The present disclosure concerns a method of estimating cardinality of a join of tables using multi-column density values, and more particularly, the present disclosure relates to a method of estimating cardinality of a join of tables using multi-column density values and additionally using coarser density values of a subset of the multi-column density attributes.
Relational databases systems allow a database user to enter queries into the database and return the data that meets the conditions of the query. The data present within the database system is in one or more tables or relations. Each relation consists of a number of records or tuples containing specific information possibly grouped in some ordered sequence. Each tuple consists of one or more fields called attributes. In any single attribute of a tuple there can be only a single value. However, each tuple can have a different value for the same attribute.
Some characteristics of a database relation are typically maintained. For example, the database system may maintain the cardinality, the density and the number of distinct values of each relation. Cardinality is the number of tuples or records in a relation. The number of distinct values is the number of distinct values of a given attribute or set of attributes. The density is the average number of tuples per distinct value.
One operation performed by a database system is known as a join operation. A join operation is used to combine related tuples from two relations into single tuples. Typically, the join operation matches tuples from two relations on the values taken from each relation on a common attribute and creates a joined table or relation. If another relation needs to be joined with the result relation, the result relation may be referred to as an intermediate relation because it is created in the process of generating a result relation.
Another operation performed by a database system is the select operation. The select operation is used to select a subset of tuples from a relation that satisfy a selection condition. One can consider the select operation to be a filter that keeps only those tuples that satisfy a qualifying condition.
A query entered into a relational database system may result in multiple operations being performed. For example, selection operations on one or more relations can be used together with multiple join operations in the same query. In many cases, the operations of the query can be performed in several different orders without changing the result of the query. Each possible order of operations is referred to as a query execution plan. There may be several alternative query execution plans, each specifying a set of operations to be executed by the database system. Each different query execution plan will have a different “cost.” The costs may include the cost of accessing secondary storage, computation cost, memory usage cost, and communication cost.
Relational database systems typically include a component called a query optimizer. The query optimizer may identify several query execution plans, estimate the cost of each different query execution plan, and select the plan with the lowest estimated cost for execution. Query plans generated for a query will differ in their total cost of obtaining the desired data. The query optimizer evaluates these cost estimates for each query plan in order to determine which plan is likely to have the lowest execution cost.
The join operation can be quite expensive, since joining together two or more entire relations can result in a very large relation. When multiple joins are present in a query, the cost of a bad execution plan may increase dramatically. It is important for the query optimizer to identify a query execution plan that minimizes cost. The join ordering chosen by the query optimizer is often a key factor in the ultimate cost of the query execution plan.
In view of these considerations, query optimizers often use estimates of the cardinality of joins in attempting to select the most efficient query plan. It may be desirable to estimate the cardinality of the join of a table R with a table S where attribute a of table R is equal to attribute b of table S, denoted RR.a=S.bS. One prior art method of estimating the cardinality of such a join relied on what is referred to as the “containment assumption.” According to the containment assumption, each group of distinct valued tuples belonging to the relation with the minimal number of distinct values joins with some group of distinct valued tuples in the other table. The containment assumption has been extended to estimate the cardinality of a multi-predicate join. One problem with an existing method of estimating the cardinality of a multi-predicate using the containment assumption join is that cardinality is often overestimated.
There is a need for a method of estimating cardinality using multi-column density values and additionally using coarser density values of a subset of the multi-column density attributes to reduce the overestimation problem.
The present disclosure concerns a method of estimating cardinality of a join of tables using multi-column density values and additionally using coarser density values of a subset of the multi-column density attributes. In one embodiment, the sequence of attributes for the coarser densities is a prefix of the sequence of multi-column density attributes.
In one embodiment of the method, a multi-column density over a set of attributes is identified for each table. A density of a subset of the set of attributes is also identified for each table. In this embodiment, a number of tuples from each table that participate in the join is estimated using densities of the subsets. The cardinality of the join is estimated using the multi-column density for each table and the estimated number of tuples that participate in the join from each table.
In one embodiment of the method, a multi-column density over a set of attributes and a cardinality is identified for each table. A density of a subset of the set of attributes is also identified for each table. The cardinality of the join is estimated using the multi-column density for each table, the density of said subset of said set of attributes for each table, and the cardinality of each table.
In one embodiment, the method improves an existing method of estimating cardinality of a join of tables using a multi-column density over a set of attributes. The method improves the existing method by additionally using a density of a subset of the set of multi-column density attributes for each table to estimate the cardinality of the join.
The number of tuples from each table that participate in the join may be estimated by multiplying an estimated minimum number of distinct values for the subset of attributes and a density of the subset of attributes in the table. The number of distinct values of a subset of attributes can be calculated by dividing the cardinality of the table by the density of the subset of attributes in the table.
In one embodiment, the join cardinality estimation is calculated by selecting the minimum of an estimated number of tuples from a first table that participate in the join divided by a multi-column density over a set of attributes for the first table and an estimated number of tuples from a second table that participate in the join divided by a multi-column density over a set of attributes for the second table. This minimum, the multi-column density for the first table, and the multi-column density for the second table are multiplied to calculate the estimated cardinality of the join.
In one embodiment, the cardinality estimation is calculated by identifying a first minimum value of a cardinality of a first table divided by a density of a subset of the set of multi-column density attributes for the first table and a cardinality of a second table divided by a density of a subset of the set multi-column density attributes for the second table. The method also identifies a second minimum value of the density of the subset of the set of multi-column density attributes for the first table divided by a multi-column density over the set of multi-column density attributes for the first table and the density of the subset of the set of multi-column density attributes for the second table divided by a multi-column density over the set of attributes for the second table. The first minimum value, the second minimum value, the multi-column density over the set of attributes for the first table, and the multi-column density over the set of attributes for the second table are multiplied to calculate the estimated cardinality of the join.
These and other objects, advantages, and features of an exemplary embodiment are described in conjunction with the accompanying drawings.
Exemplary Operating Environment
With reference to
A number of program modules may be stored on the hard disk 27, magnetic disk 29, optical disk 31, ROM 24 or RAM 25, including an operating system 35, one or more application programs 36, other program modules 37, and program data 38. A database system 55 may also be stored on the hard disk, magnetic disk 29, optical disk 31, ROM 24 or RAM 25. A user may enter commands and information into personal computer 20 through input devices such as a keyboard 40 and pointing device 42. Other input devices may include a microphone, joystick, game pad, satellite dish, scanner, or the like. These and other input devices are often connected to processing unit 21 through a serial port interface 46 that is coupled to system bus 23, but may be connected by other interfaces, such as a parallel port, game port or a universal serial bus (USB). A monitor 47 or other type of display device is also connected to system bus 23 via an interface, such as a video adapter 48. In addition to the monitor, personal computers typically include other peripheral output devices such as speakers and printers.
Personal computer 20 may operate in a networked environment using logical connections to one or more remote computers, such as a remote computer 49. Remote computer 49 may be another personal computer, a server, a router, a network PC, a peer device or other common network node, and typically includes many or all of the elements described above relative to personal computer 20, although only a memory storage device 50 has been illustrated in
When using a LAN networking environment, personal computer 20 is connected to local network 51 through a network interface or adapter 53. When used in a WAN networking environment, personal computer 20 typically includes a modem 54 or other means for establishing communication over wide area network 52, such as the Internet. Modem 54, which may be internal or external, is connected to system bus 23 via serial port interface 46. In a networked environment, program modules depicted relative to personal computer 20, or portions thereof, may be stored in remote memory storage, device 50. It will be appreciated that the network connections shown are exemplary and other means of establishing a communications link between the computers may be used.
Notation
In this disclosure, tables are represented by uppercase letters. For example, tables may be represented by uppercase letters R; S; T . . . Table attributes are represented by lowercase letters. For example, table attributes may be represented by lowercase letters a; b; c . . . The cardinality of a table R is denoted as cardR or |R|. For a given attribute a of table R, the density of attribute a in table R, denoted δR.a, is the average number of duplicates per distinct value of attribute a in R. The number of distinct values of attribute a in table R is denoted dvR.a. As a result, dvR.a=cardR/δR.a.
A multi-column density MCD for attributes a1, . . . ,an of table R is denoted as δR.a1, . . . ,an. The multi-column density δR.a1, . . . ,an is the average number of duplicates per distinct combination of attributes a1, . . . ,an. In this disclosure, the number of distinct combinations of a1, . . . ,an in table R are denoted as dvR.a1, . . . ,an. As such, the number of distinct attribute combinations dvR.a1, . . . ,an=cardR/δR.a1, . . . ,an. It is assumed in this disclosure that whenever multi-column densities are available over attributes a1, . . . ,an the densities of prefix subsets of the original multi-column density are also available. For the multi-column density of a1, . . . ,an the densities of the prefixes: (a1), (a1, a2); (a1, a2, a3), . . . ,(a1,a2, . . . ,an−1) are also available.
Join Estimation Using Containment Assumptions
The cardinality of the join of table R with table S where attribute a of table R is equal to attribute b of table S is denoted RR.a=S.bS. According to the containment assumption, each group of distinct valued tuples belonging to the relation with the minimal number of distinct values joins with some group of distinct valued tuples in the other table. This is illustrated by the example of
Using the containment assumption, an estimated cardinality of the join is calculated as:
Applying Equation 1 to the example of
Considering the multi-predicate join of table R with table S where attribute a of table R is equal to attribute b of table S and attribute c of table R is equal to attribute d of table S, denoted RR.a=S.b
R.c=S.dS. The estimation calculation of Equation 1 can be extended to this multi-predicate join to define the estimated cardinality of the multi-predicate join as:
This estimation is correct under the containment assumption in absence of additional information, such as partial densities. If the densities of the prefixes of the multi-column densities (e.g., δR.a,δR.b) are also available, the estimation of Equation 2 could be an overestimate of the cardinality. The example illustrated by
R.a=S.b
R.c=S.dS. Suppose cardR=1000, cardS=1000 and that δR.a,c=10 and δR.b,d=1. Referring to
Using Coarser Densities to Reduce Overestimation
Using Equation 2 in the Example of
In one embodiment, the multicolumn density δR.a,c, the multi-column density δR.b,d, and the density information δR.a,δR.b, about a prefix subset of the multi-column density attributes is also available. This density information can be used to estimate the cardinality of RR.a=S.b
R.c=S.dS by using a two-step approach. This two-step approach is illustrated by
R.a=S.bS using the containment assumption and the coarser subset density information δR.a and δS.b. The estimated number of tuples in table R that participate in the simpler join R
R.a=S.bS is:
Analogously, the number of tuples in S that participate in this simpler join is:
In a second step, the cardinality of the original join RR.a=S.b
R.c=S.dS is estimated 132 using the estimated number tuples from R and S that participate in the simpler join R
R.a=S.bS (using the Equations 3 and 4) instead of the cardinality of R cardR and the cardinality of S cardS. The observation is that any tuple from table R or table S that does not participate in the simpler join will not participate in the original join, since the latter contains an additional equality predicate. Therefore, in the exemplary embodiment the method replaces the actual cardinalities cardR and cardS in estimation Equation 2 with the estimated number of tuples from table R and table S that participate in the simpler join defined by Equations 3 and 4 respectively. In this way, the method removes from table R and table S tuples that are do not to participate in the join when the containment assumption is applied. This assumes that the values of densities δR.a,c and δS.b,d are unchanged after removing the tuples that do not participate in the join. The resulting estimation equation is:
By extracting the common factor
Equation 5 reduces to:
If the densities of the other prefixes are available from the construction of the original multi-column densities i.e., densities δR.c, δR.c,a, δS.d, δS.d,b using the arguments applied above, the cardinality RR.a=S.b
R.c=S.dS can be estimated by Equation 7.
The two estimations returned using Equations 6 and 7 can be different. In such a case, since all these estimations are estimated upper bounds for the join, the exemplary method takes the minimal one estimate.
In one embodiment, Equation 7 is generalized for any number n (greater than 2) of equalities in the join predicate, by recursively applying the method described above. In general, for a join with a number n predicates there can be n! (n factorial) different permutations of multi-column densities and subsets of prefixes. If all of the densities are available, the method selects the minimum estimation of the cardinality in one embodiment. However, in one embodiment only a selected number of the possible estimations are calculated. This may be done either to be efficient or because not all combinations of the multi-column densities are available. Regardless of how many of the possible estimations are calculated using the disclosed method, the disclosed estimation method guarantees that the minimal value among the estimations is at least as accurate as the estimation obtained if the single multi-column densities were just used by themselves in Equation 2.
Situations When Using Coarser Subset Densities Reduces Overestimation
Equation 6 has the following basic form:
min(a1/b1, a2/b2)*min(b1/c1, b2/c2)*c1*c2. where:
Analogously, if both min operators of Equation 6 return the second parameter, then the min operator of Equation 2 returns the second parameter, and the result of Equations 2 and 6 is the same. That is, Equation 6 reduces to Equation 2, since
However, if the first min operator of Equation 6 returns the first parameter, and the second min operator of Equation 6 returns the second parameter, both equations might return different values. That is, Equation 6 does not reduce to Equation 2, since, in general,
is different from either
Similarly, if the first min operator of Equation 6 returns the second parameter, and the second min operator of Equation 6 returns the first parameter, both equations might return different values. Again, Equation 6 doe not reduce to Equation 2, since
is in general different from either
In such cases, Equation 6 provides a tighter upper bound than Equation 2 and is thus more accurate. The following example query illustrates how this situation arises:
SELECT*FROM part, supplier, lineitem
WHERE
OPTION (force order)
In the example, the database TPC-H is used. The following values are obtained from the TPC-H database where S=supplier, P=part, and L=lineitem.
cardS=1000, δS.suppkey=1
cardP=20000, δP.partkey=1
cardL=600000, δL.partkey,suppkey=7.5, δL.partkey=30, δL.suppkey=600
The filter σpname like ‘green’ returns the following information where σgreenP is denoted σP:
cardσP=200, δσP.partkey=1
After applying the Cartesian product, it can be estimated (denoting “S×σP” as x) that
cardx=1000*200=200000
δx.partkey,suppkey=1 (it is the product of the original densities).
δx.partkey=1000 (since each distinct partkey value was combined with all 1000 suppliers).
δx.suppkey=200 (since each distinct suppkey value was combined with all 200 green parts).
All the information in the above example can be derived from the Cartesian product alone, without considering key information. This information can be used to estimate the size of the join Lpartkey=partkey
suppkey=suppkeyx. If the multi-column densities δL.partkey,suppkey and δx.partkey,suppkey are used with Equation 2, the estimated cardinality is calculated as follows:
By applying Equation 6 it will be shown that 600,000 is an overestimate of the cardinality. The estimated cardinality using equation 6 multi-column densities δL.partkey,suppkey and δx.partkey,suppkey, and the additional prefix density information δL.partkey and δx.partkey is calculated as follows:
Note that the results of the first min operation is the second parameter, while the result for the second min operation is the first parameter. Had the other possible multi-column densities, i.e., δL.suppkey,partkey and δx.suppkey,partkey, and the additional prefix density information δL.suppkey and δx.suppkey, the cardinality is calculated as follows:
In the exemplary embodiment, the method takes the minimum over these two refined estimations (min(600000, 6000)=6000) to obtain the estimated cardinality estimation. The estimated cardinality returned in this example is an accurate estimation (the actual value obtained by running the query on TPC-H is 6629).
Extending the Method to Histograms
In one embodiment, the disclosed method of estimating cardinality using multi-column densities can be extended to histograms. Referring to
Referring to
While the present invention has been described with a degree of particularity, it is the intent that the invention include all modifications and alterations falling within the spirit or scope of the appended claims.
Number | Name | Date | Kind |
---|---|---|---|
6865567 | Oommen et al. | Mar 2005 | B1 |
Number | Date | Country | |
---|---|---|---|
20040260675 A1 | Dec 2004 | US |