DBA SGA+PGA Memory Configuration

Description

Categories
Enginatics
Repository
GitHub
Current SGA and PGA memory configuration in gigabytes.

A frequent configuration problem is not making full use of the available hardware resources, especially the physical RAM of the database server.
This report shows the servers SGA, PGA and CPU configuration in comparison to the available hardware.
For maximum performance, configure the SGA+PGA to use the full available memory of your server minus a few gig for OS level caching, e.g. for writing and reading of PLSQL output files on the DB node and for process memory (an estimated 4MB per process, see below).

Oracle's performance tuning guide:
https://docs.oracle.com/en/database/oracle/oracle-database/12.2/tgdba/database-memory-allocation.html

Oracle's exadata best practices whitepaper:
SUM of databases (SGA_TARGET + PGA_AGGREGATE_TARGET) + 4 MB * (Maximum PROCESSES) < Physical Memory per Database Node

Tom Kyte's recommendation:
"... if you want it all automatic, give ALL FREE MEMORY on your machine to the SGA and be done with it. No more monitoring, no more thinking about how it could be, should be, would be used.
Set the PGA aggregate target and the SGA target to be a tad less than physical memory on the machine and you are done."
https://asktom.oracle.com/pls/asktom/f?p=100:11:0::::P11_QUESTION_ID:30011178429375

Database created or cloned is the date of the last clone: the latest of the controlfile creation date (reset by Rapid Clone and RMAN DUPLICATE), the last open with resetlogs (a clone restored from a backup or a storage snapshot keeps the source's creation date) and, on a multitenant database, the PDB creation date.

The report also shows the status of system statistics.
select * from sys.aux_stats$

If CPUSPEEDNW still shows the default of 4096, it either means that the stats have never been gathered or it could also mean that your storage is too fast to gather them with the standard NOWORKLOAD method and you should set reasonable values manually then.
exec dbms_stats.gather_system_stats();
exec dbms_stats.set_system_stats('IOSEEKTIM',1);
exec dbms_stats.set_system_stats('IOTFRSPEED',85782);

There is also an Exadata mode, considering the faster storage correctly and updating the MBRC too
exec dbms_stats.gather_system_stats('EXADATA');

The following command checks if hugepages are configured and how many are in use:
grep Huge /proc/meminfo
https://access.redhat.com/solutions/320303

The SGA and PGA of every other Oracle instance on the host (other databases, ASM, a second EBS environment) share its memory. To list them and the free memory:
ps -eo user,args | grep -E '[o]ra_pmon_|[a]sm_pmon_'
grep -v '^#' /etc/oratab
free -g

To configure hugepages, set the vm.nr_hugepages in the /etc/sysctl.conf file and hard and soft memlock in /etc/security/limits.conf for the oracle user.
/etc/sysctl.conf
vm.nr_hugepages = SGA GB x 512 + 5
/etc/security/limits.conf
oracle soft memlock (SGA GB x 512 + 5) * 2048
oracle hard memlock (SGA GB x 512 + 5) * 2048
https://www.enginatics.com/blog/how-to-configure-hugepages-for-oracle-ebs/

Check if all database parameters are set according to Oracle's requirement:
Database Initialization Parameters for Oracle E-Business Suite Release 12 (KA1002)
https://support.oracle.com/support/?kmContentId=11158187

To perform this check automatically, install and run Oracle's 'Database Performance and Statistics Analyzer' concurrent program (KB627702)
https://support.oracle.com/rs?type=doc&id=2126712

SQL 105 lines

Parameters 1

ParameterTypeList of values SQL
Container DataLOV
dhs.dbid=(select vd.dbid from v$database vd)