Capacity Planning and Architecture - Expert Oracle RAC Performance Diagnostics and Tuning - page 41

4

SSKY2 SSKY3

SSKY1

1

R7

R5

R3

R1

R8

500:9996

R4

3

R2

2

SCN 9996

Block 450
Block 490

Block 459 Block 550

Block 500

Block 600

SSKYDB

Figure 2-7. Read/read behavior with no transfer

The following steps are undertaken by SSKY3 to acquire the block from disk:

1. A user session or process attached to instance SSKY3 makes a request for a specific row of

data. SSKY3 determines that the master for this specific resource is SSKY4 . The request is
directed to instance SSKY4 where the GRD for the object is maintained. Oracle allocates a
node to be the resource master based on the demand for the resource on a specific instance.
If the object access increases on another node, Oracle performs a re-mastering operation to
move the resource master for the object to the node .

2. GCS, on verifying the GRD, determines that no other instance in the cluster has a copy of

the block. GCS sends a message to SSKY3 requesting it to read the block from disk.

Instance SSKY3 initiates the I/O request to read the row from disk. The row is contained in
block 500 and has SCN 9996. Because Oracle reads a block of data at a time, other rows are
also retrieved as part of this read operation. The block is read into the buffer of instance
SSKY3 . Instance SSKY3 holds the block with SCN 9996 using a shared local mode.

3.

4. SSKY3 now informs the GCS that the operation is successful. The GCS makes an entry in
the GRD on instance SSKY4 .

Read/Read Behavior with Transfer
We continue with the previous example/illustration. The Oracle process accessed the disk to retrieve a row contained
in block 500 via instance SSKY3 . The block is held in local shared mode, that is, no other instance has a copy of the
block. We assume another user requires access to another row that is part of the same data block 500. This request is
made by a user connected to instance SSKY2 .