diff --git a/cinder/common/config.py b/cinder/common/config.py
index 9a987d14966..604028f2e1b 100644
--- a/cinder/common/config.py
+++ b/cinder/common/config.py
@@ -117,7 +117,9 @@ global_opts = [
     # NOTE(vish): default to nova for compatibility with nova installs
     cfg.StrOpt('storage_availability_zone',
                default='nova',
-               help='Availability zone of this node'),
+               help='Availability zone of this node. Can be overridden per '
+                    'volume backend with the option '
+                    '"backend_availability_zone".'),
     cfg.StrOpt('default_availability_zone',
                help='Default availability zone for new volumes. If not set, '
                     'the storage_availability_zone option value is used as '
diff --git a/cinder/manager.py b/cinder/manager.py
index 3091c7a4f57..d5bc6ef25bf 100644
--- a/cinder/manager.py
+++ b/cinder/manager.py
@@ -91,6 +91,7 @@ class Manager(base.Base, PeriodicTasks):
         self.host = host
         self.cluster = cluster
         self.additional_endpoints = []
+        self.availability_zone = CONF.storage_availability_zone
         super(Manager, self).__init__(db_driver)
 
     @property
diff --git a/cinder/service.py b/cinder/service.py
index 39f061a9227..1a5295d8c71 100644
--- a/cinder/service.py
+++ b/cinder/service.py
@@ -145,6 +145,11 @@ class Service(service.Service):
             manager_class = profiler.trace_cls("rpc")(manager_class)
 
         self.service = None
+        self.manager = manager_class(host=self.host,
+                                     cluster=self.cluster,
+                                     service_name=service_name,
+                                     *args, **kwargs)
+        self.availability_zone = self.manager.availability_zone
 
         # NOTE(geguileo): We need to create the Service DB entry before we
         # create the manager, otherwise capped versions for serializer and rpc
@@ -201,10 +206,6 @@ class Service(service.Service):
             # start while we are still doing the rolling upgrade.
             self.added_to_cluster = not self.is_upgrading_to_n
 
-        self.manager = manager_class(host=self.host,
-                                     cluster=self.cluster,
-                                     service_name=service_name,
-                                     *args, **kwargs)
         self.report_interval = report_interval
         self.periodic_interval = periodic_interval
         self.periodic_fuzzy_delay = periodic_fuzzy_delay
@@ -356,13 +357,12 @@ class Service(service.Service):
                     pass
 
     def _create_service_ref(self, context, rpc_version=None):
-        zone = CONF.storage_availability_zone
         kwargs = {
             'host': self.host,
             'binary': self.binary,
             'topic': self.topic,
             'report_count': 0,
-            'availability_zone': zone,
+            'availability_zone': self.availability_zone,
             'rpc_current_version': rpc_version or self.manager.RPC_API_VERSION,
             'object_current_version': objects_base.OBJ_VERSIONS.get_current(),
         }
@@ -486,7 +486,6 @@ class Service(service.Service):
             return
 
         ctxt = context.get_admin_context()
-        zone = CONF.storage_availability_zone
         try:
             try:
                 service_ref = objects.Service.get_by_id(ctxt,
@@ -499,8 +498,8 @@ class Service(service.Service):
                                                         Service.service_id)
 
             service_ref.report_count += 1
-            if zone != service_ref.availability_zone:
-                service_ref.availability_zone = zone
+            if self.availability_zone != service_ref.availability_zone:
+                service_ref.availability_zone = self.availability_zone
 
             service_ref.save()
 
diff --git a/cinder/tests/unit/volume/test_volume.py b/cinder/tests/unit/volume/test_volume.py
index 90fb954d170..7a15851e9bb 100644
--- a/cinder/tests/unit/volume/test_volume.py
+++ b/cinder/tests/unit/volume/test_volume.py
@@ -116,6 +116,36 @@ class VolumeTestCase(base.BaseVolumeTestCase):
                          scheduler_rpcapi.client.serializer._base.version_cap)
         self.assertIsNone(scheduler_rpcapi.client.serializer._base.manifest)
 
+    @mock.patch('oslo_utils.importutils.import_object')
+    def test_backend_availability_zone(self, mock_import_object):
+        # NOTE(smcginnis): This isn't really the best place for this test,
+        # but we don't currently have a pure VolumeManager test class. So
+        # until we create a good suite for that class, putting here with
+        # other tests that use VolumeManager.
+
+        opts = {
+            'backend_availability_zone': 'caerbannog'
+        }
+
+        def conf_get(option):
+            if option in opts:
+                return opts[option]
+            return None
+
+        mock_driver = mock.Mock()
+        mock_driver.configuration.safe_get.side_effect = conf_get
+        mock_driver.configuration.extra_capabilities = 'null'
+
+        def import_obj(*args, **kwargs):
+            return mock_driver
+
+        mock_import_object.side_effect = import_obj
+
+        manager = vol_manager.VolumeManager(volume_driver=mock_driver)
+        self.assertIsNotNone(manager)
+        self.assertEqual(opts['backend_availability_zone'],
+                         manager.availability_zone)
+
     @mock.patch.object(vol_manager.VolumeManager,
                        'update_service_capabilities')
     def test_report_filter_goodness_function(self, mock_update):
diff --git a/cinder/volume/driver.py b/cinder/volume/driver.py
index ba13a2f71a7..1477d114795 100644
--- a/cinder/volume/driver.py
+++ b/cinder/volume/driver.py
@@ -266,6 +266,11 @@ volume_opts = [
                      "working CI system and testing are marked as unsupported "
                      "until CI is working again.  This also marks a driver as "
                      "deprecated and may be removed in the next release."),
+    cfg.StrOpt('backend_availability_zone',
+               default=None,
+               help='Availability zone for this volume backend. If not set, '
+                    'the storage_availability_zone option value is used as '
+                    'the default for all backends.'),
 ]
 
 # for backward compatibility
diff --git a/cinder/volume/manager.py b/cinder/volume/manager.py
index 36804ec8ddc..b89a8c5f0f1 100644
--- a/cinder/volume/manager.py
+++ b/cinder/volume/manager.py
@@ -268,6 +268,12 @@ class VolumeManager(manager.CleanableManager,
                 LOG.error(_LE("Invalid JSON: %s"),
                           self.driver.configuration.extra_capabilities)
 
+        # Check if a per-backend AZ has been specified
+        backend_zone = self.driver.configuration.safe_get(
+            'backend_availability_zone')
+        if backend_zone:
+            self.availability_zone = backend_zone
+
         if self.driver.configuration.safe_get(
                 'image_volume_cache_enabled'):
 
diff --git a/releasenotes/notes/per-backend-az-28727aca360a1cc8.yaml b/releasenotes/notes/per-backend-az-28727aca360a1cc8.yaml
new file mode 100644
index 00000000000..db5148e6a52
--- /dev/null
+++ b/releasenotes/notes/per-backend-az-28727aca360a1cc8.yaml
@@ -0,0 +1,7 @@
+---
+features:
+  - |
+    Availability zones may now be configured per backend in a multi-backend
+    configuration. Individual backend sections can now set the configuration
+    option ``backend_availability_zone``. If set, this value will override
+    the [DEFAULT] ``storage_availability_zone`` setting.