GPU and Node Utilization JSON API (CPU, GPU, Memory)

src/variorum/IBM/Power9.c

@@ -6,6 +6,7 @@
 #include <stdio.h>
 #include <stdlib.h>
 #include <sys/time.h>
+#include <sys/resource.h>
 
 #include <config_architecture.h>
 #include <Power9.h>
@@ -440,6 +441,42 @@ int ibm_cpu_p9_get_node_power_json(char **get_power_obj_str)
     return 0;
 }
 
+int ibm_cpu_p9_get_node_util_json(char **get_power_obj_str)
+{
+    char *val = ("VARIORUM_LOG");
+    if (val != NULL && atoi(val) == 1)
+    {
+        printf("Running %s\n", __FUNCTION__);
+    }
+
+    int ru;
+    char hostname[1024];
+    struct timeval tv;
+    struct rusage rusge;
+    uint64_t ts;
+    long mem;
+    json_t *get_util_obj = json_object();
+
+    gethostname(hostname, 1024);
+    gettimeofday(&tv, NULL);
+    ts = tv.tv_sec * (uint64_t)1000000 + tv.tv_usec;
+    json_object_set_new(get_util_obj, "host", json_string(hostname));
+    json_object_set_new(get_util_obj, "timestamp", json_integer(ts));
+
+    ru = getrusage(RUSAGE_SELF, &rusge);
+    if (ru < 0)
+    {
+        printf("Failed to get utilizations\n");
+        return -1;
+    }
+
+    mem = rusge.ru_maxrss;
+    json_object_set_new(get_util_obj, "memory util", json_integer(mem));
+    *get_power_obj_str = json_dumps(get_util_obj, 0);
+    json_decref(get_util_obj);
+    return 0;
+}
+
 int ibm_cpu_p9_get_node_power_domain_info_json(char **get_domain_obj_str)
 {
     char *val = ("VARIORUM_LOG");

src/variorum/IBM/Power9.h

@@ -22,6 +22,8 @@ int ibm_cpu_p9_monitoring(FILE *output);
 
 int ibm_cpu_p9_get_node_power_json(char **get_power_obj_str);
 
+int ibm_cpu_p9_get_node_util_json(char **get_util_obj_str);
+
 int ibm_cpu_p9_get_node_power_domain_info_json(char **get_domain_obj_str);
 
 #endif

src/variorum/IBM/config_ibm.c

@@ -34,6 +34,7 @@ int set_ibm_func_ptrs(int idx)
         g_platform[idx].variorum_cap_gpu_power_ratio = ibm_cpu_p9_cap_gpu_power_ratio;
         g_platform[idx].variorum_monitoring = ibm_cpu_p9_monitoring;
         g_platform[idx].variorum_get_node_power_json = ibm_cpu_p9_get_node_power_json;
+        g_platform[idx].variorum_get_node_util_json = ibm_cpu_p9_get_node_util_json;
         g_platform[idx].variorum_get_node_power_domain_info_json =
             ibm_cpu_p9_get_node_power_domain_info_json;
     }

src/variorum/IBM/ibm_power_features.h

@@ -129,7 +129,6 @@ struct occ_sensor_counter
     uint8_t  pad[5];
 } __attribute__((__packed__));
 
-
 void print_power_sensors(int chipid,
                          int long_ver,
                          FILE *output,

src/variorum/config_architecture.c

@@ -74,7 +74,6 @@ int variorum_enter(const char *filename, const char *func_name, int line_num)
         printf("Number of registered platforms: %d\n", P_NUM_PLATFORMS);
     }
 
-
     variorum_init_func_ptrs();
 
     //Triggers initialization on first call.  Errors assert.
@@ -200,7 +199,6 @@ int variorum_detect_arch(void)
     return 0;
 }
 
-
 void variorum_get_topology(unsigned *nsockets, unsigned *ncores,
                            unsigned *nthreads, int idx)
 {
@@ -358,6 +356,7 @@ void variorum_init_func_ptrs()
         g_platform[i].variorum_print_gpu_utilization = NULL;
         g_platform[i].variorum_monitoring = NULL;
         g_platform[i].variorum_get_node_power_json = NULL;
+        g_platform[i].variorum_get_node_util_json = NULL;
         g_platform[i].variorum_get_node_power_domain_info_json = NULL;
         g_platform[i].variorum_print_energy = NULL;
     }

src/variorum/config_architecture.h

@@ -240,6 +240,11 @@ struct platform
     /// @return Error code.
     int (*variorum_get_node_power_json)(char **get_power_obj_str);
 
+    /// @brief Function pointer to get JSON object node utilization data.
+    ///
+    /// @return Error code.
+    int (*variorum_get_node_util_json)(char **get_power_obj_str);
+
     /// @brief Function pointer to get JSON object for power domain information.
     ///
     /// @return Error code.

src/variorum/variorum.c

@@ -947,7 +947,6 @@ int variorum_disable_turbo(void)
     return err;
 }
 
-
 // The variorum_get_node_power_json is a node-level API, and cannot be implemented
 // at a per-component (eg CPU, GPU) level. This can only be captured by what we
 // define as the 'primary' platform, e.g. IBM Power9 CPU or Intel and AMD CPUs,
@@ -1011,6 +1010,60 @@ int variorum_get_node_power_json(char **get_power_obj_str)
     return err;
 }
 
+int variorum_get_node_util_json(char **get_util_obj_str)
+{
+    int err = 0;
+    int i;
+    err = variorum_enter(__FILE__, __FUNCTION__, __LINE__);
+    if (err)
+    {
+        return -1;
+    }
+
+    // Obtain the index corresponding to the primary platform.
+    for (i = 0; i < P_NUM_PLATFORMS; i++)
+    {
+#ifdef VARIORUM_WITH_INTEL_CPU
+        i = P_INTEL_CPU_IDX;
+        break;
+#endif
+#ifdef VARIORUM_WITH_IBM_CPU
+        i = P_IBM_CPU_IDX;
+        break;
+#endif
+#ifdef VARIORUM_WITH_AMD_CPU
+        i = P_AMD_CPU_IDX;
+        break;
+#endif
+#ifdef VARIORUM_WITH_ARM_CPU
+        i = P_ARM_CPU_IDX;
+        break;
+#endif
+    }
+
+    if (g_platform[i].variorum_get_node_util_json == NULL)
+    {
+        variorum_error_handler("Feature not yet implemented or is not supported",
+                               VARIORUM_ERROR_FEATURE_NOT_IMPLEMENTED,
+                               getenv("HOSTNAME"), __FILE__,
+                               __FUNCTION__, __LINE__);
+        // For the JSON functions, we return a -1 here, so users don't need
+        // to explicitly check for NULL strings.
+        return -1;
+    }
+
+    err = g_platform[i].variorum_get_node_util_json(get_util_obj_str);
+    if (err)
+    {
+        return -1;
+    }
+    err = variorum_exit(__FILE__, __FUNCTION__, __LINE__);
+    if (err)
+    {
+        return -1;
+    }
+    return err;
+}
 // The variorum_get_node_power_domain_info_json is a node-level API, and cannot
 // be implemented at a per-component (eg CPU, GPU) level. This can only be available
 // on what we define as the 'primary' platform, e.g. IBM Power9 CPU or Intel and AMD CPUs,

src/variorum/variorum.h

@@ -538,6 +538,15 @@ int variorum_disable_turbo(void);
 /// check for NULL strings.
 int variorum_get_node_power_json(char **get_power_obj_str);
 
+///@brief Populate a string in JSON format with total node utilization.
+///
+/// @supparcg
+/// - IBM
+///
+/// @param [out] output String (passed by refrence) that contains node-level utilization information.
+/// returns 0 if successful, otherwise -1.
+int variorum_get_node_util_json(char **get_util_obj_str);
+
 /// @brief Populate a string in JSON format with measurable and controllable
 /// power domains, along with the ranges.
 ///