-
+a size
-
Suggested stack size, in kilowords, for threads in the
async thread pool. Valid range is 16-8192 kilowords. The
default suggested stack size is 16 kilowords, that is, 64
kilobyte on 32-bit architectures. This small default size
has been chosen because the number of async threads can
be large. The default size is enough for drivers
delivered with Erlang/OTP, but might not be large
enough for other dynamically linked-in drivers that use the
driver_async() functionality.
Notice that the value passed is only a suggestion,
and it can even be ignored on some platforms.
- +A size
-
Sets the number of threads in async thread pool. Valid range
is 1-1024. The async thread pool is used by linked-in drivers to
handle work that may take a very long time. Since OTP 21 there are
very few linked-in drivers in the default Erlang/OTP distribution
that uses the async thread pool. Most of them have been migrated to
dirty IO schedulers. Defaults to 1.
- +B [c | d | i]
-
Option c makes Ctrl-C
interrupt the current shell instead of invoking the emulator break
handler. Option d (same as specifying
+B without an extra option) disables the break
handler. Option i makes the emulator ignore any
break signal.
If option c is used with
oldshell on Unix, Ctrl-C will
restart the shell process rather than interrupt it.
Notice that on Windows, this flag is only applicable for
werl, not erl
(oldshell). Notice also that
Ctrl-Break is used instead of
Ctrl-C on Windows.
- +c true | false
-
Enables or disables
time
correction:
- true
- Enables time correction. This is the default if
time correction is supported on the specific platform.
- false
- Disables time correction.
For backward compatibility, the boolean value can be omitted.
This is interpreted as +c false.
- +C no_time_warp | single_time_warp |
multi_time_warp
-
Sets time warp
mode:
- no_time_warp
-
No time warp mode (the default)
- single_time_warp
-
Single time warp mode
- multi_time_warp
-
Multi-time warp mode
- +d
-
If the emulator detects an internal error (or runs out of memory),
it, by default, generates both a crash dump and a core dump.
The core dump is, however, not very useful as the content
of process heaps is destroyed by the crash dump generation.
Option +d instructs the emulator to produce only a
core dump and no crash dump if an internal error is detected.
Calling
erlang:halt/1 with a string argument still
produces a crash dump. On Unix systems, sending an emulator process
a SIGUSR1 signal also forces a crash dump.
- +dcg DecentralizedCounterGroupsLimit
-
Limits the number of decentralized counter groups used by
decentralized counters optimized for update operations in the
Erlang runtime system. By default, the limit is 256.
When the number of schedulers is less than or equal to the
limit, each scheduler has its own group. When the
number of schedulers is larger than the groups limit,
schedulers share groups. Shared groups degrade
the performance for updating counters while many reader groups
degrade the performance for reading counters. So, the limit is a tradeoff
between performance for update operations and performance for
read operations. Each group consumes 64 bytes in each
counter.
Notice that a runtime system using decentralized
counter groups benefits from binding
schedulers to logical processors, as the groups are
distributed better between schedulers with this option.
This option only affects decentralized counters used for
the counters that are keeping track of the memory consumption
and the number of terms in ETS tables of type ordered_set with
the write_concurrency option activated.
- +e Number
-
Sets the maximum number of ETS tables. This limit is
partially obsolete.
- +ec
-
Forces option compressed on all ETS tables.
Only intended for test and evaluation.
-
+fnl
-
The virtual machine works with filenames as if they are encoded
using the ISO Latin-1 encoding, disallowing Unicode characters with
code points > 255.
For more information about Unicode filenames, see section
Unicode
Filenames in the STDLIB User's Guide. Notice that
this value also applies to command-line parameters and environment
variables (see section
Unicode in Environment and Parameters in the STDLIB
User's Guide).
- +fnu[{w|i|e}]
-
The virtual machine works with filenames as if they are encoded
using UTF-8 (or some other system-specific Unicode encoding). This is
the default on operating systems that enforce Unicode encoding, that
is, Windows MacOS X and Android.
The +fnu switch can be followed by w, i, or
e to control how wrongly encoded filenames are to be
reported:
-
w means that a warning is sent to the error_logger
whenever a wrongly encoded filename is "skipped" in directory
listings. This is the default.
-
i means that those wrongly encoded filenames are silently
ignored.
-
e means that the API function returns an error whenever a
wrongly encoded filename (or directory name) is encountered.
Notice that
file:read_link/1 always returns an error if the link
points to an invalid filename.
For more information about Unicode filenames, see section
Unicode
Filenames in the STDLIB User's Guide. Notice that
this value also applies to command-line parameters and environment
variables (see section
Unicode in Environment and Parameters in the STDLIB
User's Guide).
- +fna[{w|i|e}]
-
Selection between +fnl and +fnu is done based
on the current locale settings in the OS. This means that if you
have set your terminal for UTF-8 encoding, the filesystem is
expected to use the same encoding for filenames. This is the default
on all operating systems, except Android, MacOS X and Windows.
The +fna switch can be followed by w, i, or
e. This has effect if the locale settings cause the behavior
of +fnu to be selected; see the description of +fnu
above. If the locale settings cause the behavior of +fnl to be
selected, then w, i, or e have no effect.
For more information about Unicode filenames, see section
Unicode
Filenames in the STDLIB User's Guide. Notice that
this value also applies to command-line parameters and environment
variables (see section
Unicode in Environment and Parameters in the STDLIB
User's Guide).
- +hms Size
-
Sets the default heap size of processes to the size
Size.
- +hmbs Size
-
Sets the default binary virtual heap size of processes to the size
Size.
- +hmax Size
-
Sets the default maximum heap size of processes to the size
Size words. Defaults to 0, which means that no
maximum heap size is used. For more information, see
process_flag(max_heap_size, MaxHeapSize).
- +hmaxel true|false
-
Sets whether to send an error logger message or not for processes
reaching the maximum heap size. Defaults to true.
For more information, see
process_flag(max_heap_size, MaxHeapSize).
- +hmaxib true|false
-
Sets whether to include the size of shared off-heap binaries
in the sum compared against the maximum heap size. Defaults to
false. For more information, see
process_flag(max_heap_size, MaxHeapSize).
- +hmaxk true|false
-
Sets whether to kill processes reaching the maximum heap size or not.
Default to true. For more information, see
process_flag(max_heap_size, MaxHeapSize).
- +hpds Size
-
Sets the initial process dictionary size of processes to the size
Size.
- +hmqd off_heap|on_heap
-
Sets the default value of the message_queue_data process flag.
Defaults to on_heap. If +hmqd is not
passed, on_heap will be the default. For more information, see
process_flag(message_queue_data, MQD).
- +IOp PollSets
-
Sets the number of IO pollsets to use when polling for I/O.
This option is only used on platforms that support concurrent
updates of a pollset, otherwise the same number of pollsets
are used as IO poll threads.
The default is 1.
- +IOt PollThreads
-
Sets the number of IO poll threads to use when polling for I/O.
The maximum number of poll threads allowed is 1024. The default is 1.
A good way to check if more IO poll threads are needed is to use
microstate accounting
and see what the load of the IO poll thread is. If it is high it could
be a good idea to add more threads.
- +IOPp PollSetsPercentage
-
Similar to +IOp but uses
percentages to set the number of IO pollsets to create, based on the
number of poll threads configured. If both +IOPp and +IOp
are used, +IOPp is ignored.
- +IOPt PollThreadsPercentage
-
Similar to +IOt but uses
percentages to set the number of IO poll threads to create, based on
the number of schedulers configured. If both +IOPt and
+IOt are used, +IOPt is ignored.
- +IOs true|false
-
Enable or disable scheduler thread poll optimization. Default is
true.
If enabled, file descriptors that are frequently read may be moved to
a special pollset used by scheduler threads. The objective is to
reduce the number of system calls and thereby CPU load, but it can
in some cases increase scheduling latency for individual file
descriptor input events.
- +JPperf true|false|dump|map|fp|no_fp
-
Enables or disables support for the perf profiler when
running with the JIT on Linux. Defaults to false.
This option can be combined multiple times to enable several
options:
- dump
- Gives perf detailed line information, so that the
perf annotate feature works.
- map
- Gives perf a map over all module code, letting it
translate machine code addresses to Erlang source code
locations. This also enables frame pointers for Erlang code so that
perf can walk the call stacks of Erlang processes, which
costs one extra word per stack frame.
- fp
- Enables frame pointers independently of the map
option.
- no_fp
- Disables the frame pointers added by the map
option.
- true
- Enables map and dump.
- false
- Disables all other options.
For more details about how to run perf see the
perf support
section in the BeamAsm internal documentation.
- +JMsingle true|false
-
Enables or disables the use of single-mapped RWX memory for JIT
code. The default is to map JIT:ed machine code into two
regions sharing the same physical pages, where one region is
executable but not writable, and the other writable but not
executable. As some tools, such as QEMU user mode emulation,
cannot deal with the dual mapping, this flags allows it to be
disabled. This flag is automatically enabled by the
+JPperf flag.
- +L
-
Prevents loading information about source filenames and line
numbers. This saves some memory, but exceptions do not contain
information about the filenames and line numbers.
- +MFlag Value
-
Memory allocator-specific flags. For more information, see
erts_alloc(3).
- +pad true|false
-
Since: OTP 25.3
The boolean value used with the +pad parameter determines
the default value of the
async_dist process flag of newly spawned processes.
By default, if no +pad command line option is passed,
the async_dist flag will be set to false.
The value used in runtime can be inspected by calling
erlang:system_info(async_dist).
-
+pc Range
-
Sets the range of characters that the system considers printable in
heuristic detection of strings. This typically affects the shell,
debugger, and io:format functions (when ~tp is used in
the format string).
Two values are supported for Range:
- latin1
- The default. Only characters in the ISO Latin-1 range can be
considered printable. This means that a character with a code point
> 255 is never considered printable and that lists containing
such characters are displayed as lists of integers rather than text
strings by tools.
- unicode
- All printable Unicode characters are considered when
determining if a list of integers is to be displayed in
string syntax. This can give unexpected results if, for
example, your font does not cover all Unicode characters.
See also
io:printable_range/0 in STDLIB.
- +P Number
-
Sets the maximum number of simultaneously existing processes for this
system if a Number is passed as value. Valid range for
Number is [1024-134217727]
NOTE: The actual maximum chosen may be much larger than
the Number passed. Currently the runtime system often,
but not always, chooses a value that is a power of 2. This might,
however, be changed in the future. The actual value chosen can be
checked by calling
erlang:system_info(process_limit).
The default value is 262144
- +Q Number
-
Sets the maximum number of simultaneously existing ports for this
system if a Number is passed as value. Valid range for Number
is [1024-134217727]
NOTE: The actual maximum chosen may be much larger than
the actual Number passed. Currently the runtime system often,
but not always, chooses a value that is a power of 2. This might,
however, be changed in the future. The actual value chosen can be
checked by calling
erlang:system_info(port_limit).
The default value used is normally 65536. However, if
the runtime system is able to determine maximum amount of file
descriptors that it is allowed to open and this value is larger
than 65536, the chosen value will increased to a value
larger or equal to the maximum amount of file descriptors that
can be opened.
On Windows the default value is set to 8196 because the
normal OS limitations are set higher than most machines can handle.
- +R ReleaseNumber
-
Sets the compatibility mode.
The distribution mechanism is not backward compatible by
default. This flag sets the emulator in compatibility mode
with an earlier Erlang/OTP release ReleaseNumber.
The release number must be in the range
<current release>-2..<current release>. This
limits the emulator, making it possible for it to communicate
with Erlang nodes (as well as C- and Java nodes) running that
earlier release.
Note
Ensure that all nodes (Erlang-, C-, and Java nodes) of
a distributed Erlang system is of the same Erlang/OTP release,
or from two different Erlang/OTP releases X and Y, where
all Y nodes have compatibility mode X.
- +r
-
Forces ETS memory block to be moved on realloc.
- +rg ReaderGroupsLimit
-
Limits the number of reader groups used by read/write locks
optimized for read operations in the Erlang runtime system. By
default the reader groups limit is 64.
When the number of schedulers is less than or equal to the reader
groups limit, each scheduler has its own reader group. When the
number of schedulers is larger than the reader groups limit,
schedulers share reader groups. Shared reader groups degrade
read lock and read unlock performance while many
reader groups degrade write lock performance. So, the limit is a
tradeoff between performance for read operations and performance
for write operations. Each reader group consumes 64 byte
in each read/write lock.
Notice that a runtime system using shared reader groups benefits from
binding schedulers to logical
processors, as the reader groups are distributed better
between schedulers.
-
+S Schedulers:SchedulerOnline
-
Sets the number of scheduler threads to create and scheduler threads
to set online. The maximum for both values is 1024. If the Erlang
runtime system is able to determine the number of logical processors
configured and logical processors available, Schedulers
defaults to logical processors configured, and
SchedulersOnline defaults to logical processors available;
otherwise the default values are 1. If the emulator detects that it
is subject to a CPU
quota, the default value for SchedulersOnline will
be limited accordingly.
Schedulers can be omitted if :SchedulerOnline is not
and conversely. The number of schedulers online can be changed at
runtime through
erlang:system_flag(schedulers_online,
SchedulersOnline).
If Schedulers or SchedulersOnline is specified as a
negative number, the value is subtracted from the default number of
logical processors configured or logical processors available,
respectively.
Specifying value 0 for Schedulers or
SchedulersOnline resets the number of scheduler threads or
scheduler threads online, respectively, to its default value.
- +SP
SchedulersPercentage:SchedulersOnlinePercentage
-
Similar to +S but uses
percentages to set the number of scheduler threads to create, based
on logical processors configured, and scheduler threads to set online,
based on logical processors available.
Specified values must be > 0. For example,
+SP 50:25 sets the number of scheduler threads to 50% of the
logical processors configured, and the number of scheduler threads
online to 25% of the logical processors available.
SchedulersPercentage can be omitted if
:SchedulersOnlinePercentage is not and conversely. The number
of schedulers online can be changed at runtime through
erlang:system_flag(schedulers_online,
SchedulersOnline).
This option interacts with +S
settings. For example, on a system with 8 logical cores configured
and 8 logical cores available, the combination of the options
+S 4:4 +SP 50:25 (in either order) results in 2 scheduler
threads (50% of 4) and 1 scheduler thread online (25% of 4).
- +SDcpu
DirtyCPUSchedulers:DirtyCPUSchedulersOnline
-
Sets the number of dirty CPU scheduler threads to create and dirty
CPU scheduler threads to set online.
The maximum for both values is 1024, and each value is
further limited by the settings for normal schedulers:
- The number of dirty CPU scheduler threads created cannot exceed
the number of normal scheduler threads created.
- The number of dirty CPU scheduler threads online cannot exceed
the number of normal scheduler threads online.
For details, see the +S and
+SP. By default, the number
of dirty CPU scheduler threads created equals the number of normal
scheduler threads created, and the number of dirty CPU scheduler
threads online equals the number of normal scheduler threads online.
DirtyCPUSchedulers can be omitted if
:DirtyCPUSchedulersOnline is not and conversely. The number of
dirty CPU schedulers online can be changed at runtime through
erlang:system_flag(dirty_cpu_schedulers_online,
DirtyCPUSchedulersOnline).
The amount of dirty CPU schedulers is limited by the amount of
normal schedulers in order to limit the effect on processes
executing on ordinary schedulers. If the amount of dirty CPU
schedulers was allowed to be unlimited, dirty CPU bound jobs would
potentially starve normal jobs.
Typical users of the dirty CPU schedulers are large garbage collections,
json protocol encode/decoders written as nifs and matrix manipulation
libraries.
You can use msacc(3)
in order to see the current load of the dirty CPU schedulers threads
and adjust the number used accordingly.
- +SDPcpu
DirtyCPUSchedulersPercentage:DirtyCPUSchedulersOnlinePercentage
-
Similar to +SDcpu but
uses percentages to set the number of dirty CPU scheduler threads to
create and the number of dirty CPU scheduler threads to set online.
Specified values must be
> 0. For example, +SDPcpu 50:25 sets the number of dirty
CPU scheduler threads to 50% of the logical processors configured
and the number of dirty CPU scheduler threads online to 25% of the
logical processors available. DirtyCPUSchedulersPercentage can
be omitted if :DirtyCPUSchedulersOnlinePercentage is not and
conversely. The number of dirty CPU schedulers online can be changed
at runtime through
erlang:system_flag(dirty_cpu_schedulers_online,
DirtyCPUSchedulersOnline).
This option interacts with +SDcpu settings. For example, on a
system with 8 logical cores configured and 8 logical cores available,
the combination of the options +SDcpu 4:4 +SDPcpu 50:25 (in
either order) results in 2 dirty CPU scheduler threads (50% of 4) and
1 dirty CPU scheduler thread online (25% of 4).
- +SDio DirtyIOSchedulers
-
Sets the number of dirty I/O scheduler threads to create.
Valid range is 1-1024. By
default, the number of dirty I/O scheduler threads created is 10.
The amount of dirty IO schedulers is not limited by the amount of
normal schedulers like the amount of
dirty CPU schedulers. This since only I/O bound work is
expected to execute on dirty I/O schedulers. If the user should schedule CPU
bound jobs on dirty I/O schedulers, these jobs might starve ordinary
jobs executing on ordinary schedulers.
Typical users of the dirty IO schedulers are reading and writing to files.
You can use msacc(3)
in order to see the current load of the dirty IO schedulers threads
and adjust the number used accordingly.
- +sFlag Value
-
Scheduling specific flags.
- +sbt BindType
-
Sets scheduler bind type.
Schedulers can also be bound using flag
+stbt. The only
difference between these two flags is how the following errors
are handled:
- Binding of schedulers is not supported on the specific
platform.
- No available CPU topology. That is, the runtime system was
not able to detect the CPU topology automatically, and no
user-defined CPU topology
was set.
If any of these errors occur when +sbt has been passed,
the runtime system prints an error message, and refuses to
start. If any of these errors occur when +stbt has been
passed, the runtime system silently ignores the error, and
start up using unbound schedulers.
Valid BindTypes:
- u
-
unbound - Schedulers are not bound to logical
processors, that is, the operating system decides where the
scheduler threads execute, and when to migrate them. This is
the default.
- ns
-
no_spread - Schedulers with close scheduler
identifiers are bound as close as possible in hardware.
- ts
-
thread_spread - Thread refers to hardware threads
(such as Intel's hyper-threads). Schedulers with low scheduler
identifiers, are bound to the first hardware thread of
each core, then schedulers with higher scheduler identifiers
are bound to the second hardware thread of each core,and so on.
- ps
-
processor_spread - Schedulers are spread like
thread_spread, but also over physical processor chips.
- s
-
spread - Schedulers are spread as much as possible.
- nnts
-
no_node_thread_spread - Like thread_spread,
but if multiple Non-Uniform Memory Access (NUMA) nodes exist,
schedulers are spread over one NUMA node at a time,
that is, all logical processors of one NUMA node are bound
to schedulers in sequence.
- nnps
-
no_node_processor_spread - Like
processor_spread, but if multiple NUMA nodes exist,
schedulers are spread over one NUMA node at a time, that is,
all logical processors of one NUMA node are bound to
schedulers in sequence.
- tnnps
-
thread_no_node_processor_spread - A combination of
thread_spread, and no_node_processor_spread.
Schedulers are spread over hardware threads across NUMA
nodes, but schedulers are only spread over processors
internally in one NUMA node at a time.
- db
-
default_bind - Binds schedulers the default way.
Defaults to thread_no_node_processor_spread
(which can change in the future).
Binding of schedulers is only supported on newer
Linux, Solaris, FreeBSD, and Windows systems.
If no CPU topology is available when flag +sbt
is processed and BindType is any other type than
u, the runtime system fails to start. CPU
topology can be defined using flag
+sct. Notice
that flag +sct can have to be passed before flag
+sbt on the command line (if no CPU topology
has been automatically detected).
The runtime system does by default not bind schedulers
to logical processors.
Note
If the Erlang runtime system is the only operating system
process that binds threads to logical processors, this
improves the performance of the runtime system. However,
if other operating system processes (for example
another Erlang runtime system) also bind threads to
logical processors, there can be a performance penalty
instead. This performance penalty can sometimes be
severe. If so, you are advised not to
bind the schedulers.
How schedulers are bound matters. For example, in
situations when there are fewer running processes than
schedulers online, the runtime system tries to migrate
processes to schedulers with low scheduler identifiers.
The more the schedulers are spread over the hardware,
the more resources are available to the runtime
system in such situations.
Note
If a scheduler fails to bind, this is
often silently ignored, as it is not always
possible to verify valid logical processor identifiers. If
an error is reported, it is reported to the
error_logger. If you want to verify that the
schedulers have bound as requested, call
erlang:system_info(scheduler_bindings).
-
+sbwt none|very_short|short|medium|long|very_long
-
Sets scheduler busy wait threshold. Defaults to medium.
The threshold determines how long schedulers are to busy
wait when running out of work before going to sleep.
Note
This flag can be removed or changed at any time
without prior notice.
-
+sbwtdcpu none|very_short|short|medium|long|very_long
-
As +sbwt but affects
dirty CPU schedulers. Defaults to short.
Note
This flag can be removed or changed at any time
without prior notice.
-
+sbwtdio none|very_short|short|medium|long|very_long
-
As +sbwt but affects
dirty IO schedulers. Defaults to short.
Note
This flag can be removed or changed at any time
without prior notice.
- +scl true|false
-
Enables or disables scheduler compaction of load. By default
scheduler compaction of load is enabled. When enabled, load
balancing strives for a load distribution, which causes
as many scheduler threads as possible to be fully loaded (that is,
not run out of work). This is accomplished by migrating load
(for example, runnable processes) into a smaller set of schedulers
when schedulers frequently run out of work. When disabled,
the frequency with which schedulers run out of work is
not taken into account by the load balancing logic.
+scl false is similar to
+sub true, but
+sub true also balances scheduler utilization
between schedulers.
- +sct CpuTopology
-
-
<Id> = integer(); when 0 =< <Id> =< 65535
- <IdRange> = <Id>-<Id>
- <IdOrIdRange> = <Id> | <IdRange>
- <IdList> = <IdOrIdRange>,<IdOrIdRange> |
<IdOrIdRange>
- <LogicalIds> = L<IdList>
-
<ThreadIds> = T<IdList> | t<IdList>
- <CoreIds> = C<IdList> | c<IdList>
-
<ProcessorIds> = P<IdList> | p<IdList>
- <NodeIds> = N<IdList> | n<IdList>
-
<IdDefs> =
<LogicalIds><ThreadIds><CoreIds><ProcessorIds><NodeIds> |
<LogicalIds><ThreadIds><CoreIds><NodeIds><ProcessorIds>
- CpuTopology = <IdDefs>:<IdDefs> |
<IdDefs>
Sets a user-defined CPU topology. The user-defined
CPU topology overrides any automatically detected
CPU topology. The CPU topology is used when
binding schedulers to logical
processors.
Uppercase letters signify real identifiers and lowercase
letters signify fake identifiers only used for description
of the topology. Identifiers passed as real identifiers can
be used by the runtime system when trying to access specific
hardware; if they are incorrect the behavior is
undefined. Faked logical CPU identifiers are not accepted,
as there is no point in defining the CPU topology without
real logical CPU identifiers. Thread, core, processor, and
node identifiers can be omitted. If omitted, the thread ID
defaults to t0, the core ID defaults to c0,
the processor ID defaults to p0, and the node ID is
left undefined. Either each logical processor must
belong to only one NUMA node, or no logical
processors must belong to any NUMA nodes.
Both increasing and decreasing <IdRange>s
are allowed.
NUMA node identifiers are system wide. That is, each NUMA
node on the system must have a unique identifier. Processor
identifiers are also system wide. Core identifiers are
processor wide. Thread identifiers are core wide.
The order of the identifier types implies the hierarchy of the
CPU topology. The valid orders are as follows:
-
<LogicalIds><ThreadIds><CoreIds><ProcessorIds><NodeIds>,
that is, thread is part of a core that is part of a processor,
which is part of a NUMA node.
-
<LogicalIds><ThreadIds><CoreIds><NodeIds><ProcessorIds>,
that is, thread is part of a core that is part of a NUMA node,
which is part of a processor.
A CPU topology can consist of both processor external, and
processor internal NUMA nodes as long as each logical processor
belongs to only one NUMA node. If
<ProcessorIds> is omitted, its default position
is before <NodeIds>. That is, the default is
processor external NUMA nodes.
If a list of identifiers is used in an
<IdDefs>:
-
<LogicalIds> must be a list
of identifiers.
- At least one other identifier type besides
<LogicalIds> must also have a
list of identifiers.
- All lists of identifiers must produce the
same number of identifiers.
A simple example. A single quad core processor can be
described as follows:
% erl +sct L0-3c0-3
1> erlang:system_info(cpu_topology).
[{processor,[{core,{logical,0}},
{core,{logical,1}},
{core,{logical,2}},
{core,{logical,3}}]}]A more complicated example with two quad core
processors, each processor in its own NUMA node.
The ordering of logical processors is a bit weird.
This to give a better example of identifier lists:
% erl +sct L0-1,3-2c0-3p0N0:L7,4,6-5c0-3p1N1
1> erlang:system_info(cpu_topology).
[{node,[{processor,[{core,{logical,0}},
{core,{logical,1}},
{core,{logical,3}},
{core,{logical,2}}]}]},
{node,[{processor,[{core,{logical,7}},
{core,{logical,4}},
{core,{logical,6}},
{core,{logical,5}}]}]}]As long as real identifiers are correct, it is OK
to pass a CPU topology that is not a correct
description of the CPU topology. When used with
care this can be very useful. This
to trick the emulator to bind its schedulers
as you want. For example, if you want to run multiple
Erlang runtime systems on the same machine, you
want to reduce the number of schedulers used and
manipulate the CPU topology so that they bind to
different logical CPUs. An example, with two Erlang
runtime systems on a quad core machine:
% erl +sct L0-3c0-3 +sbt db +S3:2 -detached -noinput -noshell -sname one
% erl +sct L3-0c0-3 +sbt db +S3:2 -detached -noinput -noshell -sname two
In this example, each runtime system have two
schedulers each online, and all schedulers online
will run on different cores. If we change to one
scheduler online on one runtime system, and three
schedulers online on the other, all schedulers
online will still run on different cores.
Notice that a faked CPU topology that does not reflect
how the real CPU topology looks like is likely to
decrease the performance of the runtime system.
For more information, see
erlang:system_info(cpu_topology).
- +ssrct
-
Skips reading CPU topology.
Note
Reading CPU topology slows down startup when starting many
parallel instances of ERTS on systems with large amount of
cores, using this flag might speed up execution in such
scenarios.
- +sfwi Interval
-
Sets scheduler-forced wakeup interval. All run queues are
scanned each Interval milliseconds. While there are
sleeping schedulers in the system, one scheduler is woken
for each non-empty run queue found. Interval default
to 0, meaning this feature is disabled.
Note
This feature has been introduced as a temporary workaround
for long-executing native code, and native code that does not
bump reductions properly in OTP. When these bugs have been
fixed, this flag will be removed.
- +spp Bool
-
Sets default scheduler hint for port parallelism. If set to
true, the virtual machine schedules port tasks when it
improves parallelism in the system. If set to false, the
virtual machine tries to perform port tasks immediately,
improving latency at the expense of parallelism. Default to
false. The default used can be inspected in runtime by
calling
erlang:system_info(port_parallelism).
The default can be overridden on port creation by passing option
parallelism to
erlang:open_port/2.
-
+sss size
-
Suggested stack size, in kilowords, for scheduler threads.
Valid range is 20-8192 kilowords. The default suggested
stack size is 128 kilowords.
-
+sssdcpu size
-
Suggested stack size, in kilowords, for dirty CPU scheduler
threads. Valid range is 20-8192 kilowords. The default
suggested stack size is 40 kilowords.
-
+sssdio size
-
Suggested stack size, in kilowords, for dirty IO scheduler
threads. Valid range is 20-8192 kilowords. The default
suggested stack size is 40 kilowords.
- +stbt BindType
-
Tries to set the scheduler bind type. The same as flag
+sbt except
how some errors are handled. For more information, see
+sbt.
- +sub true|false
-
Enables or disables
scheduler utilization balancing of load. By default
scheduler utilization balancing is disabled and instead scheduler
compaction of load is enabled, which strives for a load
distribution that causes as many scheduler threads as possible
to be fully loaded (that is, not run out of work). When scheduler
utilization balancing is enabled, the system instead tries to
balance scheduler utilization between schedulers. That is,
strive for equal scheduler utilization on all schedulers.
+sub true is only supported on systems where the runtime
system detects and uses a monotonically increasing high-resolution
clock. On other systems, the runtime system fails to start.
+sub true implies
+scl false. The difference between
+sub true and +scl false is that +scl false
does not try to balance the scheduler utilization.
-
+swct very_eager|eager|medium|lazy|very_lazy
-
Sets scheduler wake cleanup threshold. Defaults to medium.
Controls how eager schedulers are to be requesting
wakeup because of certain cleanup operations. When a lazy setting
is used, more outstanding cleanup operations can be left undone
while a scheduler is idling. When an eager setting is used,
schedulers are more frequently woken, potentially increasing
CPU-utilization.
Note
This flag can be removed or changed at any time without prior
notice.
- +sws default|legacy
-
Sets scheduler wakeup strategy. Default strategy changed in
ERTS 5.10 (Erlang/OTP R16A). This strategy was known as
proposal in Erlang/OTP R15. The legacy strategy
was used as default from R13 up to and including R15.
Note
This flag can be removed or changed at any time without prior
notice.
-
+swt very_low|low|medium|high|very_high
-
Sets scheduler wakeup threshold. Defaults to medium.
The threshold determines when to wake up sleeping schedulers
when more work than can be handled by currently awake schedulers
exists. A low threshold causes earlier wakeups, and a high
threshold causes later wakeups. Early wakeups distribute work
over multiple schedulers faster, but work does more easily bounce
between schedulers.
Note
This flag can be removed or changed at any time without prior
notice.
-
+swtdcpu very_low|low|medium|high|very_high
-
As +swt but
affects dirty CPU schedulers. Defaults to medium.
Note
This flag can be removed or changed at any time
without prior notice.
-
+swtdio very_low|low|medium|high|very_high
-
As +swt but affects
dirty IO schedulers. Defaults to medium.
Note
This flag can be removed or changed at any time
without prior notice.
- +t size
-
Sets the maximum number of atoms the virtual machine can handle.
Defaults to 1,048,576.
- +T Level
-
Enables modified timing and sets the modified timing level. Valid
range is 0-9. The timing of the runtime system is changed. A high
level usually means a greater change than a low level. Changing the
timing can be very useful for finding timing-related bugs.
Modified timing affects the following:
- Process spawning
- A process calling spawn,
spawn_link, spawn_monitor,
or spawn_opt is scheduled out immediately
after completing the call. When higher modified timing levels are
used, the caller also sleeps for a while after it is scheduled out.
- Context reductions
- The number of reductions a process is allowed to use before it
is scheduled out is increased or reduced.
- Input reductions
- The number of reductions performed before checking I/O is
increased or reduced.
Note
Performance suffers when modified timing is enabled. This flag is
only intended for testing and debugging.
return_to and return_from
trace messages are lost when tracing on the spawn BIFs.
This flag can be removed or changed at any time without prior
notice.
- +v
-
Verbose.
- +V
-
Makes the emulator print its version number.
- +W w | i | e
-
Sets the mapping of warning messages for
error_logger. Messages sent to the error logger
using one of the warning routines can be mapped to errors
(+W e), warnings (+W w), or
information reports (+W i). Defaults to warnings.
The current mapping can be retrieved using
error_logger:warning_map/0. For more information,
see
error_logger:warning_map/0 in Kernel.
- +zFlag Value
-
Miscellaneous flags:
- +zdbbl size
-
Sets the distribution buffer busy limit
(dist_buf_busy_limit)
in kilobytes. Valid range is 1-2097151. Defaults to 1024.
A larger buffer limit allows processes to buffer
more outgoing messages over the distribution. When the
buffer limit has been reached, sending processes will be
suspended until the buffer size has shrunk. The buffer
limit is per distribution channel. A higher limit
gives lower latency and higher throughput at the expense
of higher memory use.
This limit only affects processes that have disabled
fully
asynchronous distributed signaling.
- +zdntgc time
-
Sets the delayed node table garbage collection time
(delayed_node_table_gc)
in seconds. Valid values are either infinity or
an integer in the range 0-100000000. Defaults to 60.
Node table entries that are not referred linger
in the table for at least the amount of time that this
parameter determines. The lingering prevents repeated
deletions and insertions in the tables from occurring.
- +zosrl limit
-
Sets a limit on the amount of outstanding requests made by
a system process orchestrating system wide changes. Valid
range of this limit is [1, 134217727]. See
erlang:system_flag(outstanding_system_requests_limit, Limit)
for more information.