Cache Fusion

Each Instance of RAC database has it’s own buffer cache in its SGA and all these caches put together forms the cache fusion. The blocks within the RAC instances are transferred using a high speed private interconnect. In order to make sure that no two or more instances are updating the same block at the same time and also to track what each instance is performing with the block, cache coherency and consistency needs to be maintained. Oracle RAC maitains this cache coherency and consistency using two main services GCS (Global Cache Service)and GES (Global Enqueue Service).

GCS with the help of LMS processes, co-ordinates within the RAC instances and records the status of the cached data blocks in the GRD (Global Resource Directory). The GRD is distributed across all the active instances and contains the information of the cached block such as the block number, which instance is owning the current version of the block, mode of the block, role of the block.

GES maintains the coherency of enqueues on the dictionary and library cache. It keeps track of all global enqueues of the resources in the RAC environment.

Based on what a resource holder is requesting to perform, a data block can be of any of the 3 modes.

1. NULL (N): A null mode indicates just a resource holder and has no access rights.

2. Exclusive (X): An exclusive mode signifies exclusive access of the block. This means that the resource holder needs to perform a write operation on the block and no other resource can write over it. However, other resources can perform read operation on the block.

3. Shared (S): A shared mode indicates that the resource holder has a shared lock on the block and is performing a read operation. As the mode name, since the lock is shared, any other resource can also read the block.

In addition to the modes, GCS also has roles for the resources.

1. Local role: When a data block is first read from the disk into the cache, it’s role is said to be LOCAL. This also means that no other modified/dirty copy of the block exists in the cache.

2. Global Role: When a data block is acquired from a remote instance and if this data block is already modified in the remote instance, then the role of the block is said to be GLOBAL. If multiple copies of modified/dirtied data block exists across multiple instances, then the role of this data block is considered to be GLOBAL.

Another important concept that we need to know of is the “Past Image” (PI). As the name says, an image copy of the modified/updated data block is saved by the modifying instance before passing on this data block to the other requesting instance that would like to perform either read or write operation. A PI block is considered to be the current version of the block and in case of a node failure, GCS would start the recovery from this PI block thereby reducing the recovery time. Once the latest version of the data block is written to the disk (checkpoint), GCS informs all the instances having PI to discard those images.

All the above details speak about the data blocks, but when a block is read into the memory, it would be stored in the buffer. The state of a buffer is determined based on the 3 characters.
 

1. Lock mode – N (NULL), S (Shared), X (Exclusive)

2. Role – L(Local), G(Global)

3. Past Image(PI): Number indicating the number of past images

GV$BH view can be used to check the status of the block against an instance of the database. The values represent:

1. “cr“: This represents a NULL lock mode on the block.

2. “SCUR“: This represents that a SHARED lock is held on the block by that particular instance.

3. “XCUR“: This represents an EXCLUSIVE lock is held on the block by that particular instance.