All Oracle data is obtained by users from the buffer cache.The basic purpose of the cache is to minimize physical disk I/O by holding (buffering) copies of requested data blocks in memory.
Buffer Aging and LRU Lists
An Oracle database typically has many more data blocks on disk than memory buffers in the buffer cache. Since not all blocks can be buffered in the cache at once, new block requests (cache misses) must find room in the cache to be read in from disk. When this happens, another block in the cache is usually discarded since the cache is normally full (and fixed in size).
The buffer cache is carefully designed to favor keeping frequently requested blocks in memory and allow less popular blocks to be replaced by new block requests.
These decisions about which blocks to replace are made using a least recently used (or LRU) algorithm. This algorithm uses a data structure called the LRU list. This list basically orders the buffers in the cache according to when they were last accessed by a user.
When a block is accessed, it is moved to the MRU (most recently used) end of the list. Blocks in the cache that are not accessed for awhile will find more and more blocks ahead of them in the list, and they will be closer to the LRU end of the list. This is also known as buffer aging in the LRU list.
Buffers are replaced in the cache from the least recently used end of the LRU list. This helps insure that frequently accessed buffers are not discarded, as they are regularly moved to the MRU end of the list with each access.
This mechanism of keeping the most requested blocks in the buffer cache is normally very effective at minimizing disk I/O. Managing a single LRU list can sometimes be a bottleneck in a heavily loaded database
DBWR and the Dirty List
In addition to the LRU list Oracle keeps a list of buffers that contain data that has been modified by users. This list is called the dirty list. Changed data committed by users must eventually be written to disk, as this is the permanent storage of the database.
The DBWR background process is responsible for moving blocks from the dirty list to their permanent locations in disk files.
Dirty blocks cannot be replaced in the cache until they have been written to disk, otherwise the changes would be lost. An overabundance of dirty buffers can negatively impact cache efficiency by reducing available slots for new blocks.
This can happen when DBWR is unable to keep up with the volume of update activity.
Multiple DBWR processes can be configured in this case to increase the capacity to write out dirty blocks.
Buffer Cache Problems
The LRU algorithm of the Oracle buffer cache is normally very good at providing efficient minimization of physical disk I/O. However, there are some situations where normal buffer cache aging may not be the best option for overall performance, for instance:
•Blocks that should not go to the MRU end of the list
•Blocks that should be excluded from aging and stay in the cache .
The first situation can occur when very large tables are accessed randomly by users with very little block use overlap between users. In this case, the MRU end of the list is flooded by blocks that will not result in subsequent cache hits yet age other blocks down to the LRU end of the list.
These other blocks may be replaced when they could have resulted in cache hits had they been kept. The second situation occurs when there are data blocks that will definitely be requested regularly and we want to insure that physical disk I/O is not necessary to obtain their data.
An example might be small lookup tables, or perhaps specific indexes.
The multiple buffer pool feature allows greater control over buffer cache usage to help address these problems and obtain even better cache performance.
Buffer Aging and LRU Lists
An Oracle database typically has many more data blocks on disk than memory buffers in the buffer cache. Since not all blocks can be buffered in the cache at once, new block requests (cache misses) must find room in the cache to be read in from disk. When this happens, another block in the cache is usually discarded since the cache is normally full (and fixed in size).
The buffer cache is carefully designed to favor keeping frequently requested blocks in memory and allow less popular blocks to be replaced by new block requests.
These decisions about which blocks to replace are made using a least recently used (or LRU) algorithm. This algorithm uses a data structure called the LRU list. This list basically orders the buffers in the cache according to when they were last accessed by a user.
When a block is accessed, it is moved to the MRU (most recently used) end of the list. Blocks in the cache that are not accessed for awhile will find more and more blocks ahead of them in the list, and they will be closer to the LRU end of the list. This is also known as buffer aging in the LRU list.
Buffers are replaced in the cache from the least recently used end of the LRU list. This helps insure that frequently accessed buffers are not discarded, as they are regularly moved to the MRU end of the list with each access.
This mechanism of keeping the most requested blocks in the buffer cache is normally very effective at minimizing disk I/O. Managing a single LRU list can sometimes be a bottleneck in a heavily loaded database
DBWR and the Dirty List
In addition to the LRU list Oracle keeps a list of buffers that contain data that has been modified by users. This list is called the dirty list. Changed data committed by users must eventually be written to disk, as this is the permanent storage of the database.
The DBWR background process is responsible for moving blocks from the dirty list to their permanent locations in disk files.
Dirty blocks cannot be replaced in the cache until they have been written to disk, otherwise the changes would be lost. An overabundance of dirty buffers can negatively impact cache efficiency by reducing available slots for new blocks.
This can happen when DBWR is unable to keep up with the volume of update activity.
Multiple DBWR processes can be configured in this case to increase the capacity to write out dirty blocks.
Buffer Cache Problems
The LRU algorithm of the Oracle buffer cache is normally very good at providing efficient minimization of physical disk I/O. However, there are some situations where normal buffer cache aging may not be the best option for overall performance, for instance:
•Blocks that should not go to the MRU end of the list
•Blocks that should be excluded from aging and stay in the cache .
The first situation can occur when very large tables are accessed randomly by users with very little block use overlap between users. In this case, the MRU end of the list is flooded by blocks that will not result in subsequent cache hits yet age other blocks down to the LRU end of the list.
These other blocks may be replaced when they could have resulted in cache hits had they been kept. The second situation occurs when there are data blocks that will definitely be requested regularly and we want to insure that physical disk I/O is not necessary to obtain their data.
An example might be small lookup tables, or perhaps specific indexes.
The multiple buffer pool feature allows greater control over buffer cache usage to help address these problems and obtain even better cache performance.
