System V Shared Memory

System V shared memory allows the creation of a segment by a process. Cooperating processes can attach to the memory segment (subject to access permissions on the segment) and gain access to the data contained in the segment. This capability is implemented as a loadable module. Entries in the /etc/system file must contain the shmsys: prefix.

A special kind of shared memory known as intimate shared memory (ISM) is used by DBMS vendors to maximize performance. When a shared memory segment is made into an ISM segment, the memory for the segment is locked. This feature enables a faster I/O path to be followed and improves memory usage. A number of kernel resources describing the segment are then shared between all processes that attach to the segment in ISM mode.

For information about the changes to shared memory resources in the Oracle Solaris 10 release, see System V IPC Configuration.

For detailed information about using the new resource controls in the Oracle Solaris 10 release, see Chapter 6, Resource Controls (Overview), in System Administration Guide: Oracle Solaris Containers-Resource Management and Oracle Solaris Zones.

For legacy information about the obsolete System V shared memory parameters, see Parameters That Are Obsolete or Have Been Removed (Oracle Solaris 10).

segspt_minfree

Description

Identifies pages of system memory that cannot be allocated for ISM shared memory.

Data Type

Unsigned long

Default

5 percent of available system memory when the first ISM segment is created

Range

0 to 50 percent of physical memory

Units

Pages

Dynamic?

Yes

Validation

None. Values that are too small can cause the system to hang or performance to severely degrade when memory is consumed with ISM segments.

When to Change

On database servers with large amounts of physical memory using ISM, the value of this parameter can be decreased. If ISM segments are not used, this parameter has no effect. A maximum value of 128 MB (0x4000) is almost certainly sufficient on large memory machines.

Commitment Level

Unstable

pr_segp_disable

Description

Disables the page lock cache flushing when trying to retire a page that might belong to ISM.

When locked or busy (heavy I/O) pages are in the pending page retirement queue, the page retire thread flushes the segp_cache to encourage retirement of pending pages that might be owned by ISM. Periodic or repeated flushes of the segp_cache can be a bottleneck for high memory machines.

Default behavior is to flush the page cache every 30 seconds and if locked pages are observed in queue, then timeout exponentially backs off until 1 hour in multiples of 2.

Enabling pr_segp_disable does not disable the system's ability to retire memory pages, such as those that are faulted as a result of system diagnostic measures.

Data Type

Boolean

Default

1 (disabled)

Range

0 (enabled) and 1 (disabled)

Dynamic?

No

Validation

No

When to Change

When locked or busy (heavy I/O) pages are in the pending page retirement queue, the page retire thread flushes the segp_cache to encourage retirement of pending pages that might be owned by ISM. Periodic or repeated flushes of the segp_cache can be a bottleneck for high memory machines.

If you have a latency sensitive database or a large shared memory application, consider disabling this parameter to completely skip segp cache flushing.

Symptoms of locked kernel pages that can't be retired are as follows:

• Brief database latency or momentary database unresponsive events along with brief periodic elevated SYS CPU events upon successful page retirements, However, locked or busy pages that repeatedly fail to retire might continue to trigger page retirement threads at slower rates.

  For example, locked memory pages that can't be retired might retry at small intervals and repeat forever at 1 hour intervals. After the system reboots, the scheduled pages might retire, or it might start trying again at 30 seconds, the default rate.

• Brief unexpected or elevated smtx lock contention might be seen when monitoring segspt_shmfault, segspt_softunlock, segspt_shmpagelock, segspt_shmfree, segspt_shmunmap, segspt_shmattach, and segspt_dismfault structures.

Commitment Level

Unstable