Tuning Recovery - Expert Oracle RAC Performance Diagnostics and Tuning - page 323

Figure 10-1. Redo log switches
For performance reasons, redo log switches should be kept to as low a value as possible when considering the
recovery times involved. Large redo log files with few switches per hour would help reduce I/O contention and wait
times. However, smaller sized redo log files with more frequent switches will ensure data is moved from the log files
to the data files, keeping the recovery required to a minimal size should an instance crash. However, more controlled
methods such as using the parameter FAST_START_MTTR_TARGET (discussed later in this chapter) using the related
advisory should be adhered to to improve recovery times.
Parallelism
This would mean instance recovery should be made efficient, which means it needs to be tuned. Instance recovery
can be tuned using the parameter RECOVERY_PARALLELISM .
RECOVERY_PARALLELISM specifies the number of processes to participate in instance or crash recovery. To force
a serial crash recovery, this parameter should be set to 0 or 1. Along with the RECOVERY_PARALLELISM parameter,
increasing the PARALLEL_EXECUTION_MESSAGE SIZE value to a higher number would help improve recovery slave
session performance. In the same context, increasing the PARALLEL_MIN_SERVER value from its default value of 0
would help pre-spawn recovery slaves at startup, thus avoiding the need to spawn them when recovery is required
and improving recovery performance.
Thu Aug 28 15:51:41 2014
Post SMON to start 1st pass IR
Thu Aug 28 15:51:41 2014
minact-scn: Inst 2 is now the master inc#:4 mmon proc-id:5372 status:0x7
minact-scn status: grec-scn:0x0000.00000000 gmin-scn:0x0000.023bc527 gcalc-scn:0x0000.023bdc42
minact-scn: master found reconf/inst-rec before recscn scan old-inc#:4 new-inc#:4
Thu Aug 28 15:51:41 2014
Instance recovery: looking for dead threads
Thu Aug 28 15:51:43 2014
Submitted all GCS remote-cache requests
Fix write in gcs resources
Thu Aug 28 15:51:43 2014
Reconfiguration complete (total time 9.4 secs)
Thu Aug 28 15:51:45 2014
Beginning instance recovery of 1 threads
parallel recovery started with 4 processes