16
Managing Hash Clusters

This chapter describes how to manage hash clusters, and includes the following topics:

See Also: Before attempting tasks described in this chapter, familiarize yourself with the concepts in Chapter 10, Guidelines for Managing Schema Objects.

Guidelines for Managing Hash Clusters

This section describes guidelines to consider before attempting to manage hash clusters, and includes the following topics:

Storing a table in a hash cluster is an optional way to improve the performance of data retrieval. A hash cluster provides an alternative to a non-clustered table with an index or an index cluster. With an indexed table or index cluster, Oracle locates the rows in a table using key values that Oracle stores in a separate index. To use hashing, you create a hash cluster and load tables into it. Oracle physically stores the rows of a table in a hash cluster and retrieves them according to the results of a hash function.

Oracle uses a hash function to generate a distribution of numeric values, called hash values, which are based on specific cluster key values. The key of a hash cluster, like the key of an index cluster, can be a single column or composite key (multiple column key). To find or store a row in a hash cluster, Oracle applies the hash function to the row's cluster key value; the resulting hash value corresponds to a data block in the cluster, which Oracle then reads or writes on behalf of the issued statement.

To find or store a row in an indexed table or cluster, a minimum of two (there are usually more) I/Os must be performed:

one or more I/Os to find or store the key value in the index
another I/O to read or write the row in the table or cluster

In contrast, Oracle uses a hash function to locate a row in a hash cluster; no I/O is required. As a result, a minimum of one I/O operation is necessary to read or write a row in a hash cluster.

Advantages of Hashing

If you opt to use indexing rather than hashing, consider whether to store a table individually or as part of a cluster.

Hashing is most advantageous when you have the following conditions:

Most queries are equality queries on the cluster key:

     SELECT . . . WHERE cluster_key = . . . ;



In such cases, the cluster key in the equality condition is hashed, and the corresponding hash key is usually found with a single read. In comparison, for an indexed table the key value must first be found in the index (usually several reads), and then the row is read from the table (another read).

The tables in the hash cluster are primarily static in size so that you can determine the number of rows and amount of space required for the tables in the cluster. If tables in a hash cluster require more space than the initial allocation for the cluster, performance degradation can be substantial because overflow blocks are required.

Disadvantages of Hashing

Hashing is not advantageous in the following situations:

Most queries on the table retrieve rows over a range of cluster key values. For example, in full table scans or queries like the following, a hash function cannot be used to determine the location of specific hash keys; instead, the equivalent of a full table scan must be done to fetch the rows for the query:

     SELECT . . . WHERE cluster_key < . . . ;




With an index, key values are ordered in the index, so cluster key values that satisfy the WHERE clause of a query can be found with relatively few I/Os.

The table is not static and continually growing. If a table grows without limit, the space required over the life of the table (its cluster) cannot be pre-determined.
Applications frequently perform full-table scans on the table and the table is sparsely populated. A full-table scan in this situation takes longer under hashing.
You cannot afford to pre-allocate the space that the hash cluster will eventually need.

See Also: For more information about creating hash clusters and specifying hash functions see the Oracle8 SQL Reference.

For information about hash functions and specifying user-defined hash functions, see Oracle8 Concepts.

Even if you decide to use hashing, a table can still have separate indexes on any columns, including the cluster key. See the Oracle8 Application Developer's Guide for additional recommendations.

Estimate Size Required by Hash Clusters and Set Storage Parameters

As with index clusters, it is important to estimate the storage required for the data in a hash cluster.

Oracle guarantees that the initial allocation of space is sufficient to store the hash table according to the settings SIZE and HASHKEYS. If settings for the storage parameters INITIAL, NEXT, and MINEXTENTS do not account for the hash table size, incremental (additional) extents are allocated until at least SIZE*HASHKEYS is reached. For example, assume that the data block size is 2K, the available data space per block is approximately 1900 bytes (data block size minus overhead), and that the STORAGE and HASH parameters are specified in the CREATE CLUSTER command as follows:

STORAGE (INITIAL 100K
    NEXT 150K
    MINEXTENTS 1
    PCTINCREASE 0)
SIZE 1500
HASHKEYS 100

In this example, only one hash key can be assigned per data block. Therefore, the initial space required for the hash cluster is at least 100*2K or 200K. The settings for the storage parameters do not account for this requirement. Therefore, an initial extent of 100K and a second extent of 150K are allocated to the hash cluster.

Alternatively, assume the HASH parameters are specified as follows:

SIZE 500 HASHKEYS 100

In this case, three hash keys are assigned to each data block. Therefore, the initial space required for the hash cluster is at least 34*2K or 68K. The initial settings for the storage parameters are sufficient for this requirement (an initial extent of 100K is allocated to the hash cluster).

See Also: To estimate the size of a hash cluster, use the procedure given in "Estimating Space Required by Clusters".

Creating Hash Clusters

After a hash cluster is created, tables can be created in the cluster. A hash cluster is created using the SQL command CREATE CLUSTER. For example, the following statement creates a cluster named TRIAL_CLUSTER that stores the TRIAL table, clustered by the TRIALNO column:

CREATE CLUSTER trial_cluster (trialno NUMBER(5,0))
    PCTUSED 80
    PCTFREE 5
    TABLESPACE users
    STORAGE (INITIAL 250K     NEXT 50K
      MINEXTENTS 1     MAXEXTENTS 3
      PCTINCREASE 0)
    HASH IS trialno HASHKEYS 150;

CREATE TABLE trial (
    trialno         NUMBER(5,0) PRIMARY KEY,
    ...)
    CLUSTER trial_cluster (trialno);

The following sections explain setting the parameters of the CREATE CLUSTER command specific to hash clusters.

See Also: For additional information about creating tables in a cluster, guidelines for setting other parameters of the CREATE CLUSTER command, and the privileges required to create a hash cluster, see "Creating Clusters".

Controlling Space Use Within a Hash Cluster

When creating a hash cluster, it is important to choose the cluster key correctly and set the HASH IS, SIZE, and HASHKEYS parameters so that performance and space use are optimal. The following guidelines describe how to set these parameters.

Choosing the Key

Choosing the correct cluster key is dependent on the most common types of queries issued against the clustered tables. For example, consider the EMP table in a hash cluster. If queries often select rows by employee number, the EMPNO column should be the cluster key; if queries often select rows by department number, the DEPTNO column should be the cluster key. For hash clusters that contain a single table, the cluster key is typically the entire primary key of the contained table.

The key of a hash cluster, like that of an index cluster, can be a single column or a composite key (multiple column key). A hash cluster with a composite key must use Oracle's internal hash function.

Setting HASH IS

Only specify the HASH IS parameter if the cluster key is a single column of the NUMBER datatype, and contains uniformly distributed integers. If the above conditions apply, you can distribute rows in the cluster so that each unique cluster key value hashes, with no collisions, to a unique hash value. If these conditions do not apply, omit this option so that you use the internal hash function.

Setting SIZE

SIZE should be set to the average amount of space required to hold all rows for any given hash key. Therefore, to properly determine SIZE, you must be aware of the characteristics of your data:

If the hash cluster is to contain only a single table and the hash key values of the rows in that table are unique (one row per value), SIZE can be set to the average row size in the cluster.
If the hash cluster is to contain multiple tables, SIZE can be set to the average amount of space required to hold all rows associated with a representative hash value.

See Also: To estimate a preliminary value for SIZE, follow the procedures given in Appendix A, "Space Estimations for Schema Objects". If the preliminary value for SIZE is small (more than four hash keys can be assigned per data block), you can use this value for SIZE in the CREATE CLUSTER command.

However, if the value of SIZE is large (fewer than five hash keys can be assigned per data block), you should also consider the expected frequency of collisions and whether performance of data retrieval or efficiency of space usage is more important to you:

If the hash cluster does not use the internal hash function (if you specified HASH IS) and you expect little or no collisions, you can set SIZE as estimated; no collisions occur and space is used as efficiently as possible.

If you expect frequent collisions on inserts, the likelihood of overflow blocks being allocated to store rows is high. To reduce the possibility of overflow blocks and maximize performance when collisions are frequent, you should increase SIZE according to Table 16-1.

Table 16-1 SIZE Increase Chart

Available Space per Block/Calculated SIZE	Setting for SIZE
1	Calculated SIZE
2	Calculated SIZE + 15%
3	Calculated SIZE + 12%
4	Calculated SIZE + 8%
>4	Calculated SIZE

Overestimating the value of SIZE increases the amount of unused space in the cluster. If space efficiency is more important than the performance of data retrieval, disregard the above adjustments and use the estimated value for SIZE.

Setting HASHKEYS

For maximum distribution of rows in a hash cluster, HASHKEYS should always be a prime number.

For example, suppose you cluster the EMP table by DEPTNO, and there are 100 DEPTNOs, with values 10, 20, . . ., 1000. Assuming you bypass the internal hash function and you create a cluster with HASHKEYS of 100, then department 10 will hash to 10, department 20 to 20, . . ., department 110 to 10 (110 mod 100), department 120 to 20, and so on. Notice that there are 10 entries for hash values of 10, 20 . . ., but none for 1, 2, . . ., and so on. As a result, there is a lot of wasted space and possibly a lot of overflow blocks because of collisions. Alternatively, if HASHKEYS is set to 101, then each department number hashes to a unique hash key value.

Controlling Space in Hash Clusters: Examples

The following examples show how to correctly choose the cluster key and set the HASH IS, SIZE, and HASHKEYS parameters. For all examples, assume that the data block size is 2K and that on average, 1950 bytes of each block is available data space (block size minus overhead).

Example 1	You decide to load the EMP table into a hash cluster. Most queries retrieve employee records by their employee number. You estimate that the maximum number of rows in the EMP table at any given time is 10000 and that the average row size is 55 bytes.
	In this case, EMPNO should be the cluster key. Since this column contains integers that are unique, the internal hash function can be bypassed. SIZE can be set to the average row size, 55 bytes; note that 34 hash keys are assigned per data block. HASHKEYS can be set to the number of rows in the table, 10000, rounded up to the next highest prime number, 10001:
	`CREATE CLUSTER emp_cluster (empnoNUMBER). . .SIZE 55HASH IS empno HASHKEYS 10001;`
Example 2	Conditions similar to the previous example exist. In this case, however, rows are usually retrieved by department number. At most, there are 1000 departments with an average of 10 employees per department. Note that department numbers increment by 10 (0, 10, 20, 30, . . . ).
	In this case, DEPTNO should be the cluster key. Since this column contains integers that are uniformly distributed, the internal hash function can be bypassed. A pre-estimated SIZE (the average amount of space required to hold all rows per department) is 55 bytes * 10, or 550 bytes. Using this value for SIZE, only three hash keys can be assigned per data block. If you expect some collisions and want maximum performance of data retrieval, slightly alter your estimated SIZE to prevent collisions from requiring overflow blocks. By adjusting SIZE by 12%, to 620 bytes (see previous section about setting SIZE for clarification), only three hash keys are assigned per data block, leaving more space for rows from expected collisions.
	HASHKEYS can be set to the number of unique department numbers, 1000, rounded up to the next highest prime number, 1009:
	`CREATE CLUSTER emp_cluster (deptno NUMBER). . .SIZE 620HASH IS deptno HASHKEYS 1009;`

Altering Hash Clusters

You can alter a hash cluster with the SQL command ALTER CLUSTER:

ALTER CLUSTER emp_dept . . . ;

The implications for altering a hash cluster are identical to those for altering an index cluster. However, note that the SIZE, HASHKEYS, and HASH IS parameters cannot be specified in an ALTER CLUSTER statement. You must re-create the cluster to change these parameters and then copy the data from the original cluster.

See Also: For more information about altering an index cluster, see "Altering Clusters".

Dropping Hash Clusters

You can drop a hash cluster using the SQL command DROP CLUSTER:

DROP CLUSTER emp_dept;

A table in a hash cluster is dropped using the SQL command DROP TABLE. The implications of dropping hash clusters and tables in hash clusters are the same for index clusters.

See Also: For more information about dropping clusters, see "Dropping Clusters".

16 Managing Hash Clusters