He was not merely a chip of the old block, but the old block itself.
Edmund Burke: On Pitt's first speech
This chapter describes the nature of and relationships among the logical storage structures in the Oracle server. It includes:
Oracle allocates logical database space for all data in a database. The units of database space allocation are data blocks, extents, and segments. The following illustration shows the relationships among these data structures:
At the finest level of granularity, Oracle stores data in data blocks (also called logical blocks, Oracle blocks, or pages). One data block corresponds to a specific number of bytes of physical database space on disk.
The next level of logical database space is called an extent. An extent is a specific number of contiguous data blocks allocated for storing a specific type of information.
The level of logical database storage above an extent is called a segment. A segment is a set of extents that have been allocated for a specific type of data structure. For example, each table's data is stored in its own data segment, while each index's data is stored in its own index segment. If the table or index is partitioned, each partition is stored in its own segment.
Oracle allocates space for segments in units of one extent. When the existing extents of a segment are full, Oracle allocates another extent for that segment. Because extents are allocated as needed, the extents of a segment may or may not be contiguous on disk.
A segment (and all its extents) ar stored in one tablespace. Within a tablespace, a segment can span datafiles (have extents with data from more than one file). However, each extent can contain data from only one datafile.
Oracle manages the storage space in the datafiles of a database in units called data blocks. A data block is the smallest unit of I/O used by a database. In contrast, at the physical, operating system level, all data is stored in bytes. Each operating system has what is called a block size. Oracle requests data in multiples of Oracle data blocks, not operating system blocks.
You set the data block size for each Oracle database when you create the database. This data block size should be a multiple of the operating system's block size within the maximum (port-specific) limit to avoid unnecessary I/O. Oracle data blocks are the smallest units of storage that Oracle can use or allocate.
The Oracle data block format is similar regardless of whether the data block contains table, index, or clustered data. Figure 2-2 illustrates the format of a data block.
The header contains general block information, such as the block address and the type of segment (for example, data, index, or rollback).
This portion of the data block contains information about the tables having rows in this block.
This portion of the data block contains information about the actual rows in the block (including addresses for each row piece in the row data area).
Once the space has been allocated in the row directory of a data block's overhead, this space is not reclaimed when the row is deleted. Therefore, a block that is currently empty but had up to 50 rows at one time continues to have 100 bytes allocated in the header for the row directory. Oracle reuses this space only when new rows are inserted in the block.
The data block header, table directory, and row directory are referred to collectively as overhead. Some block overhead is fixed in size; the total block overhead size is variable. On average, the fixed and variable portions of data block overhead total 84 to 107 bytes.
This portion of the data block contains table or index data. Rows can span blocks; see "Row Chaining and Migrating" on page 2-10.
Free space is allocated for insertion of new rows and for updates to rows that require additional space (for example, when a trailing null is updated to a non-null value). Whether issued insertions actually occur in a given data block is a function of current free space in that data block and the value of the space management parameter PCTFREE. The next section, "An Introduction to PCTFREE, PCTUSED, and Row Chaining", contains more information on space management parameters.
In data blocks allocated for the data segment of a table or cluster, or for the index segment of an index, free space can also hold transaction entries. A transaction entry is required in a block for each INSERT, UPDATE, DELETE, and SELECT...FOR UPDATE statement accessing one or more rows in the block. The space required for transaction entries is operating system dependent; however, transaction entries in most operating systems require approximately 23 bytes.
Two space management parameters, PCTFREE and PCTUSED, enable you to control the use of free space for inserts of and updates to the rows in all the data blocks of a particular segment. You specify these parameters when creating or altering a table or cluster (which has its own data segment). You can also specify the storage parameter PCTFREE when creating or altering an index (which has its own index segment).
This discussion does not apply to LOB datatypes (BLOB, CLOB, NCLOB, and BFILE) - they do not use the PCTFREE storage parameter or free lists. See "LOB Datatypes" on page 10-9 for more information.
The PCTFREE parameter sets the minimum percentage of a data block to be reserved as free space for possible updates to rows that already exist in that block. For example, assume that you specify the following parameter within a CREATE TABLE statement:
This states that 20% of each data block in this table's data segment will be kept free and available for possible updates to the existing rows already within each block. New rows can be added to the row data area, and corresponding information can be added to the variable portions of the overhead area, until the row data and overhead total 80% of the total block size. Figure 2-3 illustrates PCTFREE.
The PCTUSED parameter sets the minimum percentage of a block that can be used for row data plus overhead before new rows will be added to the block. After a data block is filled to the limit determined by PCTFREE, Oracle considers the block unavailable for the insertion of new rows until the percentage of that block falls below the parameter PCTUSED. Until this value is achieved, Oracle uses the free space of the data block only for updates to rows already contained in the data block. For example, assume that you specify the following parameter in a CREATE TABLE statement:
In this case, a data block used for this table's data segment is considered unavailable for the insertion of any new rows until the amount of used space in the block falls to 39% or less (assuming that the block's used space has previously reached PCTFREE). Figure 2-4 illustrates this.
PCTFREE and PCTUSED work together to optimize the utilization of space in the data blocks of the extents within a data segment. Figure 2-5 illustrates the interaction of these two parameters.
In a newly allocated data block, the space available for inserts is the block size minus the sum of the block overhead and free space (PCTFREE). Updates to existing data can use any available space in the block; therefore, updates can reduce the available space of a block to less than PCTFREE, the space reserved for updates but not accessible to inserts.
For each data and index segment, Oracle maintains one or more free lists - lists of data blocks that have been allocated for that segment's extents and have free space greater than PCTFREE; these blocks are available for inserts. When you issue an INSERT statement, Oracle checks a free list of the table for the first available data block and uses it if possible. If the free space in that block is not large enough to accommodate the INSERT statement, and the block is at least PCTUSED, Oracle takes the block off the free list. Multiple free lists per segment can reduce contention for free lists when concurrent inserts take place.
After you issue a DELETE or UPDATE statement, Oracle processes the statement and checks to see if the space being used in the block is now less than PCTUSED. If it is, the block goes to the beginning of the transaction free list, and it is the first of the available blocks to be used in that transaction. When the transaction commits, free space in the block becomes available for other transactions.
Two types of statements can increase the free space of one or more data blocks: DELETE statements, and UPDATE statements that update existing values to smaller values. The released space from these types of statements is available for subsequent INSERT statements under the following conditions:
Released space may or may not be contiguous with the main area of free space in a data block. Oracle coalesces the free space of a data block only when (1) an INSERT or UPDATE statement attempts to use a block that contains enough free space to contain a new row piece, and (2) the free space is fragmented so that the row piece cannot be inserted in a contiguous section of the block. Oracle does this compression only in such situations, because otherwise the performance of a database system would decrease due to the continuous compression of the free space in data blocks.
In two circumstances, the data for a row in a table may be too large to fit into a single data block. In the first case, the row is too large to fit into one data block when it is first inserted. In this case, Oracle stores the data for the row in a chain of data blocks (one or more) reserved for that segment. Row chaining most often occurs with large rows, such as rows that contain a column of datatype LONG or LONG RAW. Row chaining in these cases is unavoidable.
The format of a row and a row piece are described in "Row Format and Size" on page 8-5.
However, in the second case, a row that originally fit into one data block is updated so that the overall row length increases, and the block's free space is already completely filled. In this case, Oracle migrates the data for the entire row to a new data block, assuming the entire row can fit in a new block. Oracle preserves the original row piece of a migrated row to point to the new block containing the migrated row; the ROWID of a migrated row does not change.
For more information about ROWID, see "ROWID Datatype" on page 10-12.
When a row is chained or migrated, I/O performance associated with this row decreases because Oracle must scan more than one data block to retrieve the information for the row.
See Oracle8 Tuning for information about reducing chained and migrated rows and improving I/O performance.
An extent is a logical unit of database storage space allocation made up of a number of contiguous data blocks. One or more extents in turn make up a segment. When the existing space in a segment is completely used, Oracle allocates a new extent for the segment.
When you create a table, Oracle allocates to the table's data segment an initial extent of a specified number of data blocks. Although no rows have been inserted yet, the Oracle data blocks that correspond to the initial extent are reserved for that table's rows.
If the data blocks of a segment's initial extent become full and more space is required to hold new data, Oracle automatically allocates an incremental extent for that segment. An incremental extent is a subsequent extent of the same or greater size than the previously allocated extent in that segment. (The next section explains the factors controlling the size of incremental extents.)
For maintenance purposes, the header block of each segment contains a directory of the extents in that segment.
Rollback segments always have at least two extents. For more information, see "How Extents Are Used and Allocated for Rollback Segments" on page 2-19.
This chapter applies to serial operations, in which one server process parses and executes a SQL statement. Extents are allocated somewhat differently in parallel SQL statements, which entail multiple server processes. See "Free Space and Parallel DDL" on page 22-27 for more information.
Storage parameters expressed in terms of extents define every segment. Storage parameters apply to all types of segments. They control how Oracle allocates free database space for a given segment. For example, you can determine how much space is initially reserved for a table's data segment or you can limit the number of extents the table can allocate by specifying the storage parameters of a table in the STORAGE clause of the CREATE TABLE statement.
Oracle controls the allocation of incremental extents for a given segment as follows:
In the current example, if Oracle does not find a set of exactly 20 contiguous data blocks, Oracle searches for a set of contiguous data blocks greater than 20. If the first set it finds contains 25 or more blocks, it breaks the blocks up and allocates 20 of them to the new extent and leaves the remaining 5 or more blocks as free space. Otherwise, it allocates all of the blocks (between 21 and 24) to the new extent.
The blocks of a newly allocated extent, although they were free, may not be empty of old data. Usually, Oracle formats the blocks of a newly allocated extent when it starts using the extent, but only as needed (starting with the blocks on the segment free list). In a few cases, however, such as when a database administrator forces allocation of an incremental extent with the ALLOCATE EXTENT option of an ALTER TABLE or ALTER CLUSTER statement, Oracle formats the extent's blocks when it allocates the extent.
In general, the extents of a segment do not return to the tablespace until you drop the object whose data is stored in the segment (using a DROP TABLE or DROP CLUSTER statement). Exceptions to this include the following:
When extents are freed, Oracle updates the data dictionary to reflect the regained extents as available space. Any data in the blocks of freed extents becomes inaccessible, and Oracle clears the data when the blocks are subsequently reused for other extents.
As long as a nonclustered table exists or until you truncate the table, any data block allocated to its data segment remains allocated for the table. Oracle inserts new rows into a block if there is enough room. Even if you delete all rows of a table, Oracle does not reclaim the data blocks for use by other objects in the tablespace.
After you drop a nonclustered table, this space can be reclaimed when other extents require free space.
Oracle reclaims all the extents of its data and index segments for the tablespaces that they were in and makes the extents available for other objects in the tablespace. Subsequently, when other segments require large extents, Oracle identifies and combines contiguous reclaimed extents to form the requested larger extents.
Clustered tables store their information in the data segment created for the cluster. Therefore, if you drop one table in a cluster, the data segment remains for the other tables in the cluster, and no extents are deallocated. You can also truncate clusters (except for hash clusters) to free extents.
Oracle deallocates the extents of snapshots and snapshot logs in the same manner as for nonclustered and clustered tables.
See Oracle8 Replication for more information on snapshots and snapshot logs.
All extents allocated to an index segment remain allocated as long as the index exists. When you drop the index or associated table or cluster, Oracle reclaims the extents for other uses within the tablespace.
Oracle periodically checks to see if the rollback segments of the database have grown larger than their optimal size. If a rollback segment is larger than is optimal (that is, it has too many extents), Oracle automatically deallocates one or more extents from the rollback segment. See "How Extents Are Deallocated from a Rollback Segment" on page 2-22 for more information.
When Oracle completes the execution of a statement requiring a temporary segment, Oracle automatically drops the temporary segment and returns the extents allocated for that segment to the associated tablespace. A single sort allocates its own temporary segment, in the temporary tablespace of the user issuing the statement, and then returns the extents to the tablespace.
Multiple sorts, however, can use sort segments in a temporary tablespace designated exclusively for sorts. These sort segments are allocated only once for the instance, and they are not returned after the sort but remain available for other multiple sorts. For more information, see "Temporary Segments" on page 2-16.
A segment is a set of extents that contains all the data for a specific logical storage structure within a tablespace. For example, for each table, Oracle allocates one or more extents to form that table's data segment; for each index, Oracle allocates one or more extents to form its index segment.
Oracle databases use four types of segments:
The following sections discuss each type of segment.
Every nonclustered table or partition and every cluster in an Oracle database has a single data segment to hold all of its data. Oracle creates this data segment when you create the nonclustered table or cluster with the CREATE command.
The storage parameters for a nonclustered table or cluster determine how its data segment's extents are allocated. You can set these storage parameters directly with the appropriate CREATE or ALTER command. These storage parameters affect the efficiency of data retrieval and storage for the data segment associated with the object.
Every index in an Oracle database has a single index segment to hold all of its data. Oracle creates the index segment for the index when you issue the CREATE INDEX command. In this command, you can specify storage parameters for the extents of the index segment and a tablespace in which to create the index segment. (The segments of a table and an index associated with it do not have to occupy the same tablespace.) Setting the storage parameters directly affects the efficiency of data retrieval and storage.
When processing queries, Oracle often requires temporary workspace for intermediate stages of SQL statement parsing and execution. Oracle automatically allocates this disk space called a temporary segment. Typically, Oracle requires a temporary segment as a work area for sorting. Oracle does not create a segment if the sorting operation can be done in memory or if Oracle finds some other way to perform the operation using indexes.
The following commands may require the use of a temporary segment:
Some unindexed joins and correlated subqueries may also require use of a temporary segment. For example, if a query contains a DISTINCT clause, a GROUP BY, and an ORDER BY, Oracle can require as many as two temporary segments. If applications often issue commands in the list above, the database administrator may want to improve performance by adjusting the initialization parameter SORT_AREA_SIZE.
See the Oracle8 Reference for information on SORT_AREA_SIZE and other initialization parameters.
Oracle allocates temporary segments as needed during a user session, in the temporary tablespace of the user issuing the statement. You specify this tablespace with a CREATE USER or an ALTER USER command using the TEMPORARY TABLESPACE option. If no temporary tablespace has been defined for the user, the default temporary tablespace is the SYSTEM tablespace. The default storage characteristics of the containing tablespace determine those of the extents of the temporary segment.
Oracle drops temporary segments when the statement completes.
Because allocation and deallocation of temporary segments occur frequently, it is reasonable to create a special tablespace for temporary segments. By doing so, you can distribute I/O across disk devices, and you may avoid fragmentation of the SYSTEM and other tablespaces that otherwise would hold temporary segments.
For more information about assigning a user's temporary segment tablespace, see Chapter 25, "Controlling Database Access".
Entries for changes to temporary segments used for sort operations are not stored in the redo log, except for space management operations on the temporary segment.
Each database contains one or more rollback segments. A rollback segment records the old values of data that was changed by each transaction (whether or not committed). Rollback segments are used to provide read consistency, to roll back transactions, and to recover the database. For specific information about how rollback segments function in these situations, see the appropriate sections of this book:
Information in a rollback segment consists of several rollback entries. Among other information, a rollback entry includes block information (the filenumber and block ID corresponding to the data that was changed) and the data as it existed before an operation in a transaction. Oracle links rollback entries for the same transaction, so the entries can be found easily if necessary for transaction rollback.
Neither database users nor administrators can access or read rollback segments; only Oracle can write to or read them. (They are owned by the user SYS, no matter which user creates them.)
Rollback entries change data blocks in the rollback segment, and Oracle records all changes to data blocks, including rollback entries, in the redo log. This second recording of the rollback information is very important for active transactions (not yet committed or rolled back) at the time of a system crash. If a system crash occurs, Oracle automatically restores the rollback segment information, including the rollback entries for active transactions, as part of instance or media recovery. Once the recovery is complete, Oracle performs the actual rollbacks of transactions that had been neither committed nor rolled back at the time of the system crash.
For each rollback segment, Oracle maintains a transaction table-a list of all transactions that use the associated rollback segment and the rollback entries for each change performed by these transactions. Oracle uses the rollback entries in a rollback segment to perform a transaction rollback and to create read-consistent results for queries.
Rollback segments record the data prior to change on a per-transaction basis. For every transaction, Oracle links each new change to the previous change. If you must roll back the transaction, Oracle applies the changes in a chain to the data blocks in an order that restores the data to its previous state.
Similarly, when Oracle needs to provide a read-consistent set of results for a query, it can use information in rollback segments to create a set of data consistent with respect to a single point in time.
Each time a user's transaction begins, the transaction is assigned to a rollback segment in one of two ways:
For the duration of a transaction, the associated user process writes rollback information only to the assigned rollback segment.
When you commit a transaction, Oracle releases the rollback information but does not immediately destroy it. The information remains in the rollback segment to create read-consistent views of pertinent data for queries that started before the transaction committed. To guarantee that rollback data is available for as long as possible for such views, Oracle writes the extents of rollback segments sequentially. When the last extent of the rollback segment becomes full, Oracle continues writing rollback data by wrapping around to the first extent in the segment. A long-running transaction (idle or active) may require a new extent to be allocated for the rollback segment. See Figure 2-6, Figure 2-7, and Figure 2-8 for more information about how transactions use the extents of a rollback segment.
Each rollback segment can handle a fixed number of transactions from one instance. Unless you explicitly assign transactions to particular rollback segments, Oracle distributes active transactions across available rollback segments so that all rollback segments are assigned approximately the same number of active transactions. Distribution does not depend on the size of the available rollback segments. Therefore, in environments where all transactions generate the same amount of rollback information, all rollback segments can be the same size.
When you create a rollback segment, you can specify storage parameters to control the allocation of extents for that segment. Each rollback segment must have at least two extents allocated.
One transaction writes sequentially to a single rollback segment. Each transaction writes to only one extent of the rollback segment at any given time. Many active transactions can write concurrently to a single rollback segment-even the same extent of a rollback segment; however, each data block in a rollback segment's extent can contain information for only a single transaction.
When a transaction runs out of space in the current extent and needs to continue writing, Oracle finds an available extent of the same rollback segment in one of two ways:
The first transaction that needs to acquire more rollback space checks the next extent of the rollback segment. If the next extent of the rollback segment does not contain information from an active transaction, Oracle makes it the current extent, and all transactions that need more space from then on can write rollback information to the new current extent. Figure 2-6 illustrates two transactions, T1 and T2, which begin writing in the third extent (E3) and continue writing to the fourth extent (E4) of a rollback segment.
As the transactions continue writing and fill the current extent, Oracle checks the next extent already allocated for the rollback segment to determine if it is available. In Figure 2-7, when E4 is completely full, T1 and T2 continue any further writing to the next extent allocated for the rollback segment that is available; in this figure, E1 is the next extent. This figure shows the cyclical nature of extent use in rollback segments.
To continue writing rollback information for a transaction, Oracle always tries to reuse the next extent in the ring first. However, if the next extent contains data from active transaction, then Oracle must allocate a new extent. Oracle can allocate new extents for a rollback segment until the number of extents reaches the value set for the rollback segment's storage parameter MAXEXTENTS.
Figure 2-8 shows a new extent allocated for a rollback segment. The uncommitted transactions are long running (either idle, active, or persistent in-doubt distributed transactions). At this time, they are writing to the fourth extent, E4, in the rollback segment. However, when E4 is completely full, the transactions cannot continue further writing to the next extent in sequence, E1, because it contains active rollback entries. Therefore, Oracle allocates a new extent, E5, for this rollback segment, and the transactions continue writing to this new extent.
When you drop a rollback segment, Oracle returns all extents of the rollback segment to its tablespace. The returned extents are then available to other segments in the tablespace.
When you create or alter a rollback segment, you can use the storage parameter OPTIMAL (which applies only to rollback segments) to specify the optimal size of the segment in bytes. If a transaction needs to continue writing rollback information from one extent to another extent in the rollback segment, Oracle compares the current size of the rollback segment to the segment's optimal size. If the rollback segment is larger than its optimal size and the extents immediately following the extent just filled are inactive, Oracle deallocates consecutive nonactive extents from the rollback segment until the total size of the rollback segment is equal to or close to but not less than its optimal size. Oracle always frees the oldest inactive extents, as these are the least likely to be used by consistent reads.
A rollback segment's OPTIMAL setting cannot be less than the combined space allocated for the minimum number of extents for the segment:
Oracle creates an initial rollback segment called SYSTEM whenever a database is created. This segment is in the SYSTEM tablespace and uses that tablespace's default storage parameters. You cannot drop the SYSTEM rollback segment. An instance always acquires the SYSTEM rollback segment in addition to any other rollback segments it needs.
If there are multiple rollback segments, Oracle tries to use the SYSTEM rollback segment only for special system transactions and distributes user transactions among other rollback segments; if there are too many transactions for the non-SYSTEM rollback segments, Oracle uses the SYSTEM segment as necessary. In general, after database creation, you should create at least one additional rollback segment in the SYSTEM tablespace.
When an Oracle instance opens a database, it must acquire one or more rollback segments so that the instance can handle rollback information produced by subsequent transactions. An instance can acquire both private and public rollback segments. A private rollback segment is acquired explicitly by an instance when the instance opens a database. Public rollback segments form a pool of rollback segments that any instance requiring a rollback segment can use.
Any number of private and public rollback segments can exist in a database. As an instance opens a database, the instance attempts to acquire one or more rollback segments according to the following rules:
CEIL is a SQL function that returns the smallest integer greater than or equal to the numeric input. In the example above, if TRANSACTIONS equal 155 and TRANSACTIONS_PER_ROLLBACK_SEGMENT equal 10, then the instance will try to acquire at least 16 rollback segments. (However, an instance can open the database even if the instance cannot acquire the number of rollback segments given by the division above.)
Once an instance claims a public rollback segment, no other instance can use that segment until either the rollback segment is taken offline or the instance that claimed the rollback segment is shut down.
A database used by the Oracle Parallel Server optionally can have only public and no private segments, as long as the number of segments in the database is high enough to ensure that each instance that opens the database can acquire at least two rollback segments, one of which is the SYSTEM rollback segment. However, when using the Oracle Parallel Server, you may want to use private rollback segments.
See Oracle8 Parallel Server Concepts and Administration for more information about rollback segment use in an Oracle Parallel Server.
A rollback segment is always in one of several states, depending on whether it is offline, acquired by an instance, involved in an unresolved transaction, in need of recovery, or dropped. The state of the rollback segment determines whether it can be used in transactions, as well as which administrative procedures a DBA can perform on it.
The rollback segment states are:
Has not been acquired (brought online) by any instance.
Has been acquired (brought online) by an instance; may contain data from active transactions.
Contains data from uncommitted transactions that cannot be rolled back (because the data files involved are inaccessible), or is corrupted.
Contains data from an in-doubt transaction (that is, an unresolved distributed transaction).
Has been dropped (The space once allocated to this rollback segment will later be used when a new rollback segment is created.)
The data dictionary table DBA_ROLLBACK_SEGS lists the state of each rollback segment, along with other rollback information. Figure 2-9 shows how a rollback segment moves from one state to another.
The PARTLY AVAILABLE and NEEDS RECOVERY states are very similar. A rollback segment in either state usually contains data from an unresolved transaction.
See Oracle8 Distributed Database Systems for information about failures in distributed transactions.
If you bring a PARTLY AVAILABLE rollback segment online (by a command or during instance startup), Oracle can use it for new transactions. However, the in-doubt transaction still holds some of its transaction table entries, so the number of new transactions that can use the rollback segment is limited. (See "When Rollback Information Is Required" on page 2-18 for information on the transaction table.)
Also, until you resolve the in-doubt transaction, the transaction continues to hold the extents it acquired in the rollback segment, preventing other transactions from using them. Thus, the rollback segment might need to acquire new extents for the active transactions, and therefore grow. To prevent the rollback segment from growing, a database administrator might prefer to create a new rollback segment for transactions to use until the in-doubt transaction is resolved, rather than bring the PARTLY AVAILABLE segment online.
When a tablespace goes offline so that transactions cannot be rolled back immediately, Oracle writes to a deferred rollback segment. The deferred rollback segment contains the rollback entries that could not be applied to the tablespace, so that they can be applied when the tablespace comes back online. These segments disappear as soon as the tablespace is brought back online and recovered. Oracle automatically creates deferred rollback segments in the SYSTEM tablespace.