fixed performance issue (moved STRECH case to low priority cases)

Signed-off-by: imarom <imarom@cisco.com>

5 files changed, 72 insertions, 36 deletions

diff --git a/linux_dpdk/d b/linux_dpdk/d
index 8e511bbf..8e511bbf 100644..100755
--- a/linux_dpdk/d
+++ b/linux_dpdk/d
diff --git a/scripts/automation/regression/setups/trex09/benchmark.yaml b/scripts/automation/regression/setups/trex09/benchmark.yaml
index ed51a46c..f157d8f2 100644
--- a/scripts/automation/regression/setups/trex09/benchmark.yaml
+++ b/scripts/automation/regression/setups/trex09/benchmark.yaml
@@ -209,24 +209,24 @@ test_performance_syn_attack_single_cpu:
 
 test_performance_vm_multi_cpus:
     cfg:
-        core_count             : 3
+        core_count             : 2
         mult                   : "90%"
         mpps_per_core_golden   :
-                                  min: 15.9
-                                  max: 16.5
+                                  min: 15.7
+                                  max: 16.3
 
 
 test_performance_vm_multi_cpus_cached:
     cfg:
-        core_count             : 3
+        core_count             : 2
         mult                   : "90%"
         mpps_per_core_golden   :
-                                  min: 29.6
-                                  max: 30.5
+                                  min: 28.8
+                                  max: 29.5
 
 test_performance_syn_attack_multi_cpus:
     cfg:
-        core_count             : 3
+        core_count             : 2
         mult                   : "90%"
         mpps_per_core_golden   :
                                   min: 13.0
diff --git a/scripts/automation/regression/stateless_tests/stl_performance_test.py b/scripts/automation/regression/stateless_tests/stl_performance_test.py
index a77c5c27..27c61cc0 100644
--- a/scripts/automation/regression/stateless_tests/stl_performance_test.py
+++ b/scripts/automation/regression/stateless_tests/stl_performance_test.py
@@ -180,7 +180,7 @@ class STLPerformance_Test(CStlGeneral_Test):
         setup_cfg = self.get_benchmark_param('cfg')
         scenario_cfg = {}
 
-        scenario_cfg['name']        = "VM - 64 bytes, two CPUs"
+        scenario_cfg['name']        = "VM - 64 bytes, multi CPUs"
         scenario_cfg['streams']     = self.build_perf_profile_vm(64)
 
         scenario_cfg['core_count']           = setup_cfg['core_count']
@@ -191,12 +191,12 @@ class STLPerformance_Test(CStlGeneral_Test):
 
 
 
-    # two CPUs, VM, cached, 64 bytes
+    # multi CPUs, VM, cached, 64 bytes
     def test_performance_vm_multi_cpus_cached (self):
         setup_cfg = self.get_benchmark_param('cfg')
         scenario_cfg = {}
 
-        scenario_cfg['name']        = "VM - 64 bytes, single CPU, cache size 1024"
+        scenario_cfg['name']        = "VM - 64 bytes, multi CPU, cache size 1024"
         scenario_cfg['streams']     = self.build_perf_profile_vm(64, cache_size = 1024)
 
 
@@ -207,12 +207,12 @@ class STLPerformance_Test(CStlGeneral_Test):
         self.execute_single_scenario(scenario_cfg)
 
 
-    # two CPUs, syn attack, 64 bytes
+    # multi CPUs, syn attack, 64 bytes
     def test_performance_syn_attack_multi_cpus (self):
         setup_cfg = self.get_benchmark_param('cfg')
         scenario_cfg = {}
 
-        scenario_cfg['name']        = "syn attack - 64 bytes, two CPUs"
+        scenario_cfg['name']        = "syn attack - 64 bytes, multi CPUs"
         scenario_cfg['streams']     = self.build_perf_profile_syn_attack(64)
 
         scenario_cfg['core_count']           = setup_cfg['core_count']
@@ -254,6 +254,10 @@ class STLPerformance_Test(CStlGeneral_Test):
         self.c.add_streams(ports = [0], streams = scenario_cfg['streams'])
 
         # use one core
+        cores_per_port = self.c.system_info.get('dp_core_count_per_port', 0)
+        if cores_per_port < scenario_cfg['core_count']:
+            assert 0, "test configuration requires {0} cores but only {1} per port are available".format(scenario_cfg['core_count'], cores_per_port)
+
         core_mask = (2 ** scenario_cfg['core_count']) - 1
         self.c.start(ports = [0], mult = scenario_cfg['mult'], core_mask = [core_mask])
 
diff --git a/src/bp_sim.cpp b/src/bp_sim.cpp
index c4cb0927..db273b18 100755
--- a/src/bp_sim.cpp
+++ b/src/bp_sim.cpp
@@ -3787,22 +3787,6 @@ inline int CNodeGenerator::flush_file_realtime(dsec_t max_time,
             }
             break;
 
-         /* a case called when a time strech happens */
-         case scSTRECH:
-             {
-                 dsec_t dt = cur_time - n_time;
-                 handle_time_strech(cur_time, dt, offset, thread);
-
-                 /* re-read the top of the queue - it might have changed with messaging */
-                 node = m_p_queue.top();
-                 n_time = node->m_time + offset;
-
-                 /* go back to INIT */
-                 state = scINIT;
-
-             }
-             break;
-
          case scWORK:
             {
                 int node_count = 0;
@@ -3831,14 +3815,42 @@ inline int CNodeGenerator::flush_file_realtime(dsec_t max_time,
                 do_sleep(cur_time,thread,n_time); // estimate  loop
                 state=scWORK;
                 break;
+
+
          default:
-             assert(0);
+             handle_slow_operations(state, node, cur_time, n_time, offset, thread);
+             break;
         } /* switch */
+
     }/* while*/
 
     return (teardown(thread,always,old_offset,offset));
 }
 
+
+FORCE_NO_INLINE void CNodeGenerator::handle_slow_operations(sch_state_t &state,
+                                                            CGenNode * &node,
+                                                            dsec_t &cur_time,
+                                                            dsec_t &n_time,
+                                                            dsec_t &offset,
+                                                            CFlowGenListPerThread *thread) {
+    switch (state) {
+    case scSTRECH:
+        {
+            handle_time_strech(node, cur_time, n_time, offset, thread);
+
+            /* go back to work */
+            state = scWORK;
+
+        }
+        break;
+
+    default:
+        assert(0);
+    }
+
+}
+
 /**
  * when time is streched - the flow_sync node 
  * might be postpond too much 
@@ -3849,18 +3861,26 @@ inline int CNodeGenerator::flush_file_realtime(dsec_t max_time,
  * @author imarom (7/31/2016)
  * 
  */
-FORCE_NO_INLINE void CNodeGenerator::handle_time_strech(dsec_t cur_time,
-                                                        dsec_t dt,
-                                                        dsec_t &offset,
-                                                        CFlowGenListPerThread *thread) {
+void CNodeGenerator::handle_time_strech(CGenNode * &node,
+                                        dsec_t &cur_time,
+                                        dsec_t &n_time,
+                                        dsec_t &offset,
+                                        CFlowGenListPerThread *thread) {
+
+
+    /* fix the time offset */
+    dsec_t dt = cur_time - n_time;
+    offset += dt;
 
     /* check if flow sync message was delayed too much */
     if ( (cur_time - m_last_sync_time_sec) > SYNC_TIME_OUT ) {
         handle_maintenance(thread);
+
+        /* re-read the top of the queue - it might have changed with messaging */
+        node = m_p_queue.top();
+        n_time = node->m_time + offset;
     }
 
-    /* fix the time offset */
-    offset += dt;
 }
 
 int CNodeGenerator::flush_file_sim(dsec_t max_time,
diff --git a/src/bp_sim.h b/src/bp_sim.h
index 0af27b95..f7a1e73c 100755
--- a/src/bp_sim.h
+++ b/src/bp_sim.h
@@ -2142,7 +2142,19 @@ private:
                             CFlowGenListPerThread * thread,
                             double &old_offset);
 
-        FORCE_NO_INLINE void handle_time_strech(dsec_t cur_time, dsec_t dt, dsec_t &offset,  CFlowGenListPerThread * thread);
+        FORCE_NO_INLINE void handle_slow_operations(sch_state_t &state,
+                                                    CGenNode * &node,
+                                                    dsec_t &cur_time,
+                                                    dsec_t &n_time,
+                                                    dsec_t &offset,
+                                                    CFlowGenListPerThread *thread);
+
+        void handle_time_strech(CGenNode * &node,
+                                dsec_t &cur_time,
+                                dsec_t &n_time,
+                                dsec_t &offset,
+                                CFlowGenListPerThread *thread);
+
 
 private:        
     void handle_command(CGenNode *node, CFlowGenListPerThread *thread, bool &exit_scheduler);