diff --git a/ceilometer/gnocchi_client.py b/ceilometer/gnocchi_client.py
index e61a47a9bf..6443ad9a5d 100644
--- a/ceilometer/gnocchi_client.py
+++ b/ceilometer/gnocchi_client.py
@@ -13,6 +13,7 @@
 
 from gnocchiclient import client
 from gnocchiclient import exceptions as gnocchi_exc
+import keystoneauth1.session
 from oslo_log import log
 
 from ceilometer import keystone_client
@@ -26,6 +27,10 @@ def get_gnocchiclient(conf, timeout_override=False):
                         timeout_override)
                else conf.dispatcher_gnocchi.request_timeout)
     session = keystone_client.get_session(conf, group=group, timeout=timeout)
+    adapter = keystoneauth1.session.TCPKeepAliveAdapter(
+        pool_maxsize=conf.max_parallel_requests)
+    session.mount("http://", adapter)
+    session.mount("https://", adapter)
     return client.Client('1', session,
                          interface=conf[group].interface,
                          region_name=conf[group].region_name,
diff --git a/ceilometer/notification.py b/ceilometer/notification.py
index d571713088..0f28046a5d 100644
--- a/ceilometer/notification.py
+++ b/ceilometer/notification.py
@@ -260,7 +260,9 @@ class NotificationService(cotyledon.Service):
             # to maintain sequencing as much as possible.
             listener = messaging.get_batch_notification_listener(
                 transport, targets, endpoints)
-            listener.start()
+            listener.start(
+                override_pool_size=self.conf.max_parallel_requests
+            )
             self.listeners.append(listener)
 
     def _refresh_agent(self, event):
@@ -312,7 +314,8 @@ class NotificationService(cotyledon.Service):
             batch_timeout=self.conf.notification.batch_timeout)
         # NOTE(gordc): set single thread to process data sequentially
         # if batching enabled.
-        batch = (1 if self.conf.notification.batch_size > 1 else None)
+        batch = (1 if self.conf.notification.batch_size > 1
+                 else self.conf.max_parallel_requests)
         self.pipeline_listener.start(override_pool_size=batch)
 
     def terminate(self):
diff --git a/ceilometer/opts.py b/ceilometer/opts.py
index 52ba520da7..73a2c23a3c 100644
--- a/ceilometer/opts.py
+++ b/ceilometer/opts.py
@@ -68,6 +68,11 @@ OPTS = [
                default=600,
                help='Timeout seconds for HTTP requests. Set it to None to '
                     'disable timeout.'),
+    cfg.IntOpt('max_parallel_requests',
+               default=64,
+               min=1,
+               help='Maximum number of parallel requests for '
+               'services to handle at the same time.'),
 ]
 
 
diff --git a/ceilometer/publisher/http.py b/ceilometer/publisher/http.py
index 2dd41105f4..2075ba0590 100644
--- a/ceilometer/publisher/http.py
+++ b/ceilometer/publisher/http.py
@@ -37,7 +37,6 @@ class HttpPublisher(publisher.ConfigPublisherBase):
         - ssl certificate verification can be disabled by setting `verify_ssl`
           to False
         - batching can be configured by `batch`
-        - connection pool size configured using `poolsize`
         - Basic authentication can be configured using the URL authentication
           scheme: http://username:password@example.com
         - For certificate authentication, `clientcert` and `clientkey` are the
@@ -121,9 +120,9 @@ class HttpPublisher(publisher.ConfigPublisherBase):
         self.raw_only = strutils.bool_from_string(
             self._get_param(params, 'raw_only', False))
 
-        pool_size = self._get_param(params, 'poolsize', 10, int)
         kwargs = {'max_retries': self.max_retries,
-                  'pool_connections': pool_size, 'pool_maxsize': pool_size}
+                  'pool_connections': conf.max_parallel_requests,
+                  'pool_maxsize': conf.max_parallel_requests}
         self.session = requests.Session()
 
         # authentication & config params have been removed, so use URL with
diff --git a/ceilometer/tests/functional/test_notification.py b/ceilometer/tests/functional/test_notification.py
index 3c740d8046..b19610ea0c 100644
--- a/ceilometer/tests/functional/test_notification.py
+++ b/ceilometer/tests/functional/test_notification.py
@@ -304,7 +304,8 @@ class TestRealNotificationHA(BaseRealNotification):
     def test_notification_threads(self, m_listener):
         self.CONF.set_override('batch_size', 1, group='notification')
         self.srv.run()
-        m_listener.assert_called_with(override_pool_size=None)
+        m_listener.assert_called_with(
+            override_pool_size=self.CONF.max_parallel_requests)
         m_listener.reset_mock()
         self.CONF.set_override('batch_size', 2, group='notification')
         self.srv.run()
diff --git a/releasenotes/notes/parallel_requests_option-a3f901b6001e26e4.yaml b/releasenotes/notes/parallel_requests_option-a3f901b6001e26e4.yaml
new file mode 100644
index 0000000000..f3c1465c77
--- /dev/null
+++ b/releasenotes/notes/parallel_requests_option-a3f901b6001e26e4.yaml
@@ -0,0 +1,6 @@
+---
+features:
+  - |
+    A new option named `max_parallel_requests` is available to control the
+    maximum number of parallel requests that can be executed by the agents.
+    This option also replaces the `poolsize` option of the HTTP publisher.