9 Process Architecture

Database Writer Process (DBWn)

The database writer process (DBWn) writes the contents of buffers to datafiles. The DBWn processes are responsible for writing modified (dirty) buffers in the database buffer cache to disk. Although one database writer process (DBW0) is adequate for most systems, you can configure additional processes (DBW1 through DBW9 and DBWa through DBWj) to improve write performance if your system modifies data heavily. These additional DBWn processes are not useful on uniprocessor systems.

When a buffer in the database buffer cache is modified, it is marked dirty. A cold buffer is a buffer that has not been recently used according to the least recently used (LRU) algorithm. The DBWn process writes cold, dirty buffers to disk so that user processes are able to find cold, clean buffers that can be used to read new blocks into the cache. As buffers are dirtied by user processes, the number of free buffers diminishes. If the number of free buffers drops too low, user processes that must read blocks from disk into the cache are not able to find free buffers. DBWn manages the buffer cache so that user processes can always find free buffers.

By writing cold, dirty buffers to disk, DBWn improves the performance of finding free buffers while keeping recently used buffers resident in memory. For example, blocks that are part of frequently accessed small tables or indexes are kept in the cache so that they do not need to be read in again from disk. The LRU algorithm keeps more frequently accessed blocks in the buffer cache so that when a buffer is written to disk, it is unlikely to contain data that will be useful soon.

The initialization parameter DB_WRITER_PROCESSES specifies the number of DBWn processes. The maximum number of DBWn processes is 20. If it is not specified by the user during startup, Oracle determines how to set DB_WRITER_PROCESSES based on the number of CPUs and processor groups.

The DBWn process writes dirty buffers to disk under the following conditions:

• When a server process cannot find a clean reusable buffer after scanning a threshold number of buffers, it signals DBWn to write. DBWn writes dirty buffers to disk asynchronously while performing other processing.

• DBWn periodically writes buffers to advance the checkpoint, which is the position in the redo thread (log) from which instance recovery begins. This log position is determined by the oldest dirty buffer in the buffer cache.

In all cases, DBWn performs batched (multiblock) writes to improve efficiency. The number of blocks written in a multiblock write varies by operating system.

Log Writer Process (LGWR)

The log writer process (LGWR) is responsible for redo log buffer management—writing the redo log buffer to a redo log file on disk. LGWR writes all redo entries that have been copied into the buffer since the last time it wrote.

The redo log buffer is a circular buffer. When LGWR writes redo entries from the redo log buffer to a redo log file, server processes can then copy new entries over the entries in the redo log buffer that have been written to disk. LGWR normally writes fast enough to ensure that space is always available in the buffer for new entries, even when access to the redo log is heavy.

LGWR writes one contiguous portion of the buffer to disk. LGWR writes:

• A commit record when a user process commits a transaction

• Redo log buffers

  • Every three seconds

  • When the redo log buffer is one-third full

  • When a DBWn process writes modified buffers to disk, if necessary

LGWR writes synchronously to the active mirrored group of redo log files. If one of the files in the group is damaged or unavailable, LGWR continues writing to other files in the group and logs an error in the LGWR trace file and in the system alert log. If all files in a group are damaged, or the group is unavailable because it has not been archived, LGWR cannot continue to function.

When a user issues a COMMIT statement, LGWR puts a commit record in the redo log buffer and writes it to disk immediately, along with the transaction's redo entries. The corresponding changes to data blocks are deferred until it is more efficient to write them. This is called a fast commit mechanism. The atomic write of the redo entry containing the transaction's commit record is the single event that determines the transaction has committed. Oracle returns a success code to the committing transaction, although the data buffers have not yet been written to disk.

When a user commits a transaction, the transaction is assigned a system change number (SCN), which Oracle records along with the transaction's redo entries in the redo log. SCNs are recorded in the redo log so that recovery operations can be synchronized in Real Application Clusters and distributed databases.

In times of high activity, LGWR can write to the redo log file using group commits. For example, assume that a user commits a transaction. LGWR must write the transaction's redo entries to disk, and as this happens, other users issue COMMIT statements. However, LGWR cannot write to the redo log file to commit these transactions until it has completed its previous write operation. After the first transaction's entries are written to the redo log file, the entire list of redo entries of waiting transactions (not yet committed) can be written to disk in one operation, requiring less I/O than do transaction entries handled individually. Therefore, Oracle minimizes disk I/O and maximizes performance of LGWR. If requests to commit continue at a high rate, then every write (by LGWR) from the redo log buffer can contain multiple commit records.

Checkpoint Process (CKPT)

When a checkpoint occurs, Oracle must update the headers of all datafiles to record the details of the checkpoint. This is done by the CKPT process. The CKPT process does not write blocks to disk; DBWn always performs that work.

The statistic DBWR checkpoints displayed by the System_Statistics monitor in Enterprise Manager indicates the number of checkpoint requests completed.

System Monitor Process (SMON)

The system monitor process (SMON) performs recovery, if necessary, at instance startup. SMON is also responsible for cleaning up temporary segments that are no longer in use and for coalescing contiguous free extents within dictionary managed tablespaces. If any terminated transactions were skipped during instance recovery because of file-read or offline errors, SMON recovers them when the tablespace or file is brought back online. SMON checks regularly to see whether it is needed. Other processes can call SMON if they detect a need for it.

With Real Application Clusters, the SMON process of one instance can perform instance recovery for a failed CPU or instance.

Process Monitor Process (PMON)

The process monitor (PMON) performs process recovery when a user process fails. PMON is responsible for cleaning up the database buffer cache and freeing resources that the user process was using. For example, it resets the status of the active transaction table, releases locks, and removes the process ID from the list of active processes.

PMON periodically checks the status of dispatcher and server processes, and restarts any that have stopped running (but not any that Oracle has terminated intentionally). PMON also registers information about the instance and dispatcher processes with the network listener.

Like SMON, PMON checks regularly to see whether it is needed and can be called if another process detects the need for it.

Recoverer Process (RECO)

The recoverer process (RECO) is a background process used with the distributed database configuration that automatically resolves failures involving distributed transactions. The RECO process of a node automatically connects to other databases involved in an in-doubt distributed transaction. When the RECO process reestablishes a connection between involved database servers, it automatically resolves all in-doubt transactions, removing from each database's pending transaction table any rows that correspond to the resolved in-doubt transactions.

If the RECO process fails to connect with a remote server, RECO automatically tries to connect again after a timed interval. However, RECO waits an increasing amount of time (growing exponentially) before it attempts another connection. The RECO process is present only if the instance permits distributed transactions. The number of concurrent distributed transactions is not limited.

Job Queue Processes

Job queue processes are used for batch processing. They run user jobs. They can be viewed as a scheduler service that can be used to schedule jobs as PL/SQL statements or procedures on an Oracle instance. Given a start date and an interval, the job queue processes try to run the job at the next occurrence of the interval.

Job queue processes are managed dynamically. This allows job queue clients to use more job queue processes when required. The resources used by the new processes are released when they are idle.

Dynamic job queue processes can run a large number of jobs concurrently at a given interval. The job queue processes run user jobs as they are assigned by the CJQ process. Here's what happens:

1. The coordinator process, named CJQ0, periodically selects jobs that need to be run from the system JOB$ table. New jobs selected are ordered by time.

2. The CJQ0 process dynamically spawns job queue slave processes (J000…J999) to run the jobs.

3. The job queue process runs one of the jobs that was selected by the CJQ process for execution. The processes run one job at a time.

4. After the process finishes execution of a single job, it polls for more jobs. If no jobs are scheduled for execution, then it enters a sleep state, from which it wakes up at periodic intervals and polls for more jobs. If the process does not find any new jobs, then it aborts after a preset interval.

The initialization parameter JOB_QUEUE_PROCESSES represents the maximum number of job queue processes that can concurrently run on an instance. However, clients should not assume that all job queue processes are available for job execution.

Archiver Processes (ARCn)

The archiver process (ARCn) copies redo log files to a designated storage device after a log switch has occurred. ARCn processes are present only when the database is in ARCHIVELOG mode, and automatic archiving is enabled.

An Oracle instance can have up to 10 ARCn processes (ARC0 to ARC9). The LGWR process starts a new ARCn process whenever the current number of ARCn processes is insufficient to handle the workload. The alert log keeps a record of when LGWR starts a new ARCn process.

If you anticipate a heavy workload for archiving, such as during bulk loading of data, you can specify multiple archiver processes with the initialization parameter LOG_ARCHIVE_MAX_PROCESSES. The ALTER SYSTEM statement can change the value of this parameter dynamically to increase or decrease the number of ARCn processes. However, you do not need to change this parameter from its default value of 1, because the system determines how many ARCn processes are needed, and LGWR automatically starts up more ARCn processes when the database workload requires more.

Queue Monitor Processes (QMNn)

The queue monitor process is an optional background process for Oracle Streams Advanced Queuing, which monitors the message queues. You can configure up to 10 queue monitor processes. These processes, like the job queue processes, are different from other Oracle background processes in that process failure does not cause the instance to fail.

Other Background Processes

There are several other background processes that might be running. These can include the following:

MMON performs various manageability-related background tasks, for example:

• Issuing alerts whenever a given metrics violates its threshold value

• Taking snapshots by spawning additional process (MMON slaves)

• Capturing statistics value for SQL objects which have been recently modified

MMNL performs frequent and light-weight manageability-related tasks, such as session history capture and metrics computation.

MMAN is used for internal database tasks.

RBAL coordinates rebalance activity for disk groups in an Automatic Storage Management instance. It performs a global open on Automatic Storage Management disks.ORBn performs the actual rebalance data extent movements in an Automatic Storage Management instance. There can be many of these at a time, called ORB0, ORB1, and so forth.

OSMB is present in a database instance using an Automatic Storage Management disk group. It communicates with the Automatic Storage Management instance.

Dispatcher Processes (Dnnn)

The dispatcher processes support shared server configuration by allowing user processes to share a limited number of server processes. With the shared server, fewer shared server processes are required for the same number of users, Therefore, the shared server can support a greater number of users, particularly in client/server environments where the client application and server operate on different computers.

You can create multiple dispatcher processes for a single database instance. At least one dispatcher must be created for each network protocol used with Oracle. The database administrator starts an optimal number of dispatcher processes depending on the operating system limitation and the number of connections for each process, and can add and remove dispatcher processes while the instance runs.

In a shared server configuration, a network listener process waits for connection requests from client applications and routes each to a dispatcher process. If it cannot connect a client application to a dispatcher, the listener process starts a dedicated server process, and connects the client application to the dedicated server. The listener process is not part of an Oracle instance; rather, it is part of the networking processes that work with Oracle.

Shared Server Processes (Snnn)

Each shared server process serves multiple client requests in the shared server configuration. Shared server processes and dedicated server processes provide the same functionality, except shared server processes are not associated with a specific user process. Instead, a shared server process serves any client request in the shared server configuration.

The PGA of a shared server process does not contain user-related data (which needs to be accessible to all shared server processes). The PGA of a shared server process contains only stack space and process-specific variables.

All session-related information is contained in the SGA. Each shared server process needs to be able to access all sessions' data spaces so that any server can handle requests from any session. Space is allocated in the SGA for each session's data space. You can limit the amount of space that a session can allocate by setting the resource limit PRIVATE_SGA to the desired amount of space in the user's profile.

Oracle dynamically adjusts the number of shared server processes based on the length of the request queue. The number of shared server processes that can be created ranges between the values of the initialization parameters SHARED_SERVERS and MAX_SHARED_SERVERS.