Management of the virtual portion of shared memory

The database server uses memory pools to track memory allocations that are similar in type and size.

Keeping related memory allocations in a pool helps to reduce memory fragmentation. It also enables the database server to free a large allocation of memory at one time, as opposed to freeing each piece that makes up the pool.

All sessions have one or more memory pools. When the database server requires memory, it looks first in the specified pool. If insufficient memory is available in a pool to satisfy a request, the database server adds memory from the system pool. If the database server cannot find enough memory in the system pool, it dynamically allocates more segments to the virtual portion.

The database server allocates virtual shared memory for each of its subsystems (session pools, stacks, heaps, control blocks, system catalog, SPL routine caches, SQL statement cache, sort pools, and message buffers) from pools that track free space through a linked list. When the database server allocates a portion of memory, it first searches the pool free-list for a fragment of sufficient size. If it finds none, it brings new blocks into the pool from the virtual portion. When memory is freed, it goes back to the pool as a free fragment and remains there until the pool is deleted. When the database server starts a session for a client application, for example, it allocates memory for the session pool. When the session terminates, the database server returns the allocated memory as free fragments.