“The only thing you can do easily is be wrong, and that's hardly worth the effort.”
― Norton Juster, The Phantom Tollbooth
Oracle provides two parameters that affect the PGA that look very similar but operate very differently. One of these parameters is the well-known pga_max_size and the otheris a hidden parameter, _pga_max_size. Let’s look at both and see how one can be very effective while the other can create problems with respect to PGA memory management,
DBAs know pga_max_size from extensive documentation from Oracle Corporation and from numerous Oracle professionals writing blog posts about it. It’s a common parameter to set to restrict the overall size of the PGA in releases 11.2 and later. It’s available if Automatic Memory Management (AMM) is not in use; databases running on Linux and using hugepages would be in this group since AMM and hugepages is not a supported combination. Hugepages are available for IPC (Inter-Process Communication) shared memory; this is the ‘standard’ shared memory model (starting with UNIX System V) allowing multiple processes to access the same shared memory segment. There is also another form of shared memory segment, the memory-mapped file, and currently such segments are not supported by hugepages. Oracle, on Linux, gives you a choice of using hugepages or memory-mapped files and you implement that choice by selecting to use (or not use) Automatic Memory Management (AMM). Using Automatic Shared Memory Management (ASMM) allows the DBA to set such parameters as sga_target, sga_max_size, pga_aggregate_target and pga_max_size and have some control how those memory areas are sized.
Using pga_max_size is a simple task:
SQL> alter system set pga_max_size=2G;
Now Oracle will do its best to limit the overall PGA size to the requested value but remember this is a targeted max size, not an absolute. It is more restrictive than pga_aggregate_target, meaning it’s less likely to be exceeded.
On to its sister parameter, _pga_max_size. This parameter regulates the size of the PGA memory allocated to a single process. Oracle sets this using calculations based on pga_aggregate_target and pga_max_size and, since it is an ‘undocumented’ parameter, it should NOT be changed at the whim of the DBA. Setting this to any value prevents Oracle from setting it based on its standard calculations and can seriously impact database performance and memory usage. If, for example, the DBA does this:
SQL> alter system set "_pga_max_size"=2G;
Oracle is now capable of allocating up to 2 GB of PGA to each and every process started after that change has taken place. On an exceptionally active and busy system, with parallel processing enabled, each process can have up to 2 GB of RAM in its PGA. Since many systems still don’t have terabytes of RAM installed such allocations can bring the database, and the server, to a grinding halt, throwing ORA-04030 errors in the process. This, of course, is not what the DBA intended but it is what the DBA enabled by altering the _pga_max_size parameter. Unfortunately this parameter (_pga_max_size) is still being written on in blogs that provide ‘information’, which hasn’t been validated, to the Oracle community.
Knowledge is power; unfortunately unverified ‘information’ is seen as knowledge (especially since it’s a common misconception that ‘if it’s on the Internet it MUST be true’ which isn’t always the case) by those who don’t apply critical thinking to what they read. I know of DBAs who set _pga_max_size to match the pga_max_size parameter and found, to their dismay, that their actions seriously impacted production systems in a negative way. Sometimes in the database world prolific authors are taken as experts and their words looked upon as gospel. Unfortunately prolific doesn’t necessarily mean reliable.
It’s always best to test what others tell you before assuming the advice given to you is right.