diff --git a/nano/boost/asio/ip/network_v6.hpp b/nano/boost/asio/ip/network_v6.hpp
new file mode 100644
index 0000000000..b9f213ab63
--- /dev/null
+++ b/nano/boost/asio/ip/network_v6.hpp
@@ -0,0 +1,7 @@
+#pragma once
+
+#include <nano/boost/private/macro_warnings.hpp>
+
+DISABLE_ASIO_WARNINGS
+#include <boost/asio/ip/network_v6.hpp>
+REENABLE_WARNINGS
diff --git a/nano/core_test/network.cpp b/nano/core_test/network.cpp
index f0464c6767..a5bd86da24 100644
--- a/nano/core_test/network.cpp
+++ b/nano/core_test/network.cpp
@@ -1,3 +1,4 @@
+#include <nano/node/nodeconfig.hpp>
 #include <nano/node/transport/udp.hpp>
 #include <nano/test_common/network.hpp>
 #include <nano/test_common/system.hpp>
@@ -918,7 +919,10 @@ namespace transport
 {
 	TEST (network, peer_max_tcp_attempts_subnetwork)
 	{
-		nano::system system (1);
+		nano::node_flags node_flags;
+		node_flags.disable_max_peers_per_ip = true;
+		nano::system system;
+		system.add_node (node_flags);
 		auto node (system.nodes[0]);
 		for (auto i (0); i < node->network_params.network.max_peers_per_subnetwork; ++i)
 		{
@@ -927,9 +931,9 @@ namespace transport
 			ASSERT_FALSE (node->network.tcp_channels.reachout (endpoint));
 		}
 		ASSERT_EQ (0, node->network.size ());
-		ASSERT_EQ (0, node->stats.count (nano::stat::type::tcp, nano::stat::detail::tcp_max_per_ip, nano::stat::dir::out));
+		ASSERT_EQ (0, node->stats.count (nano::stat::type::tcp, nano::stat::detail::tcp_max_per_subnetwork, nano::stat::dir::out));
 		ASSERT_TRUE (node->network.tcp_channels.reachout (nano::endpoint (boost::asio::ip::make_address_v6 ("::ffff:127.0.0.1"), nano::get_available_port ())));
-		ASSERT_EQ (1, node->stats.count (nano::stat::type::tcp, nano::stat::detail::tcp_max_per_ip, nano::stat::dir::out));
+		ASSERT_EQ (1, node->stats.count (nano::stat::type::tcp, nano::stat::detail::tcp_max_per_subnetwork, nano::stat::dir::out));
 	}
 }
 }
diff --git a/nano/core_test/socket.cpp b/nano/core_test/socket.cpp
index a5746023b9..32d921dd01 100644
--- a/nano/core_test/socket.cpp
+++ b/nano/core_test/socket.cpp
@@ -1,3 +1,5 @@
+#include <nano/boost/asio/ip/address_v6.hpp>
+#include <nano/boost/asio/ip/network_v6.hpp>
 #include <nano/lib/threading.hpp>
 #include <nano/node/socket.hpp>
 #include <nano/test_common/system.hpp>
@@ -5,23 +7,22 @@
 
 #include <gtest/gtest.h>
 
+#include <map>
+#include <memory>
+#include <utility>
+#include <vector>
+
 using namespace std::chrono_literals;
 
 TEST (socket, max_connections)
 {
-	// this is here just so that ASSERT_TIMELY can be used
 	nano::system system;
 
-	auto node_flags = nano::inactive_node_flag_defaults ();
-	node_flags.read_only = false;
-	nano::inactive_node inactivenode (nano::unique_path (), node_flags);
-	auto node = inactivenode.node;
-
-	nano::thread_runner runner (node->io_ctx, 1);
+	auto node = system.add_node ();
 
-	auto server_port (nano::get_available_port ());
-	boost::asio::ip::tcp::endpoint listen_endpoint (boost::asio::ip::address_v6::any (), server_port);
-	boost::asio::ip::tcp::endpoint dst_endpoint (boost::asio::ip::address_v6::loopback (), server_port);
+	auto server_port = nano::get_available_port ();
+	boost::asio::ip::tcp::endpoint listen_endpoint{ boost::asio::ip::address_v6::any (), server_port };
+	boost::asio::ip::tcp::endpoint dst_endpoint{ boost::asio::ip::address_v6::loopback (), server_port };
 
 	// start a server socket that allows max 2 live connections
 	auto server_socket = std::make_shared<nano::server_socket> (*node, listen_endpoint, 2);
@@ -37,10 +38,10 @@ TEST (socket, max_connections)
 	});
 
 	// client side connection tracking
-	std::atomic<int> connection_attempts = 0;
+	std::atomic<size_t> connection_attempts = 0;
 	auto connect_handler = [&connection_attempts] (boost::system::error_code const & ec_a) {
 		ASSERT_EQ (ec_a.value (), 0);
-		connection_attempts++;
+		++connection_attempts;
 	};
 
 	// start 3 clients, 2 will persist but 1 will be dropped
@@ -102,8 +103,280 @@ TEST (socket, max_connections)
 	ASSERT_TIMELY (5s, server_sockets.size () == 5); // connections accepted by the server
 
 	node->stop ();
-	runner.stop_event_processing ();
-	runner.join ();
+}
+
+TEST (socket, max_connections_per_ip)
+{
+	nano::system system;
+
+	auto node = system.add_node ();
+	ASSERT_FALSE (node->flags.disable_max_peers_per_ip);
+
+	auto server_port = nano::get_available_port ();
+	boost::asio::ip::tcp::endpoint listen_endpoint{ boost::asio::ip::address_v6::any (), server_port };
+	boost::asio::ip::tcp::endpoint dst_endpoint{ boost::asio::ip::address_v6::loopback (), server_port };
+
+	const auto max_ip_connections = node->network_params.network.max_peers_per_ip;
+	ASSERT_TRUE (max_ip_connections >= 1);
+
+	const auto max_global_connections = 1000;
+
+	auto server_socket = std::make_shared<nano::server_socket> (*node, listen_endpoint, max_global_connections);
+	boost::system::error_code ec;
+	server_socket->start (ec);
+	ASSERT_FALSE (ec);
+
+	// successful incoming connections are stored in server_sockets to keep them alive (server side)
+	std::vector<std::shared_ptr<nano::socket>> server_sockets;
+	server_socket->on_connection ([&server_sockets] (std::shared_ptr<nano::socket> const & new_connection, boost::system::error_code const & ec_a) {
+		server_sockets.push_back (new_connection);
+		return true;
+	});
+
+	// client side connection tracking
+	std::atomic<size_t> connection_attempts = 0;
+	auto connect_handler = [&connection_attempts] (boost::system::error_code const & ec_a) {
+		ASSERT_EQ (ec_a.value (), 0);
+		++connection_attempts;
+	};
+
+	// start n clients, n-1 will persist but 1 will be dropped, where n == max_ip_connections
+	std::vector<std::shared_ptr<nano::socket>> client_list;
+	client_list.reserve (max_ip_connections + 1);
+
+	for (auto idx = 0; idx < max_ip_connections + 1; ++idx)
+	{
+		auto client = std::make_shared<nano::socket> (*node);
+		client->async_connect (dst_endpoint, connect_handler);
+		client_list.push_back (client);
+	}
+
+	auto get_tcp_max_per_ip = [&node] () {
+		return node->stats.count (nano::stat::type::tcp, nano::stat::detail::tcp_max_per_ip, nano::stat::dir::in);
+	};
+
+	auto get_tcp_accept_successes = [&node] () {
+		return node->stats.count (nano::stat::type::tcp, nano::stat::detail::tcp_accept_success, nano::stat::dir::in);
+	};
+
+	ASSERT_TIMELY (5s, get_tcp_accept_successes () == max_ip_connections);
+	ASSERT_TIMELY (5s, get_tcp_max_per_ip () == 1);
+	ASSERT_TIMELY (5s, connection_attempts == max_ip_connections + 1);
+
+	node->stop ();
+}
+
+TEST (socket, limited_subnet_address)
+{
+	auto address = boost::asio::ip::make_address ("a41d:b7b2:8298:cf45:672e:bd1a:e7fb:f713");
+	auto network = nano::socket_functions::get_ipv6_subnet_address (address.to_v6 (), 32); // network prefix = 32.
+	ASSERT_EQ ("a41d:b7b2:8298:cf45:672e:bd1a:e7fb:f713/32", network.to_string ());
+	ASSERT_EQ ("a41d:b7b2::/32", network.canonical ().to_string ());
+}
+
+TEST (socket, first_ipv6_subnet_address)
+{
+	auto address = boost::asio::ip::make_address ("a41d:b7b2:8298:cf45:672e:bd1a:e7fb:f713");
+	auto first_address = nano::socket_functions::first_ipv6_subnet_address (address.to_v6 (), 32); // network prefix = 32.
+	ASSERT_EQ ("a41d:b7b2::", first_address.to_string ());
+}
+
+TEST (socket, last_ipv6_subnet_address)
+{
+	auto address = boost::asio::ip::make_address ("a41d:b7b2:8298:cf45:672e:bd1a:e7fb:f713");
+	auto last_address = nano::socket_functions::last_ipv6_subnet_address (address.to_v6 (), 32); // network prefix = 32.
+	ASSERT_EQ ("a41d:b7b2:ffff:ffff:ffff:ffff:ffff:ffff", last_address.to_string ());
+}
+
+TEST (socket, count_subnetwork_connections)
+{
+	nano::system system;
+	auto node = system.add_node ();
+
+	auto address0 = boost::asio::ip::make_address ("a41d:b7b1:ffff:ffff:ffff:ffff:ffff:ffff"); // out of network prefix
+	auto address1 = boost::asio::ip::make_address ("a41d:b7b2:8298:cf45:672e:bd1a:e7fb:f713"); // referece address
+	auto address2 = boost::asio::ip::make_address ("a41d:b7b2::"); // start of the network range
+	auto address3 = boost::asio::ip::make_address ("a41d:b7b2::1");
+	auto address4 = boost::asio::ip::make_address ("a41d:b7b2:ffff:ffff:ffff:ffff:ffff:ffff"); // end of the network range
+	auto address5 = boost::asio::ip::make_address ("a41d:b7b3::"); // out of the network prefix
+	auto address6 = boost::asio::ip::make_address ("a41d:b7b3::1"); // out of the network prefix
+
+	auto connection0 = std::make_shared<nano::socket> (*node);
+	auto connection1 = std::make_shared<nano::socket> (*node);
+	auto connection2 = std::make_shared<nano::socket> (*node);
+	auto connection3 = std::make_shared<nano::socket> (*node);
+	auto connection4 = std::make_shared<nano::socket> (*node);
+	auto connection5 = std::make_shared<nano::socket> (*node);
+	auto connection6 = std::make_shared<nano::socket> (*node);
+
+	nano::address_socket_mmap connections_per_address;
+	connections_per_address.emplace (address0, connection0);
+	connections_per_address.emplace (address1, connection1);
+	connections_per_address.emplace (address2, connection2);
+	connections_per_address.emplace (address3, connection3);
+	connections_per_address.emplace (address4, connection4);
+	connections_per_address.emplace (address5, connection5);
+	connections_per_address.emplace (address6, connection6);
+
+	// Asserts it counts only the connections for the specified address and its network prefix.
+	ASSERT_EQ (4, nano::socket_functions::count_subnetwork_connections (connections_per_address, address1.to_v6 (), 32));
+}
+
+TEST (socket, max_connections_per_subnetwork)
+{
+	nano::system system;
+
+	nano::node_flags node_flags;
+	// disabling IP limit because it will be used the same IP address to check they come from the same subnetwork.
+	node_flags.disable_max_peers_per_ip = true;
+	node_flags.disable_max_peers_per_subnetwork = false;
+	auto node = system.add_node (node_flags);
+	ASSERT_TRUE (node->flags.disable_max_peers_per_ip);
+	ASSERT_FALSE (node->flags.disable_max_peers_per_subnetwork);
+
+	auto server_port = nano::get_available_port ();
+	boost::asio::ip::tcp::endpoint listen_endpoint{ boost::asio::ip::address_v6::any (), server_port };
+	boost::asio::ip::tcp::endpoint dst_endpoint{ boost::asio::ip::address_v6::loopback (), server_port };
+
+	const auto max_subnetwork_connections = node->network_params.network.max_peers_per_subnetwork;
+	ASSERT_TRUE (max_subnetwork_connections >= 1);
+
+	const auto max_global_connections = 1000;
+
+	auto server_socket = std::make_shared<nano::server_socket> (*node, listen_endpoint, max_global_connections);
+	boost::system::error_code ec;
+	server_socket->start (ec);
+	ASSERT_FALSE (ec);
+
+	// successful incoming connections are stored in server_sockets to keep them alive (server side)
+	std::vector<std::shared_ptr<nano::socket>> server_sockets;
+	server_socket->on_connection ([&server_sockets] (std::shared_ptr<nano::socket> const & new_connection, boost::system::error_code const & ec_a) {
+		server_sockets.push_back (new_connection);
+		return true;
+	});
+
+	// client side connection tracking
+	std::atomic<size_t> connection_attempts = 0;
+	auto connect_handler = [&connection_attempts] (boost::system::error_code const & ec_a) {
+		ASSERT_EQ (ec_a.value (), 0);
+		++connection_attempts;
+	};
+
+	// start n clients, n-1 will persist but 1 will be dropped, where n == max_subnetwork_connections
+	std::vector<std::shared_ptr<nano::socket>> client_list;
+	client_list.reserve (max_subnetwork_connections + 1);
+
+	for (auto idx = 0; idx < max_subnetwork_connections + 1; ++idx)
+	{
+		auto client = std::make_shared<nano::socket> (*node);
+		client->async_connect (dst_endpoint, connect_handler);
+		client_list.push_back (client);
+	}
+
+	auto get_tcp_max_per_subnetwork = [&node] () {
+		return node->stats.count (nano::stat::type::tcp, nano::stat::detail::tcp_max_per_subnetwork, nano::stat::dir::in);
+	};
+
+	auto get_tcp_accept_successes = [&node] () {
+		return node->stats.count (nano::stat::type::tcp, nano::stat::detail::tcp_accept_success, nano::stat::dir::in);
+	};
+
+	ASSERT_TIMELY (5s, get_tcp_accept_successes () == max_subnetwork_connections);
+	ASSERT_TIMELY (5s, get_tcp_max_per_subnetwork () == 1);
+	ASSERT_TIMELY (5s, connection_attempts == max_subnetwork_connections + 1);
+
+	node->stop ();
+}
+
+TEST (socket, disabled_max_peers_per_ip)
+{
+	nano::system system;
+
+	nano::node_flags node_flags;
+	node_flags.disable_max_peers_per_ip = true;
+	auto node = system.add_node (node_flags);
+	ASSERT_TRUE (node->flags.disable_max_peers_per_ip);
+
+	auto server_port = nano::get_available_port ();
+	boost::asio::ip::tcp::endpoint listen_endpoint{ boost::asio::ip::address_v6::any (), server_port };
+	boost::asio::ip::tcp::endpoint dst_endpoint{ boost::asio::ip::address_v6::loopback (), server_port };
+
+	const auto max_ip_connections = node->network_params.network.max_peers_per_ip;
+	ASSERT_TRUE (max_ip_connections >= 1);
+
+	const auto max_global_connections = 1000;
+
+	auto server_socket = std::make_shared<nano::server_socket> (*node, listen_endpoint, max_global_connections);
+	boost::system::error_code ec;
+	server_socket->start (ec);
+	ASSERT_FALSE (ec);
+
+	// successful incoming connections are stored in server_sockets to keep them alive (server side)
+	std::vector<std::shared_ptr<nano::socket>> server_sockets;
+	server_socket->on_connection ([&server_sockets] (std::shared_ptr<nano::socket> const & new_connection, boost::system::error_code const & ec_a) {
+		server_sockets.push_back (new_connection);
+		return true;
+	});
+
+	// client side connection tracking
+	std::atomic<size_t> connection_attempts = 0;
+	auto connect_handler = [&connection_attempts] (boost::system::error_code const & ec_a) {
+		ASSERT_EQ (ec_a.value (), 0);
+		++connection_attempts;
+	};
+
+	// start n clients, n-1 will persist but 1 will be dropped, where n == max_ip_connections
+	std::vector<std::shared_ptr<nano::socket>> client_list;
+	client_list.reserve (max_ip_connections + 1);
+
+	for (auto idx = 0; idx < max_ip_connections + 1; ++idx)
+	{
+		auto client = std::make_shared<nano::socket> (*node);
+		client->async_connect (dst_endpoint, connect_handler);
+		client_list.push_back (client);
+	}
+
+	auto get_tcp_max_per_ip = [&node] () {
+		return node->stats.count (nano::stat::type::tcp, nano::stat::detail::tcp_max_per_ip, nano::stat::dir::in);
+	};
+
+	auto get_tcp_accept_successes = [&node] () {
+		return node->stats.count (nano::stat::type::tcp, nano::stat::detail::tcp_accept_success, nano::stat::dir::in);
+	};
+
+	ASSERT_TIMELY (5s, get_tcp_accept_successes () == max_ip_connections + 1);
+	ASSERT_TIMELY (5s, get_tcp_max_per_ip () == 0);
+	ASSERT_TIMELY (5s, connection_attempts == max_ip_connections + 1);
+
+	node->stop ();
+}
+
+TEST (socket, disconnection_of_silent_connections)
+{
+	nano::system system;
+	auto node = system.add_node ();
+	auto socket = std::make_shared<nano::socket> (*node);
+	// Classify the socket type as real-time as the disconnections are done only for this connection type.
+	socket->type_set (nano::socket::type_t::realtime);
+	// Silent connections are connections open by external peers that don't contribute with any data.
+	socket->set_silent_connection_tolerance_time (std::chrono::seconds{ 5 });
+	auto bootstrap_endpoint = node->bootstrap.endpoint ();
+	std::atomic<bool> connected{ false };
+	// Opening a connection that will be closed because it remains silent during the tolerance time.
+	socket->async_connect (bootstrap_endpoint, [socket, &connected] (boost::system::error_code const & ec) {
+		ASSERT_FALSE (ec);
+		connected = true;
+	});
+	ASSERT_TIMELY (4s, connected);
+	// Checking the connection was closed.
+	ASSERT_TIMELY (10s, socket->is_closed ());
+
+	auto get_tcp_silent_connection_drops = [&node] () {
+		return node->stats.count (nano::stat::type::tcp, nano::stat::detail::tcp_silent_connection_drop, nano::stat::dir::in);
+	};
+	ASSERT_EQ (1, get_tcp_silent_connection_drops ());
+
+	node->stop ();
 }
 
 TEST (socket, drop_policy)
@@ -173,6 +446,8 @@ TEST (socket, concurrent_writes)
 {
 	auto node_flags = nano::inactive_node_flag_defaults ();
 	node_flags.read_only = false;
+	node_flags.disable_max_peers_per_ip = true;
+	node_flags.disable_max_peers_per_subnetwork = true;
 	nano::inactive_node inactivenode (nano::unique_path (), node_flags);
 	auto node = inactivenode.node;
 
diff --git a/nano/lib/config.hpp b/nano/lib/config.hpp
index cd101d921f..e6eb1c50d7 100644
--- a/nano/lib/config.hpp
+++ b/nano/lib/config.hpp
@@ -156,10 +156,12 @@ class network_constants
 		request_interval_ms = is_dev_network () ? 20 : 500;
 		cleanup_period = is_dev_network () ? std::chrono::seconds (1) : std::chrono::seconds (60);
 		idle_timeout = is_dev_network () ? cleanup_period * 15 : cleanup_period * 2;
+		silent_connection_tolerance_time = std::chrono::seconds (120);
 		syn_cookie_cutoff = std::chrono::seconds (5);
 		bootstrap_interval = std::chrono::seconds (15 * 60);
 		max_peers_per_ip = is_dev_network () ? 10 : 5;
 		max_peers_per_subnetwork = max_peers_per_ip * 4;
+		ipv6_subnetwork_prefix_for_limiting = 64; // Equivalent to network prefix /64.
 		peer_dump_interval = is_dev_network () ? std::chrono::seconds (1) : std::chrono::seconds (5 * 60);
 	}
 
@@ -188,12 +190,14 @@ class network_constants
 	}
 	/** Default maximum idle time for a socket before it's automatically closed */
 	std::chrono::seconds idle_timeout;
+	std::chrono::seconds silent_connection_tolerance_time;
 	std::chrono::seconds syn_cookie_cutoff;
 	std::chrono::seconds bootstrap_interval;
-	/** Maximum number of peers per IP */
+	/** Maximum number of peers per IP. It is also the max number of connections per IP */
 	size_t max_peers_per_ip;
 	/** Maximum number of peers per subnetwork */
 	size_t max_peers_per_subnetwork;
+	size_t ipv6_subnetwork_prefix_for_limiting;
 	std::chrono::seconds peer_dump_interval;
 
 	/** Returns the network this object contains values for */
diff --git a/nano/lib/stats.cpp b/nano/lib/stats.cpp
index 7df9eab711..3325ba0d8b 100644
--- a/nano/lib/stats.cpp
+++ b/nano/lib/stats.cpp
@@ -775,6 +775,15 @@ std::string nano::stat::detail_to_string (uint32_t key)
 		case nano::stat::detail::tcp_max_per_ip:
 			res = "tcp_max_per_ip";
 			break;
+		case nano::stat::detail::tcp_max_per_subnetwork:
+			res = "tcp_max_per_subnetwork";
+			break;
+		case nano::stat::detail::tcp_silent_connection_drop:
+			res = "tcp_silent_connection_drop";
+			break;
+		case nano::stat::detail::tcp_io_timeout_drop:
+			res = "tcp_io_timeout_drop";
+			break;
 		case nano::stat::detail::unreachable_host:
 			res = "unreachable_host";
 			break;
@@ -808,6 +817,12 @@ std::string nano::stat::detail_to_string (uint32_t key)
 		case nano::stat::detail::outdated_version:
 			res = "outdated_version";
 			break;
+		case nano::stat::detail::udp_max_per_ip:
+			res = "udp_max_per_ip";
+			break;
+		case nano::stat::detail::udp_max_per_subnetwork:
+			res = "udp_max_per_subnetwork";
+			break;
 		case nano::stat::detail::blocks_confirmed:
 			res = "blocks_confirmed";
 			break;
diff --git a/nano/lib/stats.hpp b/nano/lib/stats.hpp
index ea5ba2a1fe..3eed7d4014 100644
--- a/nano/lib/stats.hpp
+++ b/nano/lib/stats.hpp
@@ -337,6 +337,8 @@ class stat final
 		invalid_telemetry_req_message,
 		invalid_telemetry_ack_message,
 		outdated_version,
+		udp_max_per_ip,
+		udp_max_per_subnetwork,
 
 		// tcp
 		tcp_accept_success,
@@ -345,6 +347,9 @@ class stat final
 		tcp_write_no_socket_drop,
 		tcp_excluded,
 		tcp_max_per_ip,
+		tcp_max_per_subnetwork,
+		tcp_silent_connection_drop,
+		tcp_io_timeout_drop,
 
 		// ipc
 		invocations,
diff --git a/nano/node/bootstrap/bootstrap_server.cpp b/nano/node/bootstrap/bootstrap_server.cpp
index b7632ef717..a4993cd6d7 100644
--- a/nano/node/bootstrap/bootstrap_server.cpp
+++ b/nano/node/bootstrap/bootstrap_server.cpp
@@ -749,5 +749,5 @@ bool nano::bootstrap_server::is_bootstrap_connection ()
 
 bool nano::bootstrap_server::is_realtime_connection ()
 {
-	return socket->type () == nano::socket::type_t::realtime || socket->type () == nano::socket::type_t::realtime_response_server;
+	return socket->is_realtime_connection ();
 }
diff --git a/nano/node/nodeconfig.hpp b/nano/node/nodeconfig.hpp
index 8dab45086b..54a219906b 100644
--- a/nano/node/nodeconfig.hpp
+++ b/nano/node/nodeconfig.hpp
@@ -144,6 +144,7 @@ class node_flags final
 	bool disable_block_processor_republishing{ false };
 	bool allow_bootstrap_peers_duplicates{ false };
 	bool disable_max_peers_per_ip{ false }; // For testing only
+	bool disable_max_peers_per_subnetwork{ false }; // For testing only
 	bool force_use_write_database_queue{ false }; // For testing only. RocksDB does not use the database queue, but some tests rely on it being used.
 	bool disable_search_pending{ false }; // For testing only
 	bool enable_pruning{ false };
diff --git a/nano/node/socket.cpp b/nano/node/socket.cpp
index 7e6ad87bd7..1e4d78d7f6 100644
--- a/nano/node/socket.cpp
+++ b/nano/node/socket.cpp
@@ -1,12 +1,19 @@
 #include <nano/boost/asio/bind_executor.hpp>
 #include <nano/boost/asio/dispatch.hpp>
+#include <nano/boost/asio/ip/address.hpp>
+#include <nano/boost/asio/ip/address_v6.hpp>
+#include <nano/boost/asio/ip/network_v6.hpp>
 #include <nano/boost/asio/read.hpp>
 #include <nano/node/node.hpp>
 #include <nano/node/socket.hpp>
+#include <nano/node/transport/transport.hpp>
 
 #include <boost/format.hpp>
 
+#include <cstdint>
+#include <iterator>
 #include <limits>
+#include <memory>
 
 nano::socket::socket (nano::node & node_a) :
 	strand{ node_a.io_ctx.get_executor () },
@@ -14,7 +21,9 @@ nano::socket::socket (nano::node & node_a) :
 	node{ node_a },
 	next_deadline{ std::numeric_limits<uint64_t>::max () },
 	last_completion_time{ 0 },
-	io_timeout{ node_a.config.tcp_io_timeout }
+	last_receive_time{ 0 },
+	io_timeout{ node_a.config.tcp_io_timeout },
+	silent_connection_tolerance_time{ node_a.network_params.network.silent_connection_tolerance_time }
 {
 }
 
@@ -51,6 +60,7 @@ void nano::socket::async_read (std::shared_ptr<std::vector<uint8_t>> const & buf
 				[this_l, buffer_a, callback_a] (boost::system::error_code const & ec, std::size_t size_a) {
 					this_l->node.stats.add (nano::stat::type::traffic_tcp, nano::stat::dir::in, size_a);
 					this_l->stop_timer ();
+					this_l->update_last_receive_time ();
 					callback_a (ec, size_a);
 				}));
 			}));
@@ -117,6 +127,11 @@ void nano::socket::stop_timer ()
 	last_completion_time = nano::seconds_since_epoch ();
 }
 
+void nano::socket::update_last_receive_time ()
+{
+	last_receive_time = nano::seconds_since_epoch ();
+}
+
 void nano::socket::checkup ()
 {
 	std::weak_ptr<nano::socket> this_w (shared_from_this ());
@@ -124,7 +139,18 @@ void nano::socket::checkup ()
 		if (auto this_l = this_w.lock ())
 		{
 			uint64_t now (nano::seconds_since_epoch ());
+			auto condition_to_disconnect{ false };
+			if (this_l->is_realtime_connection () && now - this_l->last_receive_time > this_l->silent_connection_tolerance_time.count ())
+			{
+				this_l->node.stats.inc (nano::stat::type::tcp, nano::stat::detail::tcp_silent_connection_drop, nano::stat::dir::in);
+				condition_to_disconnect = true;
+			}
 			if (this_l->next_deadline != std::numeric_limits<uint64_t>::max () && now - this_l->last_completion_time > this_l->next_deadline)
+			{
+				this_l->node.stats.inc (nano::stat::type::tcp, nano::stat::detail::tcp_io_timeout_drop, nano::stat::dir::in);
+				condition_to_disconnect = true;
+			}
+			if (condition_to_disconnect)
 			{
 				if (this_l->node.config.logging.network_timeout_logging ())
 				{
@@ -157,6 +183,14 @@ void nano::socket::timeout_set (std::chrono::seconds io_timeout_a)
 	io_timeout = io_timeout_a;
 }
 
+void nano::socket::set_silent_connection_tolerance_time (std::chrono::seconds tolerance_time_a)
+{
+	auto this_l (shared_from_this ());
+	boost::asio::dispatch (strand, boost::asio::bind_executor (strand, [this_l, tolerance_time_a] () {
+		this_l->silent_connection_tolerance_time = tolerance_time_a;
+	}));
+}
+
 void nano::socket::close ()
 {
 	auto this_l (shared_from_this ());
@@ -221,17 +255,80 @@ void nano::server_socket::close ()
 	boost::asio::dispatch (strand, boost::asio::bind_executor (strand, [this_l] () {
 		this_l->close_internal ();
 		this_l->acceptor.close ();
-		for (auto & connection_w : this_l->connections)
+		for (auto & address_connection_pair : this_l->connections_per_address)
 		{
-			if (auto connection_l = connection_w.lock ())
+			if (auto connection_l = address_connection_pair.second.lock ())
 			{
 				connection_l->close ();
 			}
 		}
-		this_l->connections.clear ();
+		this_l->connections_per_address.clear ();
 	}));
 }
 
+boost::asio::ip::network_v6 nano::socket_functions::get_ipv6_subnet_address (boost::asio::ip::address_v6 const & ip_address, size_t network_prefix)
+{
+	return boost::asio::ip::make_network_v6 (ip_address, network_prefix);
+}
+
+boost::asio::ip::address nano::socket_functions::first_ipv6_subnet_address (boost::asio::ip::address_v6 const & ip_address, size_t network_prefix)
+{
+	auto range = get_ipv6_subnet_address (ip_address, network_prefix).hosts ();
+	debug_assert (!range.empty ());
+	return *(range.begin ());
+}
+
+boost::asio::ip::address nano::socket_functions::last_ipv6_subnet_address (boost::asio::ip::address_v6 const & ip_address, size_t network_prefix)
+{
+	auto range = get_ipv6_subnet_address (ip_address, network_prefix).hosts ();
+	debug_assert (!range.empty ());
+	return *(--range.end ());
+}
+
+size_t nano::socket_functions::count_subnetwork_connections (
+nano::address_socket_mmap const & per_address_connections,
+boost::asio::ip::address_v6 const & remote_address,
+size_t network_prefix)
+{
+	auto range = get_ipv6_subnet_address (remote_address, network_prefix).hosts ();
+	if (range.empty ())
+	{
+		return 0;
+	}
+	auto const first_ip = first_ipv6_subnet_address (remote_address, network_prefix);
+	auto const last_ip = last_ipv6_subnet_address (remote_address, network_prefix);
+	auto const counted_connections = std::distance (per_address_connections.lower_bound (first_ip), per_address_connections.upper_bound (last_ip));
+	return counted_connections;
+}
+
+bool nano::server_socket::limit_reached_for_incoming_subnetwork_connections (std::shared_ptr<nano::socket> const & new_connection)
+{
+	debug_assert (strand.running_in_this_thread ());
+	if (node.flags.disable_max_peers_per_subnetwork)
+	{
+		// If the limit is disabled, then it is unreachable.
+		return false;
+	}
+	auto const counted_connections = socket_functions::count_subnetwork_connections (
+	connections_per_address,
+	nano::transport::mapped_from_v4_or_v6 (new_connection->remote.address ()),
+	node.network_params.network.ipv6_subnetwork_prefix_for_limiting);
+	return counted_connections >= node.network_params.network.max_peers_per_subnetwork;
+}
+
+bool nano::server_socket::limit_reached_for_incoming_ip_connections (std::shared_ptr<nano::socket> const & new_connection)
+{
+	debug_assert (strand.running_in_this_thread ());
+	if (node.flags.disable_max_peers_per_ip)
+	{
+		// If the limit is disabled, then it is unreachable.
+		return false;
+	}
+	auto const address_connections_range = connections_per_address.equal_range (new_connection->remote.address ());
+	auto const counted_connections = std::distance (address_connections_range.first, address_connections_range.second);
+	return counted_connections >= node.network_params.network.max_peers_per_ip;
+}
+
 void nano::server_socket::on_connection (std::function<bool (std::shared_ptr<nano::socket> const &, boost::system::error_code const &)> callback_a)
 {
 	auto this_l (std::static_pointer_cast<nano::server_socket> (shared_from_this ()));
@@ -250,14 +347,41 @@ void nano::server_socket::on_connection (std::function<bool (std::shared_ptr<nan
 		[this_l, new_connection, callback_a] (boost::system::error_code const & ec_a) {
 			this_l->evict_dead_connections ();
 
-			if (this_l->connections.size () >= this_l->max_inbound_connections)
+			if (this_l->connections_per_address.size () >= this_l->max_inbound_connections)
 			{
-				this_l->node.logger.always_log ("Network: max_inbound_connections reached, unable to open new connection");
+				this_l->node.logger.try_log ("Network: max_inbound_connections reached, unable to open new connection");
 				this_l->node.stats.inc (nano::stat::type::tcp, nano::stat::detail::tcp_accept_failure, nano::stat::dir::in);
 				this_l->on_connection_requeue_delayed (callback_a);
 				return;
 			}
 
+			if (this_l->limit_reached_for_incoming_ip_connections (new_connection))
+			{
+				auto const remote_ip_address = new_connection->remote_endpoint ().address ();
+				auto const log_message = boost::str (
+				boost::format ("Network: max connections per IP (max_peers_per_ip) was reached for %1%, unable to open new connection")
+				% remote_ip_address.to_string ());
+				this_l->node.logger.try_log (log_message);
+				this_l->node.stats.inc (nano::stat::type::tcp, nano::stat::detail::tcp_max_per_ip, nano::stat::dir::in);
+				this_l->on_connection_requeue_delayed (callback_a);
+				return;
+			}
+
+			if (this_l->limit_reached_for_incoming_subnetwork_connections (new_connection))
+			{
+				auto const remote_ip_address = new_connection->remote_endpoint ().address ();
+				debug_assert (remote_ip_address.is_v6 ());
+				auto const remote_subnet = socket_functions::get_ipv6_subnet_address (remote_ip_address.to_v6 (), this_l->node.network_params.network.max_peers_per_subnetwork);
+				auto const log_message = boost::str (
+				boost::format ("Network: max connections per subnetwork (max_peers_per_subnetwork) was reached for subnetwork %1% (remote IP: %2%), unable to open new connection")
+				% remote_subnet.canonical ().to_string ()
+				% remote_ip_address.to_string ());
+				this_l->node.logger.try_log (log_message);
+				this_l->node.stats.inc (nano::stat::type::tcp, nano::stat::detail::tcp_max_per_subnetwork, nano::stat::dir::in);
+				this_l->on_connection_requeue_delayed (callback_a);
+				return;
+			}
+
 			if (!ec_a)
 			{
 				// Make sure the new connection doesn't idle. Note that in most cases, the callback is going to start
@@ -265,7 +389,7 @@ void nano::server_socket::on_connection (std::function<bool (std::shared_ptr<nan
 				new_connection->checkup ();
 				new_connection->start_timer (this_l->node.network_params.network.is_dev_network () ? std::chrono::seconds (2) : this_l->node.network_params.network.idle_timeout);
 				this_l->node.stats.inc (nano::stat::type::tcp, nano::stat::detail::tcp_accept_success, nano::stat::dir::in);
-				this_l->connections.push_back (new_connection);
+				this_l->connections_per_address.emplace (new_connection->remote.address (), new_connection);
 				if (callback_a (new_connection, ec_a))
 				{
 					this_l->on_connection (callback_a);
@@ -329,5 +453,13 @@ bool nano::server_socket::is_temporary_error (boost::system::error_code const ec
 void nano::server_socket::evict_dead_connections ()
 {
 	debug_assert (strand.running_in_this_thread ());
-	connections.erase (std::remove_if (connections.begin (), connections.end (), [] (auto & connection) { return connection.expired (); }), connections.end ());
+	for (auto it = connections_per_address.begin (); it != connections_per_address.end ();)
+	{
+		if (it->second.expired ())
+		{
+			it = connections_per_address.erase (it);
+			continue;
+		}
+		++it;
+	}
 }
diff --git a/nano/node/socket.hpp b/nano/node/socket.hpp
index 1b4868fd3d..cda8c2bc03 100644
--- a/nano/node/socket.hpp
+++ b/nano/node/socket.hpp
@@ -8,9 +8,21 @@
 
 #include <chrono>
 #include <deque>
+#include <map>
 #include <memory>
 #include <vector>
 
+namespace boost
+{
+namespace asio
+{
+	namespace ip
+	{
+		class network_v6;
+	}
+}
+}
+
 namespace nano
 {
 /** Policy to affect at which stage a buffer can be dropped */
@@ -60,6 +72,7 @@ class socket : public std::enable_shared_from_this<nano::socket>
 	/** This can be called to change the maximum idle time, e.g. based on the type of traffic detected. */
 	void timeout_set (std::chrono::seconds io_timeout_a);
 	void start_timer (std::chrono::seconds deadline_a);
+	void set_silent_connection_tolerance_time (std::chrono::seconds tolerance_time_a);
 	bool max () const
 	{
 		return queue_size >= queue_size_max;
@@ -76,6 +89,14 @@ class socket : public std::enable_shared_from_this<nano::socket>
 	{
 		type_m = type_a;
 	}
+	bool is_realtime_connection ()
+	{
+		return type () == nano::socket::type_t::realtime || type () == nano::socket::type_t::realtime_response_server;
+	}
+	bool is_closed ()
+	{
+		return closed;
+	}
 
 protected:
 	/** Holds the buffer and callback for queued writes */
@@ -95,8 +116,10 @@ class socket : public std::enable_shared_from_this<nano::socket>
 
 	std::atomic<uint64_t> next_deadline;
 	std::atomic<uint64_t> last_completion_time;
+	std::atomic<uint64_t> last_receive_time;
 	std::atomic<bool> timed_out{ false };
 	std::atomic<std::chrono::seconds> io_timeout;
+	std::chrono::seconds silent_connection_tolerance_time;
 	std::atomic<std::size_t> queue_size{ 0 };
 
 	/** Set by close() - completion handlers must check this. This is more reliable than checking
@@ -105,6 +128,7 @@ class socket : public std::enable_shared_from_this<nano::socket>
 	void close_internal ();
 	void start_timer ();
 	void stop_timer ();
+	void update_last_receive_time ();
 	void checkup ();
 
 private:
@@ -114,6 +138,16 @@ class socket : public std::enable_shared_from_this<nano::socket>
 	static std::size_t constexpr queue_size_max = 128;
 };
 
+using address_socket_mmap = std::multimap<boost::asio::ip::address, std::weak_ptr<socket>>;
+
+namespace socket_functions
+{
+	boost::asio::ip::network_v6 get_ipv6_subnet_address (boost::asio::ip::address_v6 const &, size_t);
+	boost::asio::ip::address first_ipv6_subnet_address (boost::asio::ip::address_v6 const &, size_t);
+	boost::asio::ip::address last_ipv6_subnet_address (boost::asio::ip::address_v6 const &, size_t);
+	size_t count_subnetwork_connections (nano::address_socket_mmap const &, boost::asio::ip::address_v6 const &, size_t);
+}
+
 /** Socket class for TCP servers */
 class server_socket final : public socket
 {
@@ -138,12 +172,15 @@ class server_socket final : public socket
 	}
 
 private:
-	std::vector<std::weak_ptr<nano::socket>> connections;
+	nano::address_socket_mmap connections_per_address;
 	boost::asio::ip::tcp::acceptor acceptor;
 	boost::asio::ip::tcp::endpoint local;
 	std::size_t max_inbound_connections;
 	void evict_dead_connections ();
 	bool is_temporary_error (boost::system::error_code const ec_a);
 	void on_connection_requeue_delayed (std::function<bool (std::shared_ptr<nano::socket> const & new_connection, boost::system::error_code const &)>);
+	/** Checks whether the maximum number of connections per IP was reached. If so, it returns true. */
+	bool limit_reached_for_incoming_ip_connections (std::shared_ptr<nano::socket> const & new_connection);
+	bool limit_reached_for_incoming_subnetwork_connections (std::shared_ptr<nano::socket> const & new_connection);
 };
 }
diff --git a/nano/node/transport/tcp.cpp b/nano/node/transport/tcp.cpp
index 75ba49ded6..d5bfffef08 100644
--- a/nano/node/transport/tcp.cpp
+++ b/nano/node/transport/tcp.cpp
@@ -362,17 +362,17 @@ void nano::transport::tcp_channels::stop ()
 
 bool nano::transport::tcp_channels::max_ip_connections (nano::tcp_endpoint const & endpoint_a)
 {
-	bool result (false);
-	if (!node.flags.disable_max_peers_per_ip)
+	if (node.flags.disable_max_peers_per_ip)
 	{
-		auto const address (nano::transport::ipv4_address_or_ipv6_subnet (endpoint_a.address ()));
-		auto const subnet (nano::transport::map_address_to_subnetwork (endpoint_a.address ()));
-		nano::unique_lock<nano::mutex> lock (mutex);
-		result = channels.get<ip_address_tag> ().count (address) >= node.network_params.network.max_peers_per_ip || channels.get<subnetwork_tag> ().count (subnet) >= node.network_params.network.max_peers_per_subnetwork;
-		if (!result)
-		{
-			result = attempts.get<ip_address_tag> ().count (address) >= node.network_params.network.max_peers_per_ip || attempts.get<subnetwork_tag> ().count (subnet) >= node.network_params.network.max_peers_per_subnetwork;
-		}
+		return false;
+	}
+	bool result{ false };
+	auto const address (nano::transport::ipv4_address_or_ipv6_subnet (endpoint_a.address ()));
+	nano::unique_lock<nano::mutex> lock (mutex);
+	result = channels.get<ip_address_tag> ().count (address) >= node.network_params.network.max_peers_per_ip;
+	if (!result)
+	{
+		result = attempts.get<ip_address_tag> ().count (address) >= node.network_params.network.max_peers_per_ip;
 	}
 	if (result)
 	{
@@ -381,11 +381,37 @@ bool nano::transport::tcp_channels::max_ip_connections (nano::tcp_endpoint const
 	return result;
 }
 
+bool nano::transport::tcp_channels::max_subnetwork_connections (nano::tcp_endpoint const & endpoint_a)
+{
+	if (node.flags.disable_max_peers_per_subnetwork)
+	{
+		return false;
+	}
+	bool result{ false };
+	auto const subnet (nano::transport::map_address_to_subnetwork (endpoint_a.address ()));
+	nano::unique_lock<nano::mutex> lock (mutex);
+	result = channels.get<subnetwork_tag> ().count (subnet) >= node.network_params.network.max_peers_per_subnetwork;
+	if (!result)
+	{
+		result = attempts.get<subnetwork_tag> ().count (subnet) >= node.network_params.network.max_peers_per_subnetwork;
+	}
+	if (result)
+	{
+		node.stats.inc (nano::stat::type::tcp, nano::stat::detail::tcp_max_per_subnetwork, nano::stat::dir::out);
+	}
+	return result;
+}
+
+bool nano::transport::tcp_channels::max_ip_or_subnetwork_connections (nano::tcp_endpoint const & endpoint_a)
+{
+	return max_ip_connections (endpoint_a) || max_subnetwork_connections (endpoint_a);
+}
+
 bool nano::transport::tcp_channels::reachout (nano::endpoint const & endpoint_a)
 {
 	auto tcp_endpoint (nano::transport::map_endpoint_to_tcp (endpoint_a));
 	// Don't overload single IP
-	bool error = node.network.excluded_peers.check (tcp_endpoint) || max_ip_connections (tcp_endpoint);
+	bool error = node.network.excluded_peers.check (tcp_endpoint) || max_ip_or_subnetwork_connections (tcp_endpoint);
 	if (!error && !node.flags.disable_tcp_realtime)
 	{
 		// Don't keepalive to nodes that already sent us something
diff --git a/nano/node/transport/tcp.hpp b/nano/node/transport/tcp.hpp
index 611791b508..353f35de37 100644
--- a/nano/node/transport/tcp.hpp
+++ b/nano/node/transport/tcp.hpp
@@ -91,7 +91,9 @@ namespace transport
 		void stop ();
 		void process_messages ();
 		void process_message (nano::message const &, nano::tcp_endpoint const &, nano::account const &, std::shared_ptr<nano::socket> const &);
-		bool max_ip_connections (nano::tcp_endpoint const &);
+		bool max_ip_connections (nano::tcp_endpoint const & endpoint_a);
+		bool max_subnetwork_connections (nano::tcp_endpoint const & endpoint_a);
+		bool max_ip_or_subnetwork_connections (nano::tcp_endpoint const & endpoint_a);
 		// Should we reach out to this endpoint with a keepalive message
 		bool reachout (nano::endpoint const &);
 		std::unique_ptr<container_info_component> collect_container_info (std::string const &);
diff --git a/nano/node/transport/transport.cpp b/nano/node/transport/transport.cpp
index 718df5d39b..79998ec594 100644
--- a/nano/node/transport/transport.cpp
+++ b/nano/node/transport/transport.cpp
@@ -2,6 +2,9 @@
 #include <nano/node/node.hpp>
 #include <nano/node/transport/transport.hpp>
 
+#include <boost/asio/ip/address.hpp>
+#include <boost/asio/ip/address_v4.hpp>
+#include <boost/asio/ip/address_v6.hpp>
 #include <boost/format.hpp>
 
 #include <numeric>
@@ -82,7 +85,7 @@ nano::tcp_endpoint nano::transport::map_endpoint_to_tcp (nano::endpoint const &
 boost::asio::ip::address nano::transport::map_address_to_subnetwork (boost::asio::ip::address const & address_a)
 {
 	debug_assert (address_a.is_v6 ());
-	static short const ipv6_subnet_prefix_length = 32; // Limits for /32 IPv6 subnetwork
+	static short const ipv6_subnet_prefix_length = 32; // Equivalent to network prefix /32.
 	static short const ipv4_subnet_prefix_length = (128 - 32) + 24; // Limits for /24 IPv4 subnetwork
 	return address_a.to_v6 ().is_v4_mapped () ? boost::asio::ip::make_network_v6 (address_a.to_v6 (), ipv4_subnet_prefix_length).network () : boost::asio::ip::make_network_v6 (address_a.to_v6 (), ipv6_subnet_prefix_length).network ();
 }
@@ -159,12 +162,14 @@ std::string nano::transport::channel_loopback::to_string () const
 	return boost::str (boost::format ("%1%") % endpoint);
 }
 
-namespace
-{
-boost::asio::ip::address_v6 mapped_from_v4_bytes (unsigned long address_a)
+boost::asio::ip::address_v6 nano::transport::mapped_from_v4_bytes (unsigned long address_a)
 {
 	return boost::asio::ip::address_v6::v4_mapped (boost::asio::ip::address_v4 (address_a));
 }
+
+boost::asio::ip::address_v6 nano::transport::mapped_from_v4_or_v6 (boost::asio::ip::address const & address_a)
+{
+	return address_a.is_v4 () ? boost::asio::ip::address_v6::v4_mapped (address_a.to_v4 ()) : address_a.to_v6 ();
 }
 
 bool nano::transport::reserved_address (nano::endpoint const & endpoint_a, bool allow_local_peers)
diff --git a/nano/node/transport/transport.hpp b/nano/node/transport/transport.hpp
index 56e5f6b134..66af0baa48 100644
--- a/nano/node/transport/transport.hpp
+++ b/nano/node/transport/transport.hpp
@@ -30,6 +30,8 @@ namespace transport
 	nano::tcp_endpoint map_endpoint_to_tcp (nano::endpoint const &);
 	boost::asio::ip::address map_address_to_subnetwork (boost::asio::ip::address const &);
 	boost::asio::ip::address ipv4_address_or_ipv6_subnet (boost::asio::ip::address const &);
+	boost::asio::ip::address_v6 mapped_from_v4_bytes (unsigned long);
+	boost::asio::ip::address_v6 mapped_from_v4_or_v6 (boost::asio::ip::address const &);
 	// Unassigned, reserved, self
 	bool reserved_address (nano::endpoint const &, bool = false);
 	static std::chrono::seconds constexpr syn_cookie_cutoff = std::chrono::seconds (5);
diff --git a/nano/node/transport/udp.cpp b/nano/node/transport/udp.cpp
index 9b4885ba9b..8013e1a81c 100644
--- a/nano/node/transport/udp.cpp
+++ b/nano/node/transport/udp.cpp
@@ -100,7 +100,7 @@ std::shared_ptr<nano::transport::channel_udp> nano::transport::udp_channels::ins
 {
 	debug_assert (endpoint_a.address ().is_v6 ());
 	std::shared_ptr<nano::transport::channel_udp> result;
-	if (!node.network.not_a_peer (endpoint_a, node.config.allow_local_peers) && (node.network_params.network.is_dev_network () || !max_ip_connections (endpoint_a)))
+	if (!node.network.not_a_peer (endpoint_a, node.config.allow_local_peers) && (node.network_params.network.is_dev_network () || !max_ip_or_subnetwork_connections (endpoint_a)))
 	{
 		nano::unique_lock<nano::mutex> lock (mutex);
 		auto existing (channels.get<endpoint_tag> ().find (endpoint_a));
@@ -373,7 +373,7 @@ class udp_message_visitor : public nano::message_visitor
 	}
 	void keepalive (nano::keepalive const & message_a) override
 	{
-		if (!node.network.udp_channels.max_ip_connections (endpoint))
+		if (!node.network.udp_channels.max_ip_or_subnetwork_connections (endpoint))
 		{
 			auto cookie (node.network.syn_cookies.assign (endpoint));
 			if (cookie)
@@ -630,21 +630,45 @@ std::shared_ptr<nano::transport::channel> nano::transport::udp_channels::create
 
 bool nano::transport::udp_channels::max_ip_connections (nano::endpoint const & endpoint_a)
 {
-	bool result (false);
-	if (!node.flags.disable_max_peers_per_ip)
+	if (node.flags.disable_max_peers_per_ip)
 	{
-		auto const address (nano::transport::ipv4_address_or_ipv6_subnet (endpoint_a.address ()));
-		auto const subnet (nano::transport::map_address_to_subnetwork (endpoint_a.address ()));
-		nano::unique_lock<nano::mutex> lock (mutex);
-		result = channels.get<ip_address_tag> ().count (address) >= node.network_params.network.max_peers_per_ip || channels.get<subnetwork_tag> ().count (subnet) >= node.network_params.network.max_peers_per_subnetwork;
+		return false;
+	}
+	auto const address (nano::transport::ipv4_address_or_ipv6_subnet (endpoint_a.address ()));
+	nano::unique_lock<nano::mutex> lock (mutex);
+	auto const result = channels.get<ip_address_tag> ().count (address) >= node.network_params.network.max_peers_per_ip;
+	if (!result)
+	{
+		node.stats.inc (nano::stat::type::udp, nano::stat::detail::udp_max_per_ip, nano::stat::dir::out);
 	}
 	return result;
 }
 
+bool nano::transport::udp_channels::max_subnetwork_connections (nano::endpoint const & endpoint_a)
+{
+	if (node.flags.disable_max_peers_per_subnetwork)
+	{
+		return false;
+	}
+	auto const subnet (nano::transport::map_address_to_subnetwork (endpoint_a.address ()));
+	nano::unique_lock<nano::mutex> lock (mutex);
+	auto const result = channels.get<subnetwork_tag> ().count (subnet) >= node.network_params.network.max_peers_per_subnetwork;
+	if (!result)
+	{
+		node.stats.inc (nano::stat::type::udp, nano::stat::detail::udp_max_per_subnetwork, nano::stat::dir::out);
+	}
+	return result;
+}
+
+bool nano::transport::udp_channels::max_ip_or_subnetwork_connections (nano::endpoint const & endpoint_a)
+{
+	return max_ip_connections (endpoint_a) || max_subnetwork_connections (endpoint_a);
+}
+
 bool nano::transport::udp_channels::reachout (nano::endpoint const & endpoint_a)
 {
 	// Don't overload single IP
-	bool error = max_ip_connections (endpoint_a);
+	bool error = max_ip_or_subnetwork_connections (endpoint_a);
 	if (!error && !node.flags.disable_udp)
 	{
 		auto endpoint_l (nano::transport::map_endpoint_to_v6 (endpoint_a));
diff --git a/nano/node/transport/udp.hpp b/nano/node/transport/udp.hpp
index 6eb0e9bc3c..25e88e56d1 100644
--- a/nano/node/transport/udp.hpp
+++ b/nano/node/transport/udp.hpp
@@ -96,7 +96,9 @@ namespace transport
 		void receive_action (nano::message_buffer *);
 		void process_packets ();
 		std::shared_ptr<nano::transport::channel> create (nano::endpoint const &);
-		bool max_ip_connections (nano::endpoint const &);
+		bool max_ip_connections (nano::endpoint const & endpoint_a);
+		bool max_subnetwork_connections (nano::endpoint const & endpoint_a);
+		bool max_ip_or_subnetwork_connections (nano::endpoint const & endpoint_a);
 		// Should we reach out to this endpoint with a keepalive message
 		bool reachout (nano::endpoint const &);
 		std::unique_ptr<container_info_component> collect_container_info (std::string const &);