diff --git a/perfkitbenchmarker/linux_benchmarks/netperf_benchmark.py b/perfkitbenchmarker/linux_benchmarks/netperf_benchmark.py
index 08a0c119ae..389858d4ca 100644
--- a/perfkitbenchmarker/linux_benchmarks/netperf_benchmark.py
+++ b/perfkitbenchmarker/linux_benchmarks/netperf_benchmark.py
@@ -55,9 +55,8 @@
 flags.DEFINE_integer(
     'netperf_warmup_test_length',
     60,
-    'The duration of the warmup run in seconds. If --netperf_benchmarks'
-    ' contains TCP_RR, a warmup run of this duration will be performed before'
-    ' the benchmark runs. If 0, no warmup is done.',
+    'The duration of the warmup run in seconds. If greater than 0, a warmup '
+    'run of this duration will be performed before each benchmark run.',
     lower_bound=0,
 )
 flags.DEFINE_bool(
@@ -241,13 +240,6 @@ def Prepare(benchmark_spec):
       ],
       2,
   )
-  if (
-      TCP_RR in FLAGS.netperf_benchmarks
-      and FLAGS.netperf_warmup_test_length > 0
-  ):
-    # If TCP_RR is requested, do a warmup run as 1st run performance could be
-    # low due to initial flow setup issue.
-    RunClientServerVMs(client_vm, server_vm, FLAGS.netperf_warmup_test_length)
 
 
 def PrepareClientVM(client_vm):
@@ -526,12 +518,8 @@ def RunNetperf(
       'netperf_test_length': test_length,
       'sending_thread_count': num_streams,
       'max_iter': FLAGS.netperf_max_iter or 1,
+      'netperf_warmup_test_length': FLAGS.netperf_warmup_test_length,
   }
-  if (
-      TCP_RR in FLAGS.netperf_benchmarks
-      and FLAGS.netperf_warmup_test_length > 0
-  ):
-    metadata['netperf_warmup_test_length'] = FLAGS.netperf_warmup_test_length
 
   remote_cmd_list = []
   assert server_ips, 'Server VM does not have an IP to use for netperf.'
@@ -803,6 +791,20 @@ def RunClientServerVMs(client_vm, server_vm, test_length=None):
 
     for netperf_benchmark in FLAGS.netperf_benchmarks:
       if vm_util.ShouldRunOnExternalIpAddress():
+        if FLAGS.netperf_warmup_test_length > 0:
+          logging.info(
+              'Running warmup for %s for %d seconds',
+              netperf_benchmark,
+              FLAGS.netperf_warmup_test_length,
+          )
+          RunNetperf(
+              client_vm,
+              netperf_benchmark,
+              server_vm.GetExternalIPs(),
+              num_streams,
+              [client_vm.GetInternalIPs()[0]],
+              FLAGS.netperf_warmup_test_length,
+          )
         external_ip_results = RunNetperf(
             client_vm,
             netperf_benchmark,
@@ -821,6 +823,20 @@ def RunClientServerVMs(client_vm, server_vm, test_length=None):
         results.extend(external_ip_results)
 
       if vm_util.ShouldRunOnInternalIpAddress(client_vm, server_vm):
+        if FLAGS.netperf_warmup_test_length > 0:
+          logging.info(
+              'Running warmup for %s for %d seconds',
+              netperf_benchmark,
+              FLAGS.netperf_warmup_test_length,
+          )
+          RunNetperf(
+              client_vm,
+              netperf_benchmark,
+              server_vm.GetInternalIPs(),
+              num_streams,
+              client_vm.GetInternalIPs(),
+              FLAGS.netperf_warmup_test_length,
+          )
         internal_ip_results = RunNetperf(
             client_vm,
             netperf_benchmark,
diff --git a/tests/linux_benchmarks/netperf_benchmark_test.py b/tests/linux_benchmarks/netperf_benchmark_test.py
index 45be03d268..be803cc998 100644
--- a/tests/linux_benchmarks/netperf_benchmark_test.py
+++ b/tests/linux_benchmarks/netperf_benchmark_test.py
@@ -54,6 +54,7 @@ def setUp(self):
     self.should_run_internal = p.start()
     self.addCleanup(p.stop)
     FLAGS.netperf_enable_histograms = False
+    FLAGS.netperf_warmup_test_length = 0
 
   def _ConfigureIpTypes(self, run_external=True, run_internal=True):
     self.should_run_external.return_value = run_external
@@ -244,6 +245,25 @@ def testMultiStreams(self):
         [i[:3] for i in result],
     )
 
+  @flagsaver.flagsaver(netperf_benchmarks=['TCP_STREAM'])
+  @flagsaver.flagsaver(netperf_num_streams=[1])
+  @flagsaver.flagsaver(netperf_warmup_test_length=5)
+  def testWarmup(self):
+    self._ConfigureIpTypes(run_external=False, run_internal=True)
+    vm_spec = mock.MagicMock(spec=benchmark_spec.BenchmarkSpec)
+    vm_spec.vms = [mock.MagicMock(), mock.MagicMock()]
+    vm_spec.vms[0].RobustRemoteCommand.side_effect = [
+        (self.expected_stdout[0], ''),  # Warmup
+        (self.expected_stdout[0], ''),  # Actual
+    ]
+    vm_spec.vms[1].GetInternalIPs.return_value = ['test_ip']
+    vm_spec.vms[0].GetInternalIPs.return_value = ['test_ip']
+    unused_run_result = netperf_benchmark.Run(vm_spec)
+    self.assertEqual(vm_spec.vms[0].RobustRemoteCommand.call_count, 2)
+    call_args_list = vm_spec.vms[0].RobustRemoteCommand.call_args_list
+    self.assertIn('-l 5', call_args_list[0][0][0])
+    self.assertIn('-l 60', call_args_list[1][0][0])
+
 
 if __name__ == '__main__':
   unittest.main()