Tuning Global Cache - Expert Oracle RAC Performance Diagnostics and Tuning - page 394

gc current block pin time: Once the block is received, it needs to be pinned into the
buffer before the block is modified. This statistic means that a buffer is pinned, that is, a
user is making a change in it when a BAST 2 (the signal from another instance to send the
data) arrives. This counter indicates the time taken by the current instance to process
BAST. The value is influenced by the current state of the buffer: if the buffer is currently
available, does the DBWR need to flush the dirty buffers to disk before space is found in the
buffer to pin the block? The counter provides the total time taken. A pinned buffer cannot
be sent immediately and will wait on a down convert queue until the user has finished
modifying the buffer. The user will notify LMS when it releases the pin on the buffer. The
value is stored in centiseconds. The following formula determines the average pin time in
milliseconds:

10 × gc current block pin time ÷ gc current blocks served

gc current block flush time: When blocks are modified and commit the change,
Oracle needs to flush the blocks from the buffer cache to the redo logs by the LGWR
background process on the instance holding the block. If BAST (the signal from another
instance to send the data) arrives and the redo for a modification has not been written
to the redo log yet, the block cannot be sent immediately. It will be placed on a log flush
queue until the LGWR has completed writing to the redo logs. When the redo is written, the
LMS can send the block. The value in this statistic provides the time taken for the redo sync
before the data block could be sent.

Depending on the type of data block requests, the following wait events could also be updated:

• gc current block 2 way

• gc current grant busy

• gc current grant congested

• gc cr block 2 way

■ Note Wait events are discussed in Chapter 17.

Scenario 2: Block Request Involving Three Instances
In scenarios where the block requested by another instance is not found on the instance that masters the resource,
the GCS will grant permission to read the block from disk; or if the block is found on another instance, it will send a
message to the holding instance to send a copy of the block to the requesting instance.
As illustrated in Figure 13-5 , there are two possibilities when the block is not found on the instance that is the
master of the object (resource). Figure 13-5 is a four-node RAC cluster. All instances maintain masters for various
objects represented by R1 thru R8. In our discussions, we refer to data blocks that are mastered on instance 4 ( SSKY4 ).

2 BAST (Blocking asynchronous system trap), a notification mechanism to release a block used by the global cache service.