diff --git a/glance/common/wsgi.py b/glance/common/wsgi.py
index a94576b80d..13785f2097 100644
--- a/glance/common/wsgi.py
+++ b/glance/common/wsgi.py
@@ -207,7 +207,10 @@ Number of Glance worker processes to start.
 
 Provide a non-negative integer value to set the number of child
 process workers to service requests. By default, the number of CPUs
-available is set as the value for ``workers``.
+available is set as the value for ``workers`` limited to 8. For
+example if the processor count is 6, 6 workers will be used, if the
+processor count is 24 only 8 workers will be used. The limit will only
+apply to the default value, if 24 workers is configured, 24 is used.
 
 Each worker process is made to listen on the port set in the
 configuration file and contains a greenthread pool of size 1000.
@@ -332,8 +335,10 @@ except ImportError:
 def get_num_workers():
     """Return the configured number of workers."""
     if CONF.workers is None:
-        # None implies the number of CPUs
-        return processutils.get_worker_count()
+        # None implies the number of CPUs limited to 8
+        # See Launchpad bug #1748916 and the config help text
+        workers = processutils.get_worker_count()
+        return workers if workers < 8 else 8
     return CONF.workers
 
 
diff --git a/glance/tests/unit/common/test_wsgi.py b/glance/tests/unit/common/test_wsgi.py
index 0f4c5ad566..710d94807e 100644
--- a/glance/tests/unit/common/test_wsgi.py
+++ b/glance/tests/unit/common/test_wsgi.py
@@ -604,7 +604,7 @@ class ServerTest(test_utils.BaseTestCase):
                                                 socket_timeout=900)
 
     def test_number_of_workers(self):
-        """Ensure the default number of workers matches num cpus."""
+        """Ensure the number of workers matches num cpus limited to 8."""
         def pid():
             i = 1
             while True:
@@ -612,12 +612,30 @@ class ServerTest(test_utils.BaseTestCase):
                 yield i
 
         with mock.patch.object(os, 'fork') as mock_fork:
+            with mock.patch('oslo_concurrency.processutils.get_worker_count',
+                            return_value=4):
+                mock_fork.side_effect = pid
+                server = wsgi.Server()
+                server.configure = mock.Mock()
+                fake_application = "fake-application"
+                server.start(fake_application, None)
+                self.assertEqual(4, len(server.children))
+            with mock.patch('oslo_concurrency.processutils.get_worker_count',
+                            return_value=24):
+                mock_fork.side_effect = pid
+                server = wsgi.Server()
+                server.configure = mock.Mock()
+                fake_application = "fake-application"
+                server.start(fake_application, None)
+                self.assertEqual(8, len(server.children))
             mock_fork.side_effect = pid
             server = wsgi.Server()
             server.configure = mock.Mock()
             fake_application = "fake-application"
             server.start(fake_application, None)
-            self.assertEqual(processutils.get_worker_count(),
+            cpus = processutils.get_worker_count()
+            expected_workers = cpus if cpus < 8 else 8
+            self.assertEqual(expected_workers,
                              len(server.children))