Decentralized counters for ETS ordered_set with write_concurrency

Previously, all ETS tables used centralized counter variables to keep
track of the number of items stored and the amount of memory
consumed. These counters can cause scalability problems (especially on
big NUMA systems). This commit adds an implementation of a
decentralized counter and modifies the implementation of ETS so that
ETS tables of type ordered_set with write_concurrency enabled use the
decentralized counter. [Experiments][1] indicate that this change
substantially improves the scalability of ETS ordered_set tables with
write_concurrency enabled in scenarios with frequent `ets:insert/2`
and `ets:delete/2` calls.

The new counter is implemented in the module erts_flxctr
(`erts_flxctr.h` and `erts_flxctr.c`). The module has the suffix
flxctr as it contains the implementation of a flexible counter (i.e.,
counter instances can be configured to be either centralized or
decentralized). Counters that are configured to be centralized are
implemented with a single counter variable which is modified with
atomic operations. Decentralized counters are spread over several
cache lines (how many can be configured with the parameter
`+dcg`). The scheduler threads are mapped to cache lines so that there
is no single point of contention when decentralized counters are
updated. The thread progress functionality of the Erlang VM is
utilized to implement support for linearizable snapshots of
decentralized counters. The snapshot functionality is used by the
`ets:info/1` and `ets:info/2` functions.

[1]: http://winsh.me/ets_catree_benchmark/flxctr_res.html

1 files changed, 23 insertions, 0 deletions

diff --git a/erts/doc/src/erl.xml b/erts/doc/src/erl.xml
index 88ddb03e97..471d7caa5a 100644
--- a/erts/doc/src/erl.xml
+++ b/erts/doc/src/erl.xml
@@ -636,6 +636,29 @@
           produces a crash dump. On Unix systems, sending an emulator process
           a <c>SIGUSR1</c> signal also forces a crash dump.</p>
       </item>
+      <tag><marker id="+dcg"/><c><![CDATA[+rg DecentralizedCounterGroupsLimit]]></c></tag>
+      <item>
+        <p>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.</p>
+        <p>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.</p>
+        <p>Notice that a runtime system using decentralized
+           counter groups benefits from <seealso marker="#+sbt">binding
+           schedulers to logical processors</seealso>, as the groups are
+           distributed better between schedulers with this option.</p>
+        <p>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.</p>
+      </item>
       <tag><marker id="+e"/><c><![CDATA[+e Number]]></c></tag>
       <item>
         <p>Sets the maximum number of ETS tables. This limit is