erts_alloc
erts_alloc
is an Erlang Run-Time System internal memory
allocator library. erts_alloc
provides the Erlang
Run-Time System with a number of memory allocators.
Allocators
Currently the following allocators are present:
temp_alloc
eheap_alloc
binary_alloc
ets_alloc
driver_alloc
literal_alloc
sl_alloc
ll_alloc
fix_alloc
exec_alloc
std_alloc
sys_alloc
malloc
implementation
used on the specific OS.mseg_alloc
mseg_alloc
is used by other
allocators for allocating memory segments and is currently only
available on systems that have the mmap
system
call. Memory segments that are deallocated are kept for a
while in a segment cache before they are destroyed. When
segments are allocated, cached segments are used if possible
instead of creating new segments. This in order to reduce
the number of system calls made.sys_alloc
and literal_alloc
are always enabled and
cannot be disabled. exec_alloc
is only available if it is needed
and cannot be disabled. mseg_alloc
is always enabled if it is
available and an allocator that uses it is enabled. All other
allocators can be enabled or disabled.
By default all allocators are enabled.
When an allocator is disabled, sys_alloc
is used instead of
the disabled allocator.
The main idea with the erts_alloc
library is to separate
memory blocks that are used differently into different memory
areas, and by this achieving less memory fragmentation. By
putting less effort in finding a good fit for memory blocks that
are frequently allocated than for those less frequently
allocated, a performance gain can be achieved.
The alloc_util framework
Internally a framework called alloc_util
is used for
implementing allocators. sys_alloc
, and
mseg_alloc
do not use this framework; hence, the
following does not apply to them.
An allocator manages multiple areas, called carriers, in which
memory blocks are placed. A carrier is either placed in a
separate memory segment (allocated via mseg_alloc
), or in
the heap segment (allocated via sys_alloc
). Multiblock
carriers are used for storage of several blocks. Singleblock
carriers are used for storage of one block. Blocks that are
larger than the value of the singleblock carrier threshold
(sbct) parameter are placed
in singleblock carriers. Blocks that are smaller than the value
of the sbct
parameter are placed in multiblock
carriers. Normally an allocator creates a "main multiblock
carrier". Main multiblock carriers are never deallocated. The
size of the main multiblock carrier is determined by the value
of the mmbcs parameter.
Sizes of multiblock carriers
allocated via mseg_alloc
are
decided based on the values of the largest multiblock carrier
size (lmbcs), the smallest
multiblock carrier size (smbcs),
and the multiblock carrier growth stages
(mbcgs) parameters. If
nc
is the current number of multiblock carriers (the main
multiblock carrier excluded) managed by an allocator, the size
of the next mseg_alloc
multiblock carrier allocated by
this allocator will roughly be
smbcs+nc*(lmbcs-smbcs)/mbcgs
when
nc <= mbcgs
,
and lmbcs
when nc > mbcgs
. If the value of the
sbct
parameter should be larger than the value of the
lmbcs
parameter, the allocator may have to create
multiblock carriers that are larger than the value of the
lmbcs
parameter, though.
Singleblock carriers allocated via mseg_alloc
are sized
to whole pages.
Sizes of carriers allocated via sys_alloc
are
decided based on the value of the sys_alloc
carrier size
(ycs) parameter. The size of
a carrier is the least number of multiples of the value of the
ycs
parameter that satisfies the request.
Coalescing of free blocks are always performed immediately. Boundary tags (headers and footers) in free blocks are used which makes the time complexity for coalescing constant.
The memory allocation strategy used for multiblock carriers by an allocator is configurable via the as parameter. Currently the following strategies are available:
Strategy: Find the smallest block that satisfies the requested block size.
Implementation: A balanced binary search tree is used. The time complexity is proportional to log N, where N is the number of sizes of free blocks.
Strategy: Find the smallest block that satisfies the requested block size. If multiple blocks are found, choose the one with the lowest address.
Implementation: A balanced binary search tree is used. The time complexity is proportional to log N, where N is the number of free blocks.
Strategy: Find the block with the lowest address that satisfies the requested block size.
Implementation: A balanced binary search tree is used. The time complexity is proportional to log N, where N is the number of free blocks.
Strategy: Find the carrier with the lowest address that can satisfy the requested block size, then find a block within that carrier using the "best fit" strategy.
Implementation: Balanced binary search trees are used. The time complexity is proportional to log N, where N is the number of free blocks.
Strategy: Find the carrier with the lowest address that can satisfy the requested block size, then find a block within that carrier using the "adress order best fit" strategy.
Implementation: Balanced binary search trees are used. The time complexity is proportional to log N, where N is the number of free blocks.
Strategy: Try to find the best fit, but settle for the best fit found during a limited search.
Implementation: The implementation uses segregated free lists with a maximum block search depth (in each list) in order to find a good fit fast. When the maximum block search depth is small (by default 3) this implementation has a time complexity that is constant. The maximum block search depth is configurable via the mbsd parameter.
Strategy: Do not search for a fit, inspect only one free block to see if it satisfies the request. This strategy is only intended to be used for temporary allocations.
Implementation: Inspect the first block in a free-list. If it satisfies the request, it is used; otherwise, a new carrier is created. The implementation has a time complexity that is constant.
As of erts version 5.6.1 the emulator will refuse to
use this strategy on other allocators than temp_alloc
.
This since it will only cause problems for other allocators.
Apart from the ordinary allocators described above a number of pre-allocators are used for some specific data types. These pre-allocators pre-allocate a fixed amount of memory for certain data types when the run-time system starts. As long as pre-allocated memory is available, it will be used. When no pre-allocated memory is available, memory will be allocated in ordinary allocators. These pre-allocators are typically much faster than the ordinary allocators, but can only satisfy a limited amount of requests.
System Flags Effecting erts_alloc
Warning!
Only use these flags if you are absolutely sure what you are doing. Unsuitable settings may cause serious performance degradation and even a system crash at any time during operation.
Memory allocator system flags have the following syntax:
+M<S><P> <V>
where <S>
is a letter identifying a subsystem,
<P>
is a parameter, and <V>
is the
value to use. The flags can be passed to the Erlang emulator
(erl) as command line
arguments.
System flags effecting specific allocators have an upper-case
letter as <S>
. The following letters are used for
the currently present allocators:
B: binary_alloc
D: std_alloc
E: ets_alloc
F: fix_alloc
H: eheap_alloc
I: literal_alloc
L: ll_alloc
M: mseg_alloc
R: driver_alloc
S: sl_alloc
T: temp_alloc
X: exec_alloc
Y: sys_alloc
The following flags are available for configuration of
mseg_alloc
:
+MMamcbf <size>
+MMrmcbf <ratio>
+MMsco true|false
true
. When a super carrier is used and this
flag is true
, mseg_alloc
will only create carriers
in the super carrier. Note that the alloc_util
framework may
create sys_alloc
carriers, so if you want all carriers to
be created in the super carrier, you therefore want to disable use
of sys_alloc
carriers by also passing
+Musac false
. When the flag
is false
, mseg_alloc
will try to create carriers outside
of the super carrier when the super carrier is full.
NOTE: Setting this flag to
false
may not be supported
on all systems. This flag will in that case be ignored.
NOTE: The super carrier cannot be enabled nor disabled on halfword heap systems. This flag will be ignored on halfword heap systems.
+MMscrfsd <amount>
65536
.
This parameter determines the amount of memory to reserve for
free segment descriptors used by the super carrier. If the system
runs out of reserved memory for free segment descriptors, other
memory will be used. This may however cause fragmentation issues,
so you want to ensure that this never happens. The maximum amount
of free segment descriptors used can be retrieved from the
erts_mmap
tuple part of the result from calling
erlang:system_info({allocator, mseg_alloc}).
+MMscrpm true|false
true
. When this flag is
true
, physical memory will be reserved for the whole super
carrier at once when it is created. The reservation will after that
be left unchanged. When this flag is set to false
only virtual
address space will be reserved for the super carrier upon creation.
The system will attempt to reserve physical memory upon carrier
creations in the super carrier, and attempt to unreserve physical
memory upon carrier destructions in the super carrier.
NOTE: What reservation of physical memory actually means highly depends on the operating system, and how it is configured. For example, different memory overcommit settings on Linux drastically change the behaviour. Also note, setting this flag to
false
may not be supported on all systems. This flag will in that case
be ignored.
NOTE: The super carrier cannot be enabled nor disabled on halfword heap systems. This flag will be ignored on halfword heap systems.
+MMscs <size in MB>
mseg_alloc
will always try to create new carriers in the super
carrier if it exists. Note that the alloc_util
framework may
create sys_alloc
carriers. For more information on this, see the
documentation of the +MMsco
flag.
NOTE: The super carrier cannot be enabled nor disabled on halfword heap systems. This flag will be ignored on halfword heap systems.
+MMmcs <amount>
The following flags are available for configuration of
sys_alloc
:
+MYe true
sys_alloc
. Note: sys_alloc
cannot be disabled.+MYm libc
malloc
library to use. Currently only
libc
is available. libc
enables the standard
libc
malloc implementation. By default libc
is used.+MYtt <size>
sbrk
) that will be kept by malloc
(not
released to the operating system). When the amount of free
memory at the top of the heap exceeds the trim threshold,
malloc
will release it (by calling
sbrk
). Trim threshold is given in kilobytes. Default
trim threshold is 128. Note: This flag will
only have any effect when the emulator has been linked with
the GNU C library, and uses its malloc
implementation.+MYtp <size>
malloc
when
sbrk
is called to get more memory from the operating
system. Default top pad size is 0. Note: This flag
will only have any effect when the emulator has been linked
with the GNU C library, and uses its malloc
implementation.The following flags are available for configuration of allocators
based on alloc_util
. If u
is used as subsystem
identifier (i.e., <S> = u
) all allocators based on
alloc_util
will be effected. If B
, D
, E
,
F
, H
, L
, R
, S
, or T
is used as
subsystem identifier, only the specific allocator identified will be
effected:
+M<S>acul <utilization>|de
<utilization>
is an integer in the range
[0, 100]
representing utilization in percent. When a
utilization value larger than zero is used, allocator instances
are allowed to abandon multiblock carriers. If de
(default
enabled) is passed instead of a <utilization>
,
a recomended non zero utilization value will be used. The actual
value chosen depend on allocator type and may be changed between
ERTS versions. Currently the default equals de
, but this
may be changed in the future. Carriers will be abandoned when
memory utilization in the allocator instance falls below the
utilization value used. Once a carrier has been abandoned, no new
allocations will be made in it. When an allocator instance gets an
increased multiblock carrier need, it will first try to fetch an
abandoned carrier from an allocator instances of the same
allocator type. If no abandoned carrier could be fetched, it will
create a new empty carrier. When an abandoned carrier has been
fetched it will function as an ordinary carrier. This feature has
special requirements on the
allocation strategy used. Currently
only the strategies aoff
, aoffcbf
and aoffcaobf
support
abandoned carriers. This feature also requires
multiple thread specific instances
to be enabled. When enabling this feature, multiple thread specific
instances will be enabled if not already enabled, and the
aoffcbf
strategy will be enabled if current strategy does not
support abandoned carriers. This feature can be enabled on all
allocators based on the alloc_util
framework with the
exception of temp_alloc
(which would be pointless).
+M<S>as bf|aobf|aoff|aoffcbf|aoffcaobf|gf|af
bf
(best fit),
aobf
(address order best fit), aoff
(address order first fit),
aoffcbf
(address order first fit carrier best fit),
aoffcaobf
(address order first fit carrier address order best fit),
gf
(good fit), and af
(a fit). See
the description of allocation strategies in "the alloc_util
framework" section.+M<S>asbcst <size>
mseg_alloc
singleblock carrier is shrunk, the carrier
will be left unchanged if the amount of unused memory is less
than this threshold; otherwise, the carrier will be shrunk.
See also rsbcst.+M<S>e true|false
<S>
.+M<S>lmbcs <size>
mseg_alloc
) multiblock carrier size (in
kilobytes). See the description
on how sizes for mseg_alloc multiblock carriers are decided
in "the alloc_util
framework" section. On 32-bit Unix style OS
this limit can not be set higher than 128 megabyte.+M<S>mbcgs <ratio>
mseg_alloc
) multiblock carrier growth stages. See
the description on how sizes for
mseg_alloc multiblock carriers are decided
in "the alloc_util
framework" section.+M<S>mbsd <depth>
<S>
. When the good fit strategy is used, free
blocks are placed in segregated free-lists. Each free list
contains blocks of sizes in a specific range. The max block
search depth sets a limit on the maximum number of blocks to
inspect in a free list during a search for suitable block
satisfying the request.+M<S>mmbcs <size>
<S>
. The main
multiblock carrier is allocated via sys_alloc
and is
never deallocated.+M<S>mmmbc <amount>
mseg_alloc
multiblock carriers. Maximum number of
multiblock carriers allocated via mseg_alloc
by
allocator <S>
. When this limit has been reached,
new multiblock carriers will be allocated via
sys_alloc
.+M<S>mmsbc <amount>
mseg_alloc
singleblock carriers. Maximum number of
singleblock carriers allocated via mseg_alloc
by
allocator <S>
. When this limit has been reached,
new singleblock carriers will be allocated via
sys_alloc
.+M<S>ramv <bool>
+M<S>rmbcmt <ratio>
+M<S>rsbcmt <ratio>
+M<S>rsbcst <ratio>
mseg_alloc
singleblock carrier is shrunk, the carrier will be left
unchanged if the ratio of unused memory is less than this
threshold; otherwise, the carrier will be shrunk.
See also asbcst.+M<S>sbct <size>
+M<S>smbcs <size>
mseg_alloc
) multiblock carrier size (in
kilobytes). See the description
on how sizes for mseg_alloc multiblock carriers are decided
in "the alloc_util
framework" section.+M<S>t true|false
Multiple, thread specific instances of the allocator.
This option will only have any effect on the runtime system
with SMP support. Default behaviour on the runtime system with
SMP support is NoSchedulers+1
instances. Each scheduler will use
a lock-free instance of its own and other threads will use
a common instance.
It was previously (before ERTS version 5.9) possible to configure a smaller amount of thread specific instances than schedulers. This is, however, not possible any more.
Currently the following flags are available for configuration of
alloc_util
, i.e. all allocators based on alloc_util
will be effected:
+Muycs <size>
sys_alloc
carrier size. Carriers allocated via
sys_alloc
will be allocated in sizes which are
multiples of the sys_alloc
carrier size. This is not
true for main multiblock carriers and carriers allocated
during a memory shortage, though.+Mummc <amount>
mseg_alloc
carriers. Maximum number of carriers
placed in separate memory segments. When this limit has been
reached, new carriers will be placed in memory retrieved from
sys_alloc
.+Musac <bool>
sys_alloc
carriers. By default true
. If
set to false
, sys_alloc
carriers will never be
created by allocators using the alloc_util
framework.The following flag is special for literal_alloc
:
+MIscs <size in MB>
literal_alloc
super carrier size (in MB). The amount of
virtual address space reserved for literal terms in
Erlang code on 64-bit architectures. The default is 1024 (1GB)
and is usually sufficient. The flag is ignored on 32-bit
architectures.The following flag is special for exec_alloc
:
+MXscs <size in MB>
exec_alloc
super carrier size (in MB). The amount of
virtual address space reserved for native executable code
used by hipe on specific architectures (x86_64). The default is 512 MB.
Instrumentation flags:
+Mim true|false
instrument
module. +Mim true
implies +Mis true
.
+Mim true
is the same as
-instr.+Mis true|false
instrument
module.+Mit X
Note!
When instrumentation of the emulator is enabled, the emulator uses more memory and runs slower.
Other flags:
+Mea min|max|r9c|r10b|r11b|config
min
max
r9c|r10b|r11b
config
erts_alloc_config
, not when using the created
configuration.
+Mlpm all|no
no
, i.e.,
no physical memory will be locked. If set to all
, all
memory mappings made by the runtime system, will be locked into
physical memory. If set to all
, the runtime system will fail
to start if this feature is not supported, the user has not got enough
privileges, or the user is not allowed to lock enough physical memory.
The runtime system will also fail with an out of memory condition
if the user limit on the amount of locked memory is reached.
Only some default values have been presented here. erlang:system_info(allocator), and erlang:system_info({allocator, Alloc}) can be used in order to obtain currently used settings and current status of the allocators.
Note!
Most of these flags are highly implementation dependent, and they may be changed or removed without prior notice.
erts_alloc
is not obliged to strictly use the settings that
have been passed to it (it may even ignore them).
erts_alloc_config(3)
is a tool that can be used to aid creation of an
erts_alloc
configuration that is suitable for a limited
number of runtime scenarios.