bionic (3) erts_alloc_config.3erl.gz
NAME
erts_alloc_config - Configuration tool for erts_alloc
DESCRIPTION
Note:
erts_alloc_config is currently an experimental tool and might be subject to backward incompatible
changes.
erts_alloc(3erl) is an Erlang Run-Time System internal memory allocator library. erts_alloc_config is
intended to be used to aid creation of an erts_alloc(3erl) configuration that is suitable for a limited
number of runtime scenarios. The configuration that erts_alloc_config produce is intended as a
suggestion, and may need to be adjusted manually.
The configuration is created based on information about a number of runtime scenarios. It is obviously
impossible to foresee every runtime scenario that can occur. The important scenarios are those that cause
maximum or minimum load on specific memory allocators. Load in this context is total size of memory
blocks allocated.
The current implementation of erts_alloc_config concentrate on configuration of multi-block carriers.
Information gathered when a runtime scenario is saved is mainly current and maximum use of multi-block
carriers. If a parameter that change the use of multi-block carriers is changed, a previously generated
configuration is invalid and erts_alloc_config needs to be run again. It is mainly the single block
carrier threshold that effects the use of multi-block carriers, but other single-block carrier parameters
might as well. If another value of a single block carrier parameter than the default is desired, use the
desired value when running erts_alloc_config.
A configuration is created in the following way:
* Pass the +Mea config command-line flag to the Erlang runtime system you are going to use for creation
of the allocator configuration. It will disable features that prevent erts_alloc_config from doing
its job. Note, you should not use this flag when using the created configuration. Also note that it
is important that you use the same amount of schedulers when creating the configuration as you are
going the use on the system using the configuration.
* Run your applications with different scenarios (the more the better) and save information about each
scenario by calling save_scenario/0. It may be hard to know when the applications are at an (for
erts_alloc_config) important runtime scenario. A good approach may therefore be to call
save_scenario/0 repeatedly, e.g. once every tenth second. Note that it is important that your
applications reach the runtime scenarios that are important for erts_alloc_config when you are saving
scenarios; otherwise, the configuration may perform bad.
* When you have covered all scenarios, call make_config/1 in order to create a configuration. The
configuration is written to a file that you have chosen. This configuration file can later be read by
an Erlang runtime-system at startup. Pass the command line argument -args_file FileName to the erl(1)
command.
* The configuration produced by erts_alloc_config may need to be manually adjusted as already stated.
Do not modify the file produced by erts_alloc_config; instead, put your modifications in another file
and load this file after the file produced by erts_alloc_config. That is, put the -args_file FileName
argument that reads your modification file later on the command-line than the -args_file FileName
argument that reads the configuration file produced by erts_alloc_config. If a memory allocation
parameter appear multiple times, the last version of will be used, i.e., you can override parameters
in the configuration file produced by erts_alloc_config. Doing it this way simplifies things when you
want to rerun erts_alloc_config.
Note:
The configuration created by erts_alloc_config may perform bad, ever horrible, for runtime scenarios that
are very different from the ones saved when creating the configuration. You are, therefore, advised to
rerun erts_alloc_config if the applications run when the configuration was made are changed, or if the
load on the applications have changed since the configuration was made. You are also advised to rerun
erts_alloc_config if the Erlang runtime system used is changed.
erts_alloc_config saves information about runtime scenarios and performs computations in a server that is
automatically started. The server register itself under the name '__erts_alloc_config__'.
EXPORTS
save_scenario() -> ok | {error, Error}
Types:
Error = term()
save_scenario/0 saves information about the current runtime scenario. This information will later
be used when make_config/0, or make_config/1 is called.
The first time save_scenario/0 is called a server will be started. This server will save runtime
scenarios. All saved scenarios can be removed by calling stop/0.
make_config() -> ok | {error, Error}
Types:
Error = term()
This is the same as calling make_config(group_leader()).
make_config(FileNameOrIODev) -> ok | {error, Error}
Types:
FileNameOrIODev = string() | io_device()
Error = term()
make_config/1 uses the information previously saved by save_scenario/0 in order to produce an
erts_alloc configuration. At least one scenario have had to be saved. All scenarios previously
saved will be used when creating the configuration.
If FileNameOrIODev is a string(), make_config/1 will use FileNameOrIODev as a filename. A file
named FileNameOrIODev is created and the configuration will be written to that file. If
FileNameOrIODev is an io_device() (see the documentation of the module io), the configuration will
be written to the io device.
stop() -> ok | {error, Error}
Types:
Error = term()
Stops the server that saves runtime scenarios.
SEE ALSO
erts_alloc(3erl), erl(1), io(3erl)