Skip Headers
Oracle® Database Reference
10g Release 2 (10.2)

Part Number B14237-02
Go to Documentation Home
Home
Go to Book List
Book List
Go to Table of Contents
Contents
Go to Index
Index
Go to Master Index
Master Index
Go to Feedback page
Feedback

Go to previous page
Previous
Go to next page
Next
View PDF

DBA_HIST_INSTANCE_RECOVERY

DBA_HIST_INSTANCE_RECOVERY displays the historical monitoring of the mechanisms available to the user to limit recovery I/O. This view contains snapshots of V$INSTANCE_RECOVERY.

Column Datatype NULL Description
SNAP_ID NUMBER NOT NULL Unique snapshot ID
DBID NUMBER NOT NULL Database ID for the snapshot
INSTANCE_NUMBER NUMBER NOT NULL Instance number for the snapshot
RECOVERY_ESTIMATED_IOS NUMBER   Number of dirty buffers in the buffer cache. In the Standard Edition, this column is always null.
ACTUAL_REDO_BLKS NUMBER   Current actual number of redo blocks required for recovery
TARGET_REDO_BLKS NUMBER   Current target number of redo blocks that must be processed for recovery. This value is the minimum value of the following 3 columns, and identifies which of the 3 user-defined limits determines checkpointing.
LOG_FILE_SIZE_REDO_BLKS NUMBER   Maximum number of redo blocks required to guarantee that a log switch does not occur before the checkpoint completes
LOG_CHKPT_TIMEOUT_REDO_BLKS NUMBER   Number of redo blocks that need to be processed during recovery to satisfy the LOG_CHECKPOINT_TIMEOUT parameter. The value displayed is not meaningful unless LOG_CHECKPOINT_TIMEOUT has been set.
LOG_CHKPT_INTERVAL_REDO_BLKS NUMBER   Number of redo blocks that need to be processed during recovery to satisfy the LOG_CHECKPOINT_INTERVAL parameter. The value displayed is not meaningful unless LOG_CHECKPOINT_INTERVAL has been set.
FAST_START_IO_TARGET_REDO_BLKS NUMBER   This column is obsolete and maintained for backward compatibility. The value of this column is always null.
TARGET_MTTR NUMBER   Effective MTTR (mean time to recover) target value in seconds. The TARGET_MTTR value is calculated based on the value of the FAST_START_MTTR_TARGET parameter (the TARGET_MTTR value is used internally), and is usually an approximation of the parameter's value. However, if the FAST_START_MTTR_TARGET parameter value is very small (for example, one second), or very large (for example, 3600 seconds), then the calculation will produce a target value dictated by system limitations.

In such cases, the TARGET_MTTR value will be the shortest calculated time, or the longest calculated time that recovery is expected to take. If FAST_START_MTTR_TARGET is not specified, then the value of this field is the current estimated MTTR.

ESTIMATED_MTTR NUMBER   Current estimated mean time to recover (MTTR) based on the number of dirty buffers and log blocks (0 if FAST_START_MTTR_TARGET is not specified). This value tells you how long you can expect recovery to take based on the work the system is doing right now.
CKPT_BLOCK_WRITES NUMBER   Number of blocks written by checkpoint writes
OPTIMAL_LOGFILE_SIZE NUMBER   Redo log file size (in megabytes) that is considered optimal based on the current setting of FAST_START_MTTR_TARGET. It is recommended that all online redo logs be configured to be at least this value.
ESTD_CLUSTER_AVAILABLE_TIME NUMBER   Estimated time (in seconds) that the cluster would become partially available should the instance fail. This column is only meaningful in a Real Application Clusters (RAC) environment. In a non-RAC environment, the value of this column is null.
WRITES_MTTR NUMBER   Number of writes driven by the FAST_START_MTTR_TARGET parameter
WRITES_LOGFILE_SIZE NUMBER   Number of writes driven by the smallest redo log file size
WRITES_LOG_CHECKPOINT_SETTINGS NUMBER   Number of writes driven by the LOG_CHECKPOINT_INTERVAL parameter or the LOG_CHECKPOINT_TIMEOUT parameter
WRITES_OTHER_SETTINGS NUMBER   Number of writes driven by other reasons (such as the deprecated FAST_START_IO_TARGET parameter)
WRITES_AUTOTUNE NUMBER   Number of writes due to auto-tune checkpointing
WRITES_FULL_THREAD_CKPT NUMBER   Number of writes due to full thread checkpoints