Tuning HP Tru64 UNIX V5.1A and V5.1B for Oracle

7
Setting vm_swap_eager = 0 will cause the system to enter a deferred swap allocation mode. Use
lazy swap allocation mode for any system that does not overcommit memory resources. If
vm_swap_eager has been set to 0 and the system runs out of allocatable swap space, any process
that tries to allocate additional swap space will be terminated by the kernel. There are no mechanisms
available to protect processes from being terminated in this situation.
ubc_overflow
A new kernel parameter was introduced in Tru64 UNIX V5.1B PK3 Service PAK that allows UBC
memory to be allocated across RADs on certain types of systems (EV7 based systems such as the
AlphaServer ES47, ES80, and GS1280).
Unfortunately, this parameter was incorrectly initialized on some EV68 based systems such as the
AlphaServer GS80, GS160, and GS320. For these types of systems running Tru64 UNIX V5.1B-1
(PK3) Service PAK only, you can manually correct this parameter by setting ubc_overflow = 0.
Initialization of the parameter is correct with Tru64 UNIX V5.1B-2 (PK4) and later.
vm_overflow
On NUMA systems (such as the AlphaServer ES47, ES80, GS160, GS320, and GS1280 systems),
HP recommends the setting vm_overflow = 1. This is only possible with Tru64 UNIX V5.1B-3 (PK5
Service PAK) or later and should only be done when the patches described in the Tru64
UNIX/TruCluster Server V5.1B-3 with PK5 Service PAK section are applied.
With this patch enabled, the system can more easily allocate memory in other RADs when the local
RAD is low on memory. This prevents systems from paging and/or processes stalled for minutes when
allocated more VM data structures, even though sufficient free memory is available in other RADs.
Shared Memory Allocation and Granularity Hints
HP Tru64 UNIX supports several different options to allocate shared memory for Oracle, including:
• Default shared memory allocation
The traditional option is to use Segmented Shared Memory (SSM), which is controlled with the
ssm_threshold tunable parameter in the ipc subsystem. The default for Tru64 UNIX V5.x is to
use SSM type shared memory (with ipc:ssm_threshold set to its default value of 8838608). HP
recommends this type of shared memory allocation for all systems, except those that need to get the
maximum performance out of the system.
• Big pages
Another option available is to use granularity hints memory. This can either be done by statically
reserving an area of memory specifically for this purpose at boot time, or by using a new option in
Tru64 UNIX V5.1B called big pages. Although the big pages option offers the best performance in
addition to a significant management advantage over statically allocated granularity hints memory,
Oracle has not been certified for use on a system with big pages enabled. As a result, Oracle does
not support the use of big pages.
The recommended value for use with Oracle is to leave the parameter vm_bigpg_enabled at its
default value of 0, which is to not use big pages”.
HP supports the use of big pages in general, provided that segmentation is disabled, that is,
vm:vm_bigpg_seg=0.
• Statically allocated granularity hints memory
Statically allocated granularity hints memory is reserved at boot time and cannot be used for any
other purpose, nor can it be returned to the system or reclaimed when not in use. It will be solely