aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorRickard Green <[email protected]>2011-11-18 12:46:43 +0100
committerRickard Green <[email protected]>2011-11-18 13:09:16 +0100
commit58f5f45566b827e74ff623398bdf6d0b6fcebdb3 (patch)
tree3d68d4afef9e55d4453b8550ba2488d202a2b484
parenta8ecde4e398942260153d8b2f37344af259b1007 (diff)
downloadotp-58f5f45566b827e74ff623398bdf6d0b6fcebdb3.tar.gz
otp-58f5f45566b827e74ff623398bdf6d0b6fcebdb3.tar.bz2
otp-58f5f45566b827e74ff623398bdf6d0b6fcebdb3.zip
Use unbound schedulers as default
As of ERTS version 5.9 (OTP-R15B) the runtime system will by default not bind schedulers to logical processors. 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 (as for example another Erlang runtime system) also bind threads to logical processors, there might be a performance penalty instead. In some cases this performance penalty might be severe. Due to this, we change the default so that the user must make an active decision in order to bind schedulers.
-rw-r--r--erts/doc/src/erl.xml50
-rw-r--r--erts/doc/src/erlang.xml14
-rw-r--r--erts/emulator/beam/erl_cpu_topology.c12
3 files changed, 23 insertions, 53 deletions
diff --git a/erts/doc/src/erl.xml b/erts/doc/src/erl.xml
index 02082e57c6..048ffbcd47 100644
--- a/erts/doc/src/erl.xml
+++ b/erts/doc/src/erl.xml
@@ -41,25 +41,11 @@
to scroll back to text which has scrolled off the screen.
The <c><![CDATA[erl]]></c> program must be used, however, in pipelines or if
you want to redirect standard input or output.</p>
- <note><p>As of ERTS version 5.8 (OTP-R14A) the runtime system will by
- default bind schedulers to logical processors using the
- <c>default_bind</c> bind type if the amount of schedulers are
- at least equal to the amount of logical processors configured,
- binding of schedulers is supported, and a CPU topology is
- available at startup.
- </p><p>
- 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 (as for example
- another Erlang runtime system) also bind threads to
- logical processors, there might be a performance penalty
- instead. If this is the case you, are are advised to
- unbind the schedulers using the
- <seealso marker="#+sbt">+sbtu</seealso> command line argument,
- or by invoking
- <seealso marker="erlang#system_flag_scheduler_bind_type">erlang:system_flag(scheduler_bind_type,
- unbound)</seealso>.</p>
+ <note><p>As of ERTS version 5.9 (OTP-R15B) the runtime system will by
+ default <em>not</em> bind schedulers to logical processors.
+ For more information see documentation of the
+ <seealso marker="#+sbt">+sbt</seealso> system flag.
+ </p>
</note>
</description>
<funcs>
@@ -718,22 +704,18 @@
that the <c>+sct</c> flag may have to be passed before the
<c>+sbt</c> flag on the command line (in case no CPU topology
has been automatically detected).</p>
- <p>The runtime system will by default bind schedulers to logical
- processors using the <c>default_bind</c> bind type if the amount
- of schedulers are at least equal to the amount of logical
- processors configured, binding of schedulers is supported,
- and a CPU topology is available at startup.
+ <p>The runtime system will by default <em>not</em> bind schedulers
+ to logical processors.
</p>
- <p><em>NOTE:</em> 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 (as for example another Erlang runtime
- system) also bind threads to logical processors, there might be a
- performance penalty instead. If this is the case you, are advised
- to unbind the schedulers using the <c>+sbtu</c> command line
- argument, or by invoking
- <seealso marker="erlang#system_flag_scheduler_bind_type">erlang:system_flag(scheduler_bind_type,
- unbound)</seealso>.</p>
+ <p><em>NOTE:</em> 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 (as for example
+ another Erlang runtime system) also bind threads to
+ logical processors, there might be a performance penalty
+ instead. In some cases this performance penalty might be
+ severe. If this is the case, you are advised to not
+ bind the schedulers.</p>
<p>For more information, see
<seealso marker="erlang#system_flag_scheduler_bind_type">erlang:system_flag(scheduler_bind_type, SchedulerBindType)</seealso>.
</p>
diff --git a/erts/doc/src/erlang.xml b/erts/doc/src/erlang.xml
index aef31f5b98..ae6300a728 100644
--- a/erts/doc/src/erlang.xml
+++ b/erts/doc/src/erlang.xml
@@ -5173,21 +5173,17 @@ true</pre>
For more information on how to define the CPU topology, see
<seealso marker="#system_flag_cpu_topology">erlang:system_flag(cpu_topology, CpuTopology)</seealso>.
</p>
- <p>The runtime system will by default bind schedulers to logical
- processors using the <c>default_bind</c> bind type if the amount
- of schedulers are at least equal to the amount of logical
- processors configured, binding of schedulers is supported,
- and a CPU topology is available at startup.
+ <p>The runtime system will by default <em>not</em> bind schedulers
+ to logical processors.
</p>
<p><em>NOTE:</em> 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 (as for example another Erlang
runtime system) also bind threads to logical processors, there
- might be a performance penalty instead. If this is the case you,
- are are advised to unbind the schedulers using the
- <seealso marker="erl#+sbt">+sbtu</seealso> command line argument,
- or <c>erlang:system_flag(scheduler_bind_type, unbound)</c>.</p>
+ might be a performance penalty instead. In some cases this
+ performance penalty might be severe. If this is the case, you
+ are advised to not bind the schedulers.</p>
<p>Schedulers can be bound in different ways. The <c>How</c>
argument determines how schedulers are bound. <c>How</c> can
currently be one of:</p>
diff --git a/erts/emulator/beam/erl_cpu_topology.c b/erts/emulator/beam/erl_cpu_topology.c
index 03c0ef904a..cc930ba1e3 100644
--- a/erts/emulator/beam/erl_cpu_topology.c
+++ b/erts/emulator/beam/erl_cpu_topology.c
@@ -1729,16 +1729,8 @@ erts_init_cpu_topology(void)
scheduler2cpu_map[ix].bound_id = -1;
}
- if (cpu_bind_order == ERTS_CPU_BIND_UNDEFINED) {
- int ncpus = erts_get_cpu_configured(cpuinfo);
- if (ncpus < 1 || erts_no_schedulers < ncpus)
- cpu_bind_order = ERTS_CPU_BIND_NONE;
- else
- cpu_bind_order = ((system_cpudata || user_cpudata)
- && (erts_bind_to_cpu(cpuinfo, -1) != -ENOTSUP)
- ? ERTS_CPU_BIND_DEFAULT_BIND
- : ERTS_CPU_BIND_NONE);
- }
+ if (cpu_bind_order == ERTS_CPU_BIND_UNDEFINED)
+ cpu_bind_order = ERTS_CPU_BIND_NONE;
reader_groups_map = add_cpu_groups(reader_groups,
reader_groups_callback,