eG Monitoring
 

Measures reported by OraRacSgaTest

The System Global Area (SGA) is the most important memory structure in Oracle. The SGA stores several different components of memory usage that are designed to execute processes to obtain data for user queries as quickly as possible while also maximizing the number of concurrent users that can access the Oracle instance. The main components of the SGA are

  • The buffer cache: This area of memory allows for selective performance gains on obtaining and changing data. The buffer cache stores data blocks that contain row data that has been selected or updated recently. When the user wants to select data from a table, Oracle looks in the buffer cache to see if the data block that contains the row has already been loaded. If it has, then the buffer cache has achieved its selective performance improvement by not having to look for the data block on disk. If not, then Oracle must locate the data block that contains the row, load it into memory, and present the selected output to the user.

  • The shared pool: The two main components of the shared pool are the shared SQL library cache and the data dictionary cache. The shared SQL library cache is designed to store parse information for SQL statements executing against the database. Parse information includes the set of database operations that the SQL execution mechanism will perform in order to obtain data requested by the user processes. This information is treated as a shared resource in the library cache. If another user process comes along wanting to run the same query that Oracle has already parsed for another user, the database will recognize the opportunity for reuse and let the user process utilize the parse information already available in the shared pool. The other component of the shared pool is the data dictionary cache, also referred to by many DBAs as the “row” cache. This memory structure is designed to store the data from the Oracle data dictionary in order to improve response time on data dictionary queries. Since all user processes and the Oracle database internal processes use the data dictionary, the database as a whole benefits in terms of performance from the presence of cached dictionary data in memory.

An Oracle database server brings in data into the SGA before doing any operation on it. So it is critical to monitor the various structures inside the SGA to ensure optimal database performance. The OraRacSgaTest test collects a variety of statistics relating to the various SGA components on each node in an Oracle cluster.

Outputs of the test : One set of results for every node in the Oracle cluster

The measures made by this test are as follows:

Measurement Description Measurement Unit Interpretation
Library_hit_ratio The library cache is a buffer that contains the shared SQL and PL/SQL areas. The library cache hit ratio indicates the percentage of shared SQL statements being reparsed by this cluster node. Percent For a well-tuned database, this ratio is 90% or more. A lower hit ratio may indicate that the memory allocation to the library cache is insufficient. A low value can significantly degrade the database performance. Increasing the value of the SHARED_POOL_SIZE initialization parameter will help in improving the hit ratio.
Data_buffer_hit_ratio Indicates the percentage of time that this cluster node is able to satisfy a request with information that is already available in the memory. Percent Physical I/O takes a significant amount of time, and also increases the CPU resources required. The database configuration should be tuned to ensure that a required block will most likely be in memory. The extent to which this is achieved is measured using the buffer cache hit ratio. For a well-tuned database, this ratio should be 80% or higher. A lower value indicates insufficient memory allocation to the database buffer cache. Increasing the value of the DB_BLOCK_BUFFERS initialization parameter will help in improving the hit ratio. If you are monitoring Oracle 9i or higher, then, note that the DB_BLOCK_BUFFERS parameter is not supported in Oracle 9i or above. It is therefore recommended that you use the equivalent DB_CACHE_SIZE parameter instead.
Dictionary_hit_ratio Indicates the percentage of data dictionary information pertaining to the database, file space availability and object privileges being readily available in this cluster node's memory. Percent As with the case of the library cache, the dictionary cache hit ratio should be at least 90%. A lower value may be due to the insufficient memory allocation to the dictionary cache. Increasing the value of the SHARED_POOL_SIZE parameter will help in improving the hit ratio.
Redolog_buf_miss_ratio Indicates the percentage of requests to this cluster node that had to wait before the redolog buffer is allocated to it. Percent Before any transaction could occur, the before image of the data will be stored in the redo log buffer.

It is crucial to make the redo log buffer available immediately to the transactions without any wait. The above is crucial to improve the overall performance. This measure indicates how many percentage of times it had to wait for a redo log buffer to be allocated. This can be improved by increasing the LOG_BUFFER parameter.
Pct_sorts_on_disk Indicates the percentage of sorts that is happening on the secondary storage disk of this cluster node. Percent For best performance, most sorts should occur in memory; sorts written to disk adversely affect performance. If more than 10% of sorts happen on disk, the database performance could degrade. To improve the sorting performance of a database, consider tuning the parameters SORT_AREA_SIZE and SORT_AREA_RETAINED_SIZE. The dynamically modifiable initialization parameter called SORT_AREA_SIZE specifies the maximum amount of memory to use for each sort. If a significant number of sorts require disk I/O to temporary segments, an application's performance may benefit from increasing the size of the sort area. Oracle 9i (or above) supports the SORT_AREA_SIZE and the SORT_AREA_RETAINED_SIZE parameters only to ensure backward compatibility with previous versions of Oracle. Therefore, while monitoring Oracle 9i or higher, it is recommended that you use the equivalent PGA_AGGREGATE_TARGET parameter instead.