 |
Index for
Section 5 |
|
 |
Alphabetical
listing for S |
|
 |
Bottom of
page |
|
sys_attrs_vm(5)
NAME
sys_attrs_vm - system attributes for the vm kernel subsystem
DESCRIPTION
This reference page describes system attributes for the Virtual Memory (vm)
kernel subsystem. See sys_attrs(5) for general guidelines about changing
system attributes.
In the following list, an asterisk (*) precedes the names of attributes
whose values you can change while the system is running. Changes to values
of attributes whose names are not preceded by an asterisk take effect only
when the system is rebooted.
anon_rss_enforce
A value that sets no limit (0), a soft limit (1), or a hard limit (2)
on the resident set size of a process.
Default value: 0 (no limit)
By default, applications can set a process-specific limit on the number
of pages resident in memory by specifying the RLIMIT_RSS resource value
in a setrlimit() call. However, applications are not required to limit
the resident set size of a process and there is no system-wide default
limit. Therefore, the resident set size for a process is limited only
by system memory restrictions. If the demand for memory exceeds the
number of free pages, processes with large resident set sizes are
likely candidates for swapping.
The anon_rss_enforce attribute enables different levels of control over
process set sizes and when the pages that a process is using in
anonymous memory are swapped out (blocking the process) during times of
contention for free pages. Setting anon_rss_enforce to either 1 or 2,
allows you to enforce a system-wide limit on resident set size for a
process through the vm_rss_max_percent attribute. Setting
anon_rss_enforce to 1 (a soft limit), enables finer control over
process blocking and paging of anonymous memory by allowing you to set
the vm_rss_block_target and vm_rss_wakeup_target attributes.
When anon_rss_enforce is set to 2, the resident set size for a process
cannot exceed the system-wide limit set by the vm_rss_max_percent
attribute or a process-specific limit, if any, that is set by an
application's setrlimit() call. When the resident set size exceeds
either of these limits, the system starts to swap out pages of
anonymous memory that the process is already using to keep the resident
set size within the specified limit.
When anon_rss_enforce is set to 1, any system-default and process-
specific limits on resident set size still apply and will cause
swapping to occur when exceeded. Otherwise, a process's pages are
swapped out when the number of free pages is less than the value of the
vm_rss_block_target attribute. The process remains blocked until the
number of free pages reaches the value of the vm_rss_wakeup_target.
* boost_pager_priority
This attribute supports diskless systems and enables the pager to be
more responsive. It functions under the following conditions:
·
The diskless driver is loaded and configured. Diskless system
services are part of the Dataless Management Services (DMS). DMS
enables systems to run the operating system from a server without
requiring a local hard disk on each client system.
·
The server is serving a realtime pre-emptive kernel.
Default value: 0 (off)
Maximum value: 1 (on)
* enable_yellow_zone
A value that enables (1) or disables (0) a soft guard page on the
program stack. This allows an application to enter a signal handler on
stack overflows, which otherwise would cause a core dump.
Default value: 0 (disabled)
The enable_yellow_zone attribute is intended for use by systems
programmers who are debugging kernel applications, such as device
drivers.
gh_chunks
Number of 4-MB chunks of memory reserved at boot time for shared memory
use. This memory cannot be used for any other purpose, nor can it be
returned to the system or reclaimed when not being used. On NUMA-aware
systems (GS80, GS160, and GS320), the gh_chunks attribute affects only
the first Resource Affinity Domain (RAD). See the entry for
rad_gh_regions for more information.
Default value: 0 (chunks) (The zero value means that use of granularity
hints is disabled.)
Minimum value: 0
Maximum value: 9,223,372,036,854,775,807
The attributes associated with "granularity hints" (the gh_*attributes)
are sometimes recommended specifically for database servers. Using
segmented shared memory (SSM) is the alternative to using granularity
hints and is recommended for most systems. Therefore, if the gh_chunks
attribute is not set to zero, the ssm_threshold attribute of the ipc
subsystem should be set to zero. If the gh_chunks attribute is set to
zero, the ssm_threshold attribute should not be set to zero.
See your database product documentation and the System Configuration
and Tuning manual for more information about using granularity hints or
SSM.
gh_fail_if_no_mem
A value that enables (1) or disables (0) a failure return by the shmget
function under certain conditions when granularity hints is in use.
When this attribute is set to 1, the shmget() function returns a
failure if the requested segment size is larger than the value of the
gh-min-seg-size attribute and if there is insufficient memory allocated
by the gh-chunks attribute to satisfy the request.
Default value: 1 (enabled)
gh_front_alloc
A value that specifies whether the memory reserved for granularity
hints is (1) or is not (0) allocated from low physical memory
addresses. Allocation from low physical memory addresses is useful if
you have an odd number of memory boards.
Default value: 1 (allocation from low physical memory addresses)
gh_keep_sorted
Specifies whether the memory reserved for granularity hints is (1) or
is not (0) sorted.
Default value: 0 (not sorted)
gh_min_seg_size
Size, in bytes, of the segment in which shared memory is allocated from
the memory reserved for shared memory, according to the value of the
gh-chunks attribute.
Default value: 8,388,608 (bytes, or 8 MB)
Minimum value: 0
Maximum value: 9,223,372,036,854,775,807
kernel_stack_pages
Number of pages per thread that are used for stack space in kernel
mode.
Default value: 2 (pages per thread)
Minimum value: 2
Maximum value: 3
The sysconfig command may display 0 (zero) when the actual setting is
2. This error will be corrected in a release following Tru64 UNIX
Version 5.0.
It is strongly recommended that you not modify kernel_stack_pages
unless directed to do so by your support representative. In the event
of a kernel stack not valid halt error that is caused by a kernel stack
overflow problem, increasing the value of kernel_stack_pages may work
around the problem. This workaround will not be successful if the error
occurred because the stack pointer became corrupted. In any event, a
kernel stack not valid halt error is always an unexpected error that
should be reported to your support representative for further
investigation.
* kstack_free_target
Number of freed kernel stack pages that are saved for reuse. Above this
limit, freed kernel stack pages are immediately deallocated.
Default value: 5 (pages)
Minimum value: 0
Maximum value: 2,147,483,647
Deallocation of freed kernel stack pages ensures that memory is
available for other operations. However, the processor time required
for deallocating freed kernel stack pages has a negative performance
impact that might be more noticeable on NUMA-enabled systems (GS80,
GS160, GS320) than on other systems. You can use the
kstack_free_target value to make the most appropriate tradeoff between
increased memory consumption and time spent by CPUs in a purge
operation.
You can change the value of the kstack_free_target attribute while the
system is running.
malloc_percpu_cache
A value that enables (1) or disables (0) caching of malloc memory on a
per CPU basis.
Default value: 1
Do not modify the default setting for this attribute unless instructed
to do so by support personnel or by patch kit documentation.
new_wire_method
Default value: 1 (on)
Do not modify the default setting for this attribute unless instructed
to do so by support personnel or by patch kit documentation.
private_cache_percent
Percentage of the secondary cache that is reserved for anonymous
(nonshared) memory. Increasing the cache for anonymous memory reduces
the cache space available for file-backed memory (shared). This
attribute is useful only for benchmarking.
Default value: 0 (percent)
Minimum value: 0
Maximum value: 100
rad_gh_regions[n]
For NUMA-aware systems (GS80, GS160, and GS320), the granularity hints
chunk size (in megabytes) for the Resource Affinity Domain (RAD)
identified by n. There are 64 elements in the attribute array,
rad_gh_regions[0] to rad_gh_regions[63]. Although all elements in the
array are visible on all systems, the kernel uses only the element
values corresponding to RADs that exist on the system. See the entry
for the gh_chunks attribute for general information about granularity
hints memory allocation.
Default value: 0 (MB) (Granularity hints is disabled.)
The array of rad_gh_regions[n] attributes replace the gh_chunks
attribute, which affects only the first or (for non-NUMA systems) only
RAD (rad_gh_regions[0]) supported by the system. Although gh_chunks and
the set of rad_gh_regions attributes both specify how much memory is
manipulated through granularity hints memory allocation, the unit of
measurement for the former is 4-megabyte units whereas the unit of
measurement for the latter is megabytes. Therefore:
rad_gh_regions[0] = gh_chunks * 4
Setting the gh_chunks attribute, not the rad_gh_regions[0] attribute,
is recommended if you want to use granularity hints memory allocation
on non-NUMA systems.
replicate_user_text
A value that controls whether user text can or cannot be
replicated on multiple CPUs of a NUMA-enabled system (GS80, GS160,
GS320). When the value is 1, replication of user text is enabled. When
the value is 0, replication of user text is disabled. This attribute is
sometimes used by kernel developers when debugging software for NUMA-
enabled systems; however, the attribute is not for general use. (The
value is ignored on non-NUMA systems and changing it to 0 on NUMA
systems might degrade performance.)
Default value: 1
Do not change the value of this attribute unless instructed to do so by
support personnel or patch kit instructions.
swapdevice
The device partitions reserved for swapping. This is a comma-separated
string, for example /dev/disk/dsk0g,/dev/disk/dsk0d that can be up to
256 bytes in length.
* ubc_borrowpercent
Percentage of memory above which the UBC is only borrowing memory from
the virtual memory subsystem. Paging does not occur until the UBC has
returned all its borrowed pages.
Default value: 20 (percent)
Minimum value: 0
Maximum value: 100
Increasing this value may increase UBC cache effectiveness and improve
throughput; however, the cost is a likely degradation of system
response time during a low memory condition.
ubc_ffl
Obsolete; currently ignored by the software.
* ubc_kluster_cnt
Specifies the number of pages to consolidate before initiating an I/O
operation.
Default value: 32 (pages)
Minimum value: 0
Maximum value: 512
The default value is appropriate for the vast majority of systems.
Raising this value may improve I/O efficiency if relatively few users
and applications write to only a few very large files, and there is
high probability that write operations affect contiguous pages.
However, the cost is increased time spent in memory (and holding locks
for a longer length of time) while the system determines what state
pages are in and which ones can be clustered.
* ubc_maxdirtymetadata_pcnt
A threshold value that forces cleanup of AdvFS metadata that is being
stored in the UBC. The default setting forces return of pages
containing AdvFS metadata when they reach 70 percent of the UBC.
This is not a tuning parameter. Do not modify the default setting
unless directed to do so by support personnel or patch kit
instructions.
Default value: 70 (percent)
Minimum value: 0
Maximum value: 100
* ubc_maxdirtywrites
Number of I/O operations (per second) that the virtual memory subsystem
performs when the number of dirty (modified) pages in the UBC exceeds
the value of the vm-ubcdirtypercent attribute.
Default value: 5 (operations per second)
Minimum value: 0
Maximum value: 2,147,483,647
* ubc_maxpercent
Maximum percentage of physical memory that the UBC can use at one time.
Default value: 100 (percent)
Minimum value: 0
Maximum value: 100
It is recommended that this value be set to a value in the range of 70
to 80 percent. On an overloaded system, values higher than 80 can delay
return of excess UBC pages to vm and adversely affect performance.
* ubc_minpercent
Minimum percentage of physical memory that the UBC can use.
Default value: 10 (percent)
Minimum value: 0
Maximum value: 100
* vm_aggressive_swap
A value that enables (1) or disables (0) the ability of the task
swapper to aggressively swap out idle tasks.
Default value: 0 (disabled)
Setting this attribute to 1 helps prevent a low-memory condition from
occurring and allows more jobs to be run simultaneously. However,
interactive response times are likely to be longer on a system that is
excessively paging and swapping.
* vm_asyncswapbuffers
The number of asynchronous I/O requests per swap partition that can be
outstanding at one time. Asynchronous swap requests are used for
pageout operations and for prewriting modified pages.
Default value: 4 (requests)
Minimum value: 0
Maximum value: 2,147,483,647
vm_clustermap
Size, in bytes, of the kernel cluster submap, which is used to allocate
the scatter/gather map for clustered file and swap I/O.
Default value: 1,048,576 (bytes, or 1 MB)
Minimum value: 0
Maximum value: 922,337,203,854,775,807
vm_clustersize
Maximum size, in bytes, of a single scatter/gather map for a clustered
I/O request.
Default value: 65,536 (bytes, or 64 KB)
Minimum value: 0
Maximum value: 922,337,203,854,775,807
vm_cowfaults
Number of times that the pages of an anonymous object are copy-on-write
faulted after a fork operation but before they are copied as part of
the fork operation.
Default value: 4 (faults)
Minimum value: 0
Maximum value: 2,147,483,647
vm_csubmapsize
Size, in bytes, of the kernel copy submap.
Default value: 1,048,576 (bytes, or 1 MB)
Minimum value: 0
Maximum value: 922,337,203,854,775,807
vm_ffl
Obsolete; currently ignored by the software.
* vm_inswappedmin
Minimum amount of time, in seconds, that a task remains in the
inswapped state before it is considered a candidate for outswapping.
Default value: 1 (second)
Minimum value: 0
Maximum value: 60
vm_max_rdpgio_kluster
Size, in bytes, of the largest pagein (read) cluster that is passed to
the swap device.
Default value: 16,384 (bytes) (16 KB)
Minimum value: 8192
Maximum value: 131,072
vm_max_wrpgio_kluster
Size, in bytes, of the largest pageout (write) cluster that is passed
to the swap device.
Default value: 32,768 (bytes) (32 KB)
Minimum value: 8192
Maximum value: 131,072
vm_min_kernel_address
Base address of the kernel's virtual address space. The value can be
either Oxffffffff80000000 or Oxfffffffe00000000, which sets the size of
the kernel's virtual address space to either 2 GB or 8 GB,
respectively.
Default value: 18,446,744,073,709,551,615 (2 to the power of 64)
You may need to increase the kernel's virtual address space on very
large memory (VLM) systems (for example, systems with several gigabytes
of physical memory and several thousand large processes).
* vm_page_free_hardswap
The threshold value that stops paging. When the number of pages on the
free list reaches this value, paging stops.
Default value: Varies, depending on physical memory size; about 16
times the value of vm_page_free_target
Minimum value: 0
Maximum value: 2,147,483,647
The vm_page_free_hardswap value is computed from the
vm_page_free_target value, which by default scales with physical memory
size. If you change vm_page_free_target, your change affects
vm_page_free_hardswap as well.
* vm_page_free_min
The threshold value that starts page swapping. When the number of pages
on the free page list falls below this value, paging starts.
Default value: 20 (pages, or twice the amount of vm_page_free_reserved)
Minimum value: 0
Maximum value: 2,147,483,647
* vm_page_free_optimal
The threshold value that begins hard swapping. When the number of pages
on the free list falls below this value for five seconds, hard swapping
begins.
Default value: Automatically scaled by using this formula:
vm_page_free_min + ((vm_page_free_target - vm_page_free_min) / 2)
Minimum value: 0 (pages)
Maximum value: 2,147,483,647
* vm_page_free_reserved
The threshold value that determines when memory is limited to
privileged tasks. When the number of pages on the free page list falls
below this value, only privileged tasks can get memory.
Default value: 10 (pages)
Minimum value: 1
Maximum value: 2,147,483,647
* vm_page_free_swap
The threshold value that begins swapping of idle tasks. When the number
of pages on the free page list falls below this value, idle task
swapping begins.
Default value: Automatically scaled by using this formula:
vm_page_free_min + ((vm_page_free_target - vm_page_free_min) / 2)
Minimum value: 0
Maximum value: 2,147,483,647
* vm_page_free_target
The threshold value that stops paging, When the number of pages on the
free page list reaches this value, paging stops.
Default value: Based on the amount of managed memory that is available
on the system, as shown in the following table:
__________________________________________________
Available Memory (M) vm_page_free_target (pages)
__________________________________________________
Less than 512 128
512 to 1023 256
1024 to 2047 512
2048 to 4095 768
4096 and higher 1024
__________________________________________________
Minimum value: 0 (pages)
Maximum value: 2,147,483,647
* vm_page_prewrite_target
Maximum number of modified UBC pages that the vm subsystem will
prewrite to disk if it anticipates running out of memory. The
prewritten pages are the least recently used (LRU) pages.
Default value: vm_page_free_target * 2
Minimum value: 0
Maximum value: 2,147,483,647
* vm_rss_block_target
A threshold number of free pages that will start swapping of anonymous
memory from the resident set of a process. Paging of anonymous memory
starts when the number of free pages meets or exceeds this value. The
process is blocked until the number of free pages reaches the value set
by the vm_rss_wakeup_target attribute.
Default value: Same as vm_page_free_optimal
Minimum value: 0
Maximum value: 2,147,483,647
The default value of the vm_rss_block_target attribute is the same as
the default value of the vm_page_free_optimal attribute that controls
the threshold value for hard swapping.
You can increase the value of vm_rss_block_target to start paging of
anonymous memory earlier than when hard swapping occurs or decrease the
value to delay paging of anonymous memory beyond the point at which
hard swapping occurs.
* vm_rss_max_percent
A percentage of the total pages of anonymous memory on the system that
is the system-wide limit on the resident set size for any process. The
value of this attribute has an effect only if anon_rss_enforce is set
to 1 or 2.
Default value: 100 (percent)
Minimum value: 1
Maximum value: 100
You can decrease this percentage to enforce a system-wide limit on the
resident set size for any process. Be aware, however, that this limit
applies to privileged, as well as unprivileged, processes and will
override a larger resident set size that may be specified for a process
through the setrlimit() call.
* vm_rss_wakeup_target
A threshold number of free pages that will unblock a process whose
anonymous memory is swapped out. The process is unblocked when the
number of free pages meets this value.
Default value: Same is vm_page_free_optimal
Minimum value: 0
Maximum value: 2,147,483,647
The default value of the vm_rss_wakeup_target attribute is the same as
the default value of the vm_page_free_optimal attribute that controls
the threshold value for hard swapping.
You can increase the value of vm_rss_wakeup_target to free more memory
before unblocking a process or decrease the value to unblock the
process sooner (with less freed memory).
vm_segment_cache_max
Number of text segments that can be cached in the segment cache.
(Applies only if you enable segmentation.)
Default value: 50 (segments)
Minimum value: 0
Maximum value: 8192
The vm subsystem uses the segment cache to cache inactive executables
and shared libraries. Because objects in the segment cache can be
accessed by mapping a page table entry, this cache eliminates I/O
delays for repeated executions and reloads.
Reducing the number of segments in the segment cache can free memory
and help to reduce paging overhead. (The size of each segment depends
on the text size of the executable or the shared library that is being
cached.)
* vm_segmentation
A value that enables (1) or disables (0) the ability of shared regions
of user address space to also share the page tables that map to those
shared regions.
Default value: 1 (enabled)
In a TruCluster environment, this value must be the same on all cluster
members.
vm_swap_eager
Specifies the swap allocation mode, which can be immediate mode (1) or
deferred mode (0).
Default value: 1 (immediate mode)
* vm_syncswapbuffers
The number of synchronous I/O requests that can be outstanding to the
swap partitions at one time. Synchronous swap requests are used for
page-in operations and task swapping.
Default value: 128 (requests)
Minimum value: 1
Maximum value: 2,147,483,647
vm_syswiredpercent
Maximum percentage of physical memory that can be dynamically wired.
The kernel and user processes use this memory for dynamically allocated
data structures and address space, respectively.
Default value: 80 (percent)
Minimum value: 1
Maximum value: 100
*vm_troll_percent
Enables, disables, and tunes the trolling rate for the memory troller
on systems supported by the memory troller.
When enabled, the memory troller continually reads the system's memory
to proactively discover and handle memory errors. The troll rate is
expressed as a percentage of the system's total memory trolled per hour
and you can change it at any time. Valid troll rate settings are:
·
Default value: 4 percent per hour
This default value applies if you do not specify any value for
vm_troll_percent in the /etc/sysconfigtab. At the default troll
rate, each 8 kilobyte memory page is trolled once every 24 hours.
·
Disable value: 0 (zero)
Specify a value of 0 (zero) to disable memory trolling.
·
Range: 1 - 100 percent
Specify a value in the range 1 to 100 to set the troll rate to a
percentage of memory to troll per hour. For example, a troll rate of
50 reads half the total memory in one hour. After all memory is read,
the troller starts a new pass at the beginning of memory.
·
Accelerated trolling: 101 percent
Specify a value greater than 100 percent to invoke one pass
accelerated trolling. At this rate, all system memory is trolled at a
rate of approximately 6000 pages per second, where one page equals 8
kilobytes. Trolling is then automatically disabled after a single
pass. This mode is intended for trolling all memory quickly during
off peak hours.
Low troll rates, such as the 4 percent default, have a negligible
impact on system performance. Processor usage for memory trolling
increases as the troll rate is increased. Refer to memory_trolling(5)
for additional performance information and memory troller usage
instructions.
vm_ubcbuffers
Specifies the number of I/O operations that can be outstanding while
purging dirty (modified) pages from the UBC. The dirty pages are
flushed to disk to reclaim memory. The UBC purge daemon will stop
flushing dirty pages when the number of I/Os reaches the vm_ubcbuffers
limit or there are no more dirty pages in the UBC. AdvFS software does
not use this attribute; only UFS software uses it.
Default value: 256 (I/Os)
Minimum value: 0
Maximum value: 2,147,483,647
For systems running at capacity and on which many interactive users are
performing write operations to UFS file systems, users might detect
slowed response times if many pages are flushed to disk each time the
UBC buffers are purged. Decreasing the value of vm_ubcbuffers causes
shorter but more frequent purge operations, thereby smoothing out
system response times. Do not, however, decrease vm_ubcbuffers to a
value that completely disables purging of dirty pages. One I/O for
certain file systems might be associated with many pages because of
write clustering of dirty pages.
Note
Changes to this attribute only take affect when made at boot time.
You can also set the smoothsync_age attribute of the vfs kernel
subsystem to address response-time delays that can occur during
periods of intense write activity. The smoothsync_age attribute uses
a different metric (age of dirty pages rather than number of I/Os)
to balance the frequency and duration time of purge operations and
therefore does not support the ability of UFS to flush all dirty
pages for the same write operation at the same time. However,
smoothsync_age can be changed while the system is running and is
used by AdvFS as well as UFS software. See sys_attrs_vfs(5) for
information about the smoothsync_age attribute.
* vm_ubcdirtypercent
The percentage of pages that must be dirty (modified) before the UBC
starts writing them to disk.
Default value: 10 (percent)
Minimum value: 0
Maximum value: 100
* vm_ubcfilemaxdirtypages
In the context of an application thread, the number of pages that must
be dirty (modified) before the UBC update daemon starts writing them.
This value is for internal use only.
vm_ubcpagesteal
The minimum number of pages to be available for file expansion. When
the number of available pages falls below this number, the UBC steals
additional pages to anticipate the file's expansion demands.
Default value: 24 (file pages)
Minimum value: 0
Maximum value: 2,147,483,647
* vm_ubcseqpercent
The maximum percentage of UBC memory that can be used to cache a single
file. See vm_ubcseqstartpercent for information about controlling when
the UBC checks this limit.
Default value: 10 (percent)
Minimum value: 0
Maximum value: 100
* vm_ubcseqstartpercent
A threshold value (a percentage of the UBC in terms of its current
size) that determines when the UBC starts to check the percentage of
UBC pages cached for each file object. If the cached page percentage
for any file exceeds the value of vm_ubcseqpercent, the UBC returns
that file's UBC LRU pages to virtual memory.
Default value: 50 (percent)
Minimum value: 0
Maximum value: 100
SEE ALSO
Commands: dxkerneltuner(8), sysconfig(8), and sysconfigdb(8).
Others: memory_trolling(5), sys_attrs_proc(5), and sys_attrs(5).
System Configuration and Tuning
System Administration
|
Index for
Section 5 |
|
|
Alphabetical
listing for S |
|
 |
Top of
page |
|