Implement dynamic socket buffer sizes

Cowboy will set the socket's buffer size dynamically to
better fit the current workload. When the incoming data
is small, a low buffer size reduces the memory footprint
and improves responsiveness and therefore performance.
When the incoming data is large, such as large HTTP
request bodies, a larger buffer size helps us avoid
doing too many binary appends and related allocations.

Setting a large buffer size for all use cases is
sub-optimal because allocating more than needed
necessarily results in a performance hit (not just
increased memory usage).

By default Cowboy starts with a buffer size of 8192 bytes.
It then doubles or halves the buffer size depending on
the size of the data it receives from the socket. It
stops decreasing at 8192 and increasing at 131072 by
default.

To keep track of the size of the incoming data Cowboy
maintains a moving average. It allows Cowboy to avoid
changing the buffer too often but still react quickly
when necessary. Cowboy will increase the buffer size
when the moving average is above 90% of the current
buffer size, and decrease when the moving average is
below 40% of the current buffer size.

The current buffer size and moving average are
propagated when switching protocols. The dynamic buffer
is implemented in HTTP/1, HTTP/2 and HTTP/1 Websocket.
HTTP/2 Websocket has it disabled because it doesn't
interact directly with the socket; in that case it
is HTTP/2 that has a dynamic buffer.

The dynamic buffer provides a very large performance improvement
in many scenarios, at minimal cost for others. Because it largely
depend on the underlying protocol the improvements are no all equal.
TLS and compression also impact the results.

The improvement when reading a large request body, with the
requests repeated in a fast loop are:

* HTTP: 6x to 20x faster
* HTTPS: 2x to 6x faster
* H2: 4x to 5x faster
* H2C: 20x to 40x faster

I am not sure why H2C's performance was so bad, especially compared
to H2, when using default buffer sizes. Dynamic buffers make H2C a
lot more viable with default settings.

The performance impact on "hello world" type requests is minimal,
it goes from -5% to +5% roughly.

Websocket improvements vary again depending on the protocol, but
also depending on whether compression is enabled:

* HTTP echo: roughly 2x faster
* HTTP send: roughly 4x faster
* H2C echo: roughly 2x faster
* H2C send: 3x to 4x faster

In the echo test we reply back, and Gun doesn't have the dynamic
buffer optimisation, so that probably explains the x2 difference.

With compression however there isn't much improvement. The results
are roughly within -10% to +10% of each other. Zlib compression
seems to be a bottleneck, or at least to modify the performance
profile to such an extent that the size of the buffer does not
matter. This happens to randomly generated binary data as well
so it is probably not caused by the test data.

3 files changed, 45 insertions, 3 deletions

diff --git a/doc/src/manual/cowboy_http.asciidoc b/doc/src/manual/cowboy_http.asciidoc
index 58f0435..31e2d37 100644
--- a/doc/src/manual/cowboy_http.asciidoc
+++ b/doc/src/manual/cowboy_http.asciidoc
@@ -20,6 +20,7 @@ opts() :: #{
     active_n                   => pos_integer(),
     chunked                    => boolean(),
     connection_type            => worker | supervisor,
+    dynamic_buffer             => false | {pos_integer(), pos_integer()},
     http10_keepalive           => boolean(),
     idle_timeout               => timeout(),
     inactivity_timeout         => timeout(),
@@ -53,7 +54,7 @@ Ranch functions `ranch:get_protocol_options/1` and
 
 The default value is given next to the option name:
 
-active_n (100)::
+active_n (1)::
 
 The number of packets Cowboy will request from the socket at once.
 This can be used to tweak the performance of the server. Higher
@@ -75,6 +76,17 @@ connection_type (supervisor)::
 
 Whether the connection process also acts as a supervisor.
 
+dynamic_buffer ({8192, 131072})::
+
+Cowboy will dynamically change the socket's `buffer` size
+depending on the size of the data it receives from the socket.
+This lets Cowboy use the optimal buffer size for the current
+workload.
++
+The dynamic buffer size functionality can be disabled by
+setting this option to `false`. Cowboy will also disable
+it by default when the `buffer` transport option is configured.
+
 http10_keepalive (true)::
 
 Whether keep-alive is enabled for HTTP/1.0 connections.
@@ -166,6 +178,8 @@ Ordered list of stream handlers that will handle all stream events.
 
 == Changelog
 
+* *2.13*: The `active_n` default value was changed to `1`.
+* *2.13*: The `dynamic_buffer` option was added.
 * *2.11*: The `reset_idle_timeout_on_send` option was added.
 * *2.8*: The `active_n` option was added.
 * *2.7*: The `initial_stream_flow_size` and `logger` options were added.
diff --git a/doc/src/manual/cowboy_http2.asciidoc b/doc/src/manual/cowboy_http2.asciidoc
index 1d2619c..a5fcd0b 100644
--- a/doc/src/manual/cowboy_http2.asciidoc
+++ b/doc/src/manual/cowboy_http2.asciidoc
@@ -21,6 +21,7 @@ opts() :: #{
     connection_type                => worker | supervisor,
     connection_window_margin_size  => 0..16#7fffffff,
     connection_window_update_threshold => 0..16#7fffffff,
+    dynamic_buffer                 => false | {pos_integer(), pos_integer()},
     enable_connect_protocol        => boolean(),
     goaway_initial_timeout         => timeout(),
     goaway_complete_timeout        => timeout(),
@@ -66,7 +67,7 @@ Ranch functions `ranch:get_protocol_options/1` and
 
 The default value is given next to the option name:
 
-active_n (100)::
+active_n (1)::
 
 The number of packets Cowboy will request from the socket at once.
 This can be used to tweak the performance of the server. Higher
@@ -91,6 +92,17 @@ The connection window will only get updated when its size
 becomes lower than this threshold, in bytes. This is to
 avoid sending too many `WINDOW_UPDATE` frames.
 
+dynamic_buffer ({8192, 131072})::
+
+Cowboy will dynamically change the socket's `buffer` size
+depending on the size of the data it receives from the socket.
+This lets Cowboy use the optimal buffer size for the current
+workload.
++
+The dynamic buffer size functionality can be disabled by
+setting this option to `false`. Cowboy will also disable
+it by default when the `buffer` transport option is configured.
+
 enable_connect_protocol (false)::
 
 Whether to enable the extended CONNECT method to allow
@@ -289,6 +301,8 @@ too many `WINDOW_UPDATE` frames.
 
 == Changelog
 
+* *2.13*: The `active_n` default value was changed to `1`.
+* *2.13*: The `dynamic_buffer` option was added.
 * *2.11*: Websocket over HTTP/2 is now considered stable.
 * *2.11*: The `reset_idle_timeout_on_send` option was added.
 * *2.11*: Add the option `max_cancel_stream_rate` to protect
diff --git a/doc/src/manual/cowboy_websocket.asciidoc b/doc/src/manual/cowboy_websocket.asciidoc
index e152182..d5db82f 100644
--- a/doc/src/manual/cowboy_websocket.asciidoc
+++ b/doc/src/manual/cowboy_websocket.asciidoc
@@ -203,6 +203,7 @@ opts() :: #{
     active_n       => pos_integer(),
     compress       => boolean(),
     deflate_opts   => cow_ws:deflate_opts()
+    dynamic_buffer => false | {pos_integer(), pos_integer()},
     idle_timeout   => timeout(),
     max_frame_size => non_neg_integer() | infinity,
     req_filter     => fun((cowboy_req:req()) -> map()),
@@ -224,7 +225,7 @@ init(Req, State) ->
 
 The default value is given next to the option name:
 
-active_n (100)::
+active_n (1)::
 
 The number of packets Cowboy will request from the socket at once.
 This can be used to tweak the performance of the server. Higher
@@ -248,6 +249,17 @@ options and the zlib compression options. The
 defaults optimize the compression at the expense
 of some memory and CPU.
 
+dynamic_buffer ({8192, 131072})::
+
+Cowboy will dynamically change the socket's `buffer` size
+depending on the size of the data it receives from the socket.
+This lets Cowboy use the optimal buffer size for the current
+workload.
++
+The dynamic buffer size functionality can be disabled by
+setting this option to `false`. Cowboy will also disable
+it by default when the `buffer` transport option is configured.
+
 idle_timeout (60000)::
 
 Time in milliseconds that Cowboy will keep the
@@ -287,6 +299,8 @@ normal circumstances if necessary.
 
 == Changelog
 
+* *2.13*: The `active_n` default value was changed to `1`.
+* *2.13*: The `dynamic_buffer` option was added.
 * *2.13*: The `max_frame_size` option can now be set dynamically.
 * *2.11*: Websocket over HTTP/2 is now considered stable.
 * *2.11*: HTTP/1.1 Websocket no longer traps exits by default.