMs  GetOptimalAccessMethods(pn, availQs)
31:  else if pn is a quantifier then
32:    for each relevant index idx ∈ RelevantIndexes(pn)do
33:     if HasMatchingPredicates(idx, availQs) then
34:      AMs  AMs ∪ {idx}
35:     fi
36:    fi
37:   od
38:  fi
39:  return AMs
40: end

[0203] The above “FIND ACCESS METHOD CANDIDATES” function also relies on pre-built objects in finding the next access method candidates. The set RelevantIndexes(pn) shown at line 32 is built during the pre-optimization phase to include all the partial or complete index scans for a base table that may be useful in the context of optimization of the query. For example, if a predicate of the form R.X>10 is defined in the WHERE clause, and there exists an index I=(R.X ASC) for the table R, then the RelevantIndexes(qR) contains a relevant index structure to represent a partial index scan on index I with the sargable predicate R.X>10. If the sargable predicates depend on the prefix of the plan (e.g., R.X>T.X) then a partial index scan can be used if and only if the expressions needed to evaluate the sargable predicates are generated by the prefix of the plan. For this reason, the “HasMatchingPredicates” routine at line 33 checks this property. The set RelevantIndexes(pn) at line 32 also contains complete index scans that might satisfy interesting properties of the query, such as ordering properties required by an ORDER BY clause. For a subplan S, the access methods are pre-computed before the current subplan is optimized (as described above in the “OPTIMIZE SUBPLAN” pseudocode). The function “GetOptimalAccessMethods( )” at line 30 simply recalls these access methods for a proper subplan (i.e., one or both of S<am[Dependent(S)]> and S<am[∅]>).

[0204] F. Empirical Results

[0205] The current implementation of the ASA optimizer allocates memory for optimizer objects in two ways: the fast heap-based allocation for the optimizer objects; and the stack-based allocation during enumeration for the next candidate arrays. The prebuilt optimizer's objects such as relevant indexes and join methods all use fast heap-based allocations. The arrays of the candidates generated during optimization are allocated on the stack. The following table depicts memory usage for some complex queries with the number of quantifiers for a DTB ranging from 6 to 700. The “Total #Qs” column records the total number of quantifiers built for the whole query. The column “Max #Qs” records the maximum number of quantifiers found in a subplan. The “#DTB(b+d)” records the number of base table DTBs and derived table DTBs built for the whole query. The column “#pn enum” shows the total number of plan segments enumerated during optimization. The columns “Stack(k)” and “Memory(k)” shows the memory usage in kilobytes, stack and heap, respectively, used during optimization process.

	Total
Query	#Qs	Max #Qs	#DTBs(b + d)	#pn enum	Stack(k)	Memory(k)
TPCH Q2	11	9	9 + 3	1138	4.1	56
TPCH Q5	7	6	6 + 2	3534	5.2	32
TPCH Q7	7	6	6 + 2	1464	5.2	32
TPCH Q8	9	8	8 + 2	7946	6.0	40
TPCH Q9	7	6	6 + 2	5759	5.2	32
TPCH Q11	9	6	6 + 4	66	4.1	40
TPCH Q21	7	6	6 + 4	663	5.2	40
V1	594	6	371 + 224		5.4	3100
V2	911	6	530 + 382		5.4	4792
V3	700	700	700 + 1	13031	277.3	3996

[0206] III. Cost-Based Subquery Optimization Using Left-Deep Join Enumeration Methodology

[0207] A. Overview of Cost-Based Enumeration of Subqueries

[0208] The present invention comprises a system providing methodology for cost-based enumeration of subqueries using a left-deep tree join enumeration algorithm in a relational database system. The methodology of the present invention comprises defining, for each qualifying subquery, a set of access methods (or execution methods), and a set of join methods to be used to join such subqueries with the rest of the query's tables. The access methods are defined based on the subquery execution strategies which could be employed for a subquery given its type and the semantic properties of the query the subquery is used in. Many subquery execution strategies have been proposed in research literature (see for example Kim, W., “On Optimizing an SQL-like Nested Query”, in ACM Transactions on Database Systems, 7(3): pp. 443-469, September 1982). The execution strategies utilized by the present invention are the execution of the correlated and uncorrelated subqueries in their original form, and the execution of correlated subqueries as uncorrelated derived tables without and with hash filters as illustrated in Example 8 and described below in more detail.

[0209] The enumeration methodology is further modified to treat subqueries in a similar way with the base or derived tables, by enumerating and costing subqueries using these special subquery access and join methods. Each subquery participating in the enumeration process has assigned a property vector (containing CPU cost, I/O cost, order properties, etc.) similar to the property vector used for any other base table, which is recomputed for each partial access plan ending with that subquery. Each access method of a subquery participating in the enumeration process has a property vector associated to it computed by optimizing the subquery's Derived Table Block (DTB) specifically for that access method (as described below in the discussion of the “OPTIMIZE_DB” procedure). Each special join method associated to an access method of a subquery implements the computation of its property vector during the enumeration process (e.g., each subquery join method computes its own cost based on the current subquery access method and the prefix of the current access plan under consideration). The access methods and join methods associated to a subquery are treated in the same fashion with any other access methods (e.g., a partial index scan) or join methods associated to base or derived tables. Compared to other cost-based subquery optimization techniques, the advantages of the present invention include the following:

[0210] (1) The methodology of the present invention provides that the subquery participating in the enumeration process is not transformed during optimization to reflect different subquery execution strategies. Instead, after the optimization is completed, the transformation required by the best access plan is applied in order to build the Data Flow Operator (DFO) tree corresponding to the chosen access method/join method for a subquery. This approach of the present invention uses far less memory than other transformation-based subquery optimization techniques where a SQL query is rewritten to reflect a certain subquery evaluation strategy, and, for each rewritten query, the optimization process is applied.

[0211] (2) The optimizer does not have to define new constructs to make it possible to enumerate subqueries just as any other base or derived tables. The modification of the enumeration methodology is minimal which makes it possible to easily consider new subquery execution strategies.

[0212] (3) The execution engine does not have to define new DFO operators in order to apply different subquery execution strategies. The join methods and access methods associated to a subquery correspond to DFO operators already defined for base and derived tables.

[0213] As opposed to transformation-based techniques for subquery optimization which potentially can increase the search space and, as a consequence, the memory required by the enumeration process, the present invention does not perform any transformations to the original subquery until the Data Flow Operator (DFO) tree is built from the best access plan found by the optimizer. Moreover, because the subqueries are treated like any other base or derived tables during the enumeration process, the enumeration process has a minimum number of modifications for accommodating the cost-based optimization of subqueries. Equally important, the access plans generated by the optimizer for different subquery evaluation techniques use only already available Data Flow Join Operators such as Join Hash, Join Hash Semijoins, Join Hash Antisemijoins, Join Nested Loops, etc.

[0214] B. Example of Cost-Based Subquery Optimization

[0215] Consider the following query Q1 shown below as Example 8: Example 8:

Q1: SELECT *
    FROM  Lineitem as L1, Part as P
    WHERE L1.1_partkey = P.p_partkey
          AND P.p_brand =‘Brand#23’
          AND P.p_container = ‘MED BOX’
          AND L1.1_quantity < ( SELECT
          AVG(L2.1_quantity )
          FROM Lineitem L2
          WHERE L2.1_partkey = P.p_partkey )

[0216] Some elements of the QOG for the DTB corresponding to the main query block of the Q1 are defined as follows (excluding here a description of the edges):

V={qP, qL1, qsub, S0}.

S0=({qP, qL1, qsub}, {P1, p2, p3, , p4}, . . . ).

[0217] where

[0218] qP, qL1, qsub are the quantifiers corresponding to the tables Part P, Lineitem L1, and the subquery, respectively.

[0219] p1=‘L1.1_partkey=P.p_partkey’

[0220] p2=‘P.p_brand=‘Brand#23″

[0221] p3=‘P.p_container=“MED BOX”

[0222] p4=‘L1.1_quantity=(SELECT AVG(L2.1_quantity)

[0223] FROM Lineitem L2

[0224] WHERE L2.1_partkey=P.p partkey)’

[0225] FIGS. 9A-B comprise a graphical representation of a portion of the search space for the subplan S0 for the query Q1 described above as Example 8 which illustrates the optimization process for a query using the cost-based subquery optimization methodology of the present invention. As shown at FIG. 9A, the search space 910 comprises the search space for the subplan S0. The subquery is treated as any other plan node, the costs of its access methods (shown in the box 920 at FIG. 9A) are used for each enumerated plan, the costs for its join methods are computed based on where the subquery is placed in a partial access plan.

[0226] The three inserts 930, 940, and 950 shown at FIG. 9B depict three access plans AP1, AP2, and AP3, respectively, enumerated by the enumeration algorithm of the present invention. (AP1 is illustrated at 930, AP2 is illustrated at 940, and AP3 is illustrated at 950 at FIG. 9B). As shown, this particular subquery can have three execution strategies:

[0227] (1) Its original form in the predicate p4 (i.e., correlated subquery execution strategy). The access method for this strategy is denoted by qsub<DFP> in box 920 at FIG. 9A. (The acronym DFP stands for Data Flow Predicate operator used to evaluate predicates). The subquery's DTB for this access method has the best access plan, found by the optimizer, GrByS[Lineitem L2<FK_PKEY>]. For this access method, the subquery is joined with the rest of the query block with a dummy join, denoted by JDFP in FIG. 9A. Choosing this access method for the subquery is equivalent to placing, based on cost, the predicate p4 in an access plan of the query. The cost of the access method qsub<DFP> is the cost of the best plan found for the original subquery's DTB. The cost of the dummy join, JDFP, is the cost of evaluating the subquery N times, where N is the number of rows in the left hand side of the JDFP (i.e., the number of rows in the prefix of the plan before the JDFP is placed). The plan depicted in AP2 shown in insert 940 at FIG. 9B contains the plan segment (JDFP, qsub<DFP>) and corresponds to the access plan containing the filter with the predicate p4 placed after the join of the tables Part P and Lineitem L1.

[0228] (2) As an uncorrelated derived table (i.e., aggregate and then join subquery execution strategy). The uncorrelated derived table for this example corresponds to the table:

[0229] DT(avg, partkey)=(SELECT AVG(L2.1_quantity), L2.1_partkey

[0230] FROM Lineitem L2

[0231] GROUP BY L2.1_partkey)

[0232] The access method qsub<DT> corresponds to the evaluation of this Derived Table (DT). The cost of this access method is the cost of the best plan found for the subquery's DTB optimized for the alternative of an uncorrelated derived table. The optimizer doesn't modify the subquery's DTB into a new DTB corresponding to this derived table DT. Instead, the optimization is applied to the original DTB where some compensation elements are applied to simulate the semantic of the derived table DT (for example, the correlated predicates are seen as TRUE predicates, or the GROUP BY list contains now the item “L2.1_partkey” instead of its original form of an empty list). The compensation predicates created for this access method are:

[0233] P5=‘L1.1_quantity=DT.avg’

[0234] p6=‘P.p_partkey=DT.p_partkey’

[0235] These two compensation predicates are used to decide what join methods can be used to join the subquery with the access method qsub<DT> with the rest of the main query block. Because we have equality predicate P6, one join method is JH (Join Hash) with the join predicate P6 and/or P5 One access plan enumerated by the optimizer using the plan segment (JH, qsub<DT>) is depicted in the AP1 in insert 930 at FIG. 9B.

[0236] (3) As an uncorrelated derived table with a hash filter (i.e., the filter, aggregate and then join subquery execution strategy, the new execution strategy provided by the present invention). The derived table is similar to the derived table described above in execution strategy (2), except that filter predicates using the hash filters replace the correlated equality predicates from the original subquery's definition.

[0237] DTh(avg, partkey)=(SELECT AVG(L2.1_quantity), L2.1—partkey

[0238] FROM Lineitem L2

[0239] WHERE L2.p_partkey IN HFP P.p_partkey

[0240] GROUP BY L2.1_partkey)

[0241] Note that this subquery execution strategy filters out any groups that cannot participate in the join with the outer block's tables; therefore the size of the derived table could be substantially smaller than the size of the derived table considered above in execution strategy (2). However, the methodology of the present invention provides for costing and enumerating both subquery execution strategies because the derived table DT can be placed anywhere in an access plan (e.g., DT can be the first plan node in the access plan) while DTh can be placed only after the tables that provide the hash filters used in the WHERE clause (in the above example, DTh can be used only after the table PART). The necessary conditions for a subquery to qualify for being executed using this strategy are similar to those for the “aggregate and then join” strategy, and are well known to those skilled in the art. The only additional condition is that the correlated predicates are equality predicates such that the hash filters can be built using the correlated expressions. The access method qsub<DTh> corresponds to the evaluation of this Derived Table DTh. The cost of this access method is the cost of the best plan found for subquery's DTB optimized for the alternative of an uncorrelated derived table given that the hash filter is available when the evaluation of the DTh is done. One access plan enumerated by the optimizer using the plan segment (JH, qsub<DTh >) is depicted in the AP3 shown in insert 950 at FIG. 9B. Again, the optimizer does not transform the subquery's DTB into a DTB corresponding to the derived table DTh. The property vector of this access method is set with the property vector of the best access plan found for the subquery's DTB in the context of this access method. As for the access method qsub<DT>, compensation predicates are used to set the join methods in the main query block. Each join method used in conjunction with this access method has also associated with it some conditions related to where it can be placed during enumeration. Namely, the hash filter on P.p_partkey must be valid before this access method can be used. Hence, the table Part has to be already placed in any partial access plan that considers the plan segment (JH, qsub<DTh>) as illustrated in insert 950 at FIG. 9B.

[0242] C. Enumeration Methodology Including Cost-Based Subquery Optimization

[0243] FIGS. 10A-B comprise a single flowchart 1000 illustrating at a high level the enumeration process including the cost-based subquery optimization methodology of the present invention. The following description presents method steps that may be implemented using computer-executable instructions, for directing operation of a device under processor control. The computer-executable instructions may be stored on a computer-readable medium, such as CD, DVD, flash memory, or the like. The computer-executable instructions may also be stored as a set of downloadable computer-executable instructions, for example, for downloading and installation from an Internet location (e.g., Web server).

[0244] As shown, the enumeration process incorporating cost-based subquery optimization is substantially the same as the enumeration method previously illustrated at FIG. 8A-C. In order to incorporate cost-based subquery optimization, step 803 at FIG. 8A was modified as illustrated at step 1003 at FIG. 10A to consider subqueries as plan nodes when the Query Optimization Graph (QOG) is generated for each query block or Derived Table Block (DTB). The optimization process is as follows.

[0245] The method commences, at step 1001, with the receipt of a database query including a subquery. At step 1002, rewrite optimizations are applied and a semantically equivalent rewritten query is generated. At step 1003, a Query Optimization Graph (QOG) which includes subqueries as plan nodes is generated for each query block. The QOG contains now contains access methods for each qualifying subquery, costed by optimizing the subqueries' DTBs for each particular access method. The access methods are set during this pre-optimization phase by analyzing each subquery and the context in which it is used. Certain conditions must be met to execute, for example, the subquery as an uncorrelated derived table. The decision to use a specific access method is based on this analysis. After creating the compensation predicates for each access method, the join methods associated to a subquery's quantifier are also created. The alternative components of a subquery's DTB corresponding to a specific access method are built as well here. An example of the alternative component is the set of predicates used for the uncorrelated derived table execution strategy. The alternative set of predicates contains all the original predicates less the predicates with outer references (see Example 8 for an illustrative query).

[0246] The rest of the enumeration method is substantially the same as that previously described and illustrated at FIGS. 8A-C. Specifically, the method proceeds as follows. At step 1004, the process of examining each query block in order to determine the most favorable plan for each access method of the plan node ranging over such query block commences. For each subplan in the QOG, determining the best access plans starts with the most nested plans as provided at step 1005.

[0247] As previously described, the building of access plans is performed in a step-by-step manner. At step 1006 a set of plan node candidates to be placed in the next position of the current access plan (AP) under consideration is generated. Several steps are then performed for each plan node candidate (pn). At step 1007, a plan node candidate is selected from the generated set of candidates. At step 1008, a set of join method, access method (jm, am) candidates to be used with the current plan node pn is generated. Next at step 1009, for each candidate join method and access method (jm, am), the pn(jm, am) candidate plan segment is placed in the current access plan.

[0248] At step 1010, the current access plan being considered which may (or may not) contain all the plan nodes) is evaluated based on pipeline characteristics as well as execution costs computed based upon such factors as CPU cost and input/output (I/O) cost. Based upon this evaluation, the method proceeds to step 1011 if the current access plan under consideration is less favorable than the best plan previously found. Otherwise, if it is more favorable the method proceeds to step 1012.

[0249] If the current partial access plan under consideration is less favorable than the best plan previously found, then at step 1011 the last plan node is unplaced. In other words, the candidate plan segment that was added at step 1009 is pruned, as it is less favorable than another previously found. In this case steps 1009 or 1007 are repeated. Specifically, a new join method and access method pair for the current plan node is added to the current plan node pn at step 1009 (if candidate join methods and access methods (jm, am) are available for the current plan node pn). If all candidate join methods and access methods (jm, am) for a given plan node have been examined, then new plan nodes are examined at step 1007 (if additional candidate plan nodes are available). This process may continue until all plan nodes and their join methods and access methods have been examined at this position. As previously described, this process may also be interrupted if it is determined that continuing the search of a particular subspace will not improve the best plan previously found or if the effort spent in the current search space exceeds an established threshold.

[0250] At step 1012, if the current partial access plan under consideration is more favorable than the best plan previously found, a determination is made as to whether the current access plan represents a complete plan. If the current access plan is a complete plan, the plan is retained as the new best plan and the method goes to step 1011 to examine additional alternatives while candidates remain available. Otherwise, if the current access plan is not a complete plan, the method proceeds to step 1013. At step 1013, if the current access plan is not a complete plan, the method returns to step 1006 to examine one or more plan nodes for the next position(s) in current access plan.

[0251] This process continues until all portions of the current search subspace have been considered (or until the quota available for considering alternatives is exhausted as previously described). At step 1014, steps 1005-1013 are then repeated for additional query blocks until all query blocks have been optimized. At step 1015, after all query blocks have been optimized, a detailed access plan is constructed based upon the best plans found for each query block. The operations of the system of the present invention will now be described in greater detail.

[0252] D. Incorporating Cost-Based Subquery Optimization into Optimization Process

[0253] The process of optimization of a query block or DTB commences with a call to the following “OPTIMIZE DTB” procedure for the root DTB of a query. The cost-based subquery optimization is incorporated into DTB's optimization process by optimizing a subquery's DTB for each access method which was created to correspond to a possible subquery execution strategy.

1: Procedure: OPTIMIZE DTB
2: Purpose: Optimize a Derived Table Block (DTB).
3: Create the Data Flow Operator (dfo) tree representation for
the root DTB.
4: Input:
5: - dtb0 a parse tree representation of a DTB.
6: - q0 the quantifier ranging over dtb0
7: - am0 the access method of the quantifier q0 for which dtb0 is
optimized.
8: Output:
9: - A dfo execution tree if dtb0 is the root DTB.
10:
11: begin
12: if this is the first call to OPTIMIZE_DTB for dtb0 then
13:  Q  BuildQOG(dtb0)
14:  for each quantifier q ∈ Q do
15:   if q is a subquery quantifier then
16:    for each access method am ∈ AccessMethods(q) do
17:     dtb  DTB(q, am)
18:     call OPTIMIZE DTB(dtb, q, am)
19:    od
20:   else
21:    dtb  DTB(q)
22:    call OPTIMIZE DTB(dtb, q, <seq>)
23:  fi
24:  od
25: else
26:  Q  GetQOG(dtb0)
27: fi
28: call OPTIMIZE SUBPLAN(Subplan(Q),Q)
29: if q0 is not null then
30:  set the properties of the access plan of q0<am0> to be the
properties of APbest(Subplan(Q))[ ∅].
31:  return null
32: else
33:  dfo  BuildDFO(Q)
34:  return dfo
35: fi
36: end

[0254] As shown at line 13, a Query Optimization Graph (QOG) is built for each Derived Table Block (DTB) if this is the first call to OPTIMZE_DTB. Each nested DTB is optimized once for each access method needed for the quantifier ranging over that DTB (line 18). At line 17 the DTB corresponding to the quantifier q is prepared to be optimized for the access method am. This DTB is not transformed in the same fashion the transformation-based optimizers do. Instead, a set of alternative components, such as an alternative list of predicates and an alternative GROUP BY clause, are built for each access method am. For example, for the uncorrelated derived table the predicates with outer references are replaced with TRUE predicates (i.e., predicates with 100% selectivity and always true). For the uncorrelated derived table with hash filters, the predicates with outer references are replaced with the hash filter probing predicate. At line 17, the right alternatives are set for the DTB dtb to correspond to the subquery execution strategy employed by the access method am.

[0255] The subqueries' quantifiers have, in general, more than one access method, hence for each such access method its property vector is set by optimizing the underlying DTB once. The quantifiers corresponding to base tables for example, have only one access method, denoted by <seq>, hence their DTBs are optimized exactly once (line 22). Note that if a subquery qualifies to be executed only in its original form, the only the access method defined for this subquery's quantifier is qsubb<DFP>. Enumerating this type of subquery is hence equivalent to cost-based placing of an expensive predicate in an access plan. After each access method is set for each quantifier of the current DTB dtb0, the most outer subplan is optimized by a call to “OPTIMIZE SUBPLAN” at line 28. (The “OPTIMIZE SUBPLAN” method is described above in this document. The best access plan found by this call gives the properties of the access method q0<am0> which will be used to optimize the DTB that q0 belongs to. Note that to optimize the outside DTB that q0 belongs to, only the properties of the access method q0<am0> have to be known at the time of optimization. The details about the best access plan have to be known when the DFO is built if and only if the access method q0<am0> is used in the final best access plan of the query. The best access plan for an access method it is not saved as a detailed access plan. Instead enough information is saved for the best access plan such that it can be restored if needed. This skeleton of the best access plan is saved when the access method's properties are set (line 30). The process of restoring the best access plan is described above in this document.

[0256] While the invention is described in some detail with specific reference to a single-preferred embodiment and certain alternatives, there is no intent to limit the invention to that particular embodiment or those specific alternatives. For instance, those skilled in the art will appreciate that modifications may be made to the preferred embodiment without departing from the teachings of the present invention.

Claims

1. In a database system, a method for optimizing a database query, the method comprising: receiving a database query including at least one subquery; building a query optimization graph for each query block of the database query, the query optimization graph including plan nodes representing subqueries of each query block; generating a set of access methods and join methods for each plan node, including generating at least one access method for a subquery quantifier based on subquery type and semantic properties of the database query; determining an optimal access plan for each query block based upon selecting access methods, join methods, and join order for plan nodes of said query optimization graph having favorable execution costs; and constructing a detailed access plan for execution of the database query based upon the optimal access plan determined for each query block.

2. The method of claim 1, wherein said building step includes identifying each query block in the database query.

3. The method of claim 1, wherein said building step includes building a query optimization graph including plan nodes representing subplans and quantifiers of each query block.

4. The method of claim 1, wherein said generating step includes costing each access method and join method.

5. The method of claim 1, wherein said determining step includes starting with most nested plan nodes.

6. The method of claim 1, wherein said step of generating at least one access method for a subquery includes computing a property vector for said at least one access method.

7. The method of claim 6, wherein said step of computing a property vector includes computing a property vector of an access method by optimizing the subquery's Derived Table Block (DTB) specifically for that access method without transforming the DTB for a subquery execution strategy corresponding to that access method.

8. The method of claim 1, wherein said step of generating at least one access method includes generating an access method based on subquery execution strategies appropriate for subquery type and semantic properties of the database query.

9. The method of claim 1, wherein said step of generating at least one access method includes generating an uncorrelated derived table with hash filters access method for a qualifying subquery.

10. The method of claim 1, wherein said step of generating at least one access method includes generating an access method corresponding to evaluation of a subquery in the subquery's original form.

11. The method of claim 1, wherein said step of generating at least one access method includes generating an access method using an aggregate and then join subquery execution strategy for a qualifying subquery.

12. The method of claim 1, wherein said step of generating at least one access method includes generating an access method using a filter, aggregate and then join subquery execution strategy for a qualifying subquery.

13. The method of claim 1, wherein said determining step includes evaluating execution costs of a candidate plan segment to be added to an access plan being constructed.

14. The method of claim 13, wherein said candidate plan segment comprises a selected plan node together with an access method and a join method generated for the selected plan node.

15. The method of claim 14, wherein said determining step includes evaluating execution costs of partial access plans enabling earlier pruning of unfavorable access plans.

16. The method of claim 1, wherein said determining step includes substeps of: placing a candidate plan segment in a next position in an access plan being constructed, the candidate plan segment comprising a plan node, access method and join method valid at the next position; and evaluating the access plan including the candidate plan segment.

17. The method of claim 16, wherein said determining step further comprises substeps of: if the access plan is less favorable than an access plan previously identified, replacing the candidate plan segment with another available candidate plan segment and repeating said evaluating substep.

18. The method of claim 16, wherein said determining step further comprises substeps of: if the access plan is more favorable than an access plan previously identified but is not a complete plan, placing an additional candidate plan segment in a next position of the access plan; and repeating said evaluating substep.

19. The method of claim 16, wherein said determining step further comprises substeps of: if the access plan is more favorable than an access plan previously identified and is a complete plan, considering alternative plans by repeating the placing and evaluating substeps until all alternatives are considered.

20. The method of claim 16, wherein said determining step further comprises substeps of: if the access plan is more favorable than an access plan previously identified and is a complete plan, considering alternative access plans by repeating the placing and evaluating substeps until reaching a pre-selected limitation on consideration of alternative access plans.

21. A computer-readable medium having processor-executable instructions for performing the method of claim 1.

22. A downloadable set of processor-executable instructions for performing the method of claim 1.