Added TDP and make detection; Fixed --auto mode when no parameter sup…

…plied

auto_detect.py

@@ -3,19 +3,66 @@
 import subprocess
 import re
 import logging
+import math
 
 def get_cpu_info(logger):
 
     data = {
         'freq' : None,
         'threads': None,
         'cores': None,
-        #'tdp': None,
+        'tdp': None,
         'mem': None,
-        #'make': None,
+        'make': None,
         'chips': None
     }
 
+    try:
+        file_path = '/sys/class/powercap/intel-rapl/intel-rapl:0/name'
+        with open(file_path, 'r', encoding='UTF-8') as file:
+            domain_name = file.read().strip()
+            if domain_name != 'package-0':
+                raise RuntimeError(f"Domain /sys/class/powercap/intel-rapl/intel-rapl:0/name was not package-0, but {domain_name}")
+
+        file_path = '/sys/class/powercap/intel-rapl/intel-rapl:0/constraint_0_name'
+        with open(file_path, 'r', encoding='UTF-8') as file:
+            constraint_name = file.read().strip()
+            if constraint_name != 'long_term':
+                raise RuntimeError(f"Constraint /sys/class/powercap/intel-rapl/intel-rapl:0/constraint_0_name was not long_term, but {constraint_name}")
+
+        file_path = '/sys/class/powercap/intel-rapl/intel-rapl:0/constraint_0_max_power_uw'
+        with open(file_path, 'r', encoding='UTF-8') as file:
+            tdp = file.read()
+            data['tdp'] = int(tdp) / 1_000_000
+
+        logger.info('Found TDP: %d W', data['tdp'])
+    #pylint: disable=broad-except
+    except Exception as err:
+        logger.info('Exception: %s', err)
+        logger.info('Could not read RAPL powercapping info from /sys/class/powercap/intel-rapl')
+
+    try:
+        file_paths = {
+            1: '/sys/class/powercap/intel-rapl/intel-rapl:0/name',
+            2: '/sys/class/powercap/intel-rapl/intel-rapl:1/name',
+            3: '/sys/class/powercap/intel-rapl/intel-rapl:2/name',
+            4: '/sys/class/powercap/intel-rapl/intel-rapl:3/name',
+            5: '/sys/class/powercap/intel-rapl/intel-rapl:4/name',
+            6: '/sys/class/powercap/intel-rapl/intel-rapl:5/name',
+        }
+        for chips, file_path in file_paths.items():
+            with open(file_path, 'r', encoding='UTF-8') as file:
+                domain_name = file.read().strip()
+                if domain_name != f"package-{chips-1}":
+                    raise RuntimeError(f"Domain {file_path} was not package-{chips-1}, but {domain_name}")
+                logger.info('Found Sockets: %d', chips)
+                data['chips'] = chips
+    #pylint: disable=broad-except
+    except Exception as err:
+        logger.info('Exception: %s', err)
+        logger.info('Could not find (additional) chips info under file path. Most likely reached final chip. continuing ...')
+
+
     try:
         cpuinfo = subprocess.check_output('lscpu', encoding='UTF-8')
         match = re.search(r'On-line CPU\(s\) list:\s*(0-)?(\d+)', cpuinfo)
@@ -25,14 +72,16 @@ def get_cpu_info(logger):
         else:
             logger.info('Could not find Threads. Setting to None')
 
+        # this will overwrite info we have from RAPL socket discovery, as we
+        # deem lscpu more relieable
         match = re.search(r'Socket\(s\):\s*(\d+)', cpuinfo)
         if match:
             data['chips'] = int(match.group(1))
-            logger.info('Found Sockets: %d', data['chips'])
+            logger.info('Found Sockets: %d (will take precedence if not 0)', data['chips'])
         else:
-            logger.info('Could not find Chips/Sockets. Setting to None')
+            logger.info('Could not find Chips/Sockets via lscpu')
 
-        if data['threads']:
+        if data['threads'] and data['chips']:
             match = re.search(r'Thread\(s\) per core:\s*(\d+)', cpuinfo)
             if match:
                 threads_per_core = int(match.group(1))
@@ -48,6 +97,16 @@ def get_cpu_info(logger):
         else:
             logger.info('Could not find Frequency. Setting to None')
 
+        match = re.search(r'Model name:.*Intel\(R\)', cpuinfo)
+        if match:
+            data['make'] = 'intel'
+            logger.info('Found Make: %s', data['make'])
+
+        match = re.search(r'Model name:.*AMD ', cpuinfo)
+        if match:
+            data['make'] = 'amd'
+            logger.info('Found Make: %s', data['make'])
+
 
         # we currently do not match for architecture, as this info is provided nowhere
 
@@ -79,12 +138,13 @@ def get_cpu_info(logger):
             logger.info('/proc/cpuinfo not accesible on system. Could not check for Base Frequency info. Setting value to None.')
 
 
+
     try:
         meminfo = subprocess.check_output(['cat', '/proc/meminfo'], encoding='UTF-8', stderr=subprocess.DEVNULL)
         match = re.search(r'MemTotal:\s*(\d+) kB', meminfo)
         if match:
-            data['mem'] = round(int(match.group(1)) / 1024 / 1024)
-            logger.info('Found Memory: %d', data['mem'])
+            data['mem'] = math.ceil(int(match.group(1)) / 1024 / 1024)
+            logger.info('Found Memory: %d GB', data['mem'])
         else:
             logger.info('Could not find Memory. Setting to None')
     #pylint: disable=broad-except
@@ -99,4 +159,4 @@ def get_cpu_info(logger):
     logger.addHandler(logging.StreamHandler())
     logger.setLevel(logging.INFO)
 
-    get_cpu_info(logger)
+    print(get_cpu_info(logger))

xgb.py

@@ -34,7 +34,7 @@ def train_model(cpu_chips, Z):
 
     X = X[Z.columns] # only select the supplied columns from the command line
 
-    logger.info('Model will be trained on the following variables: %s', X.columns.values)
+    logger.info('Model will be trained on the following columns and restrictions: \n%s', Z)
 
 #    params = {
 #      'max_depth': 10,
@@ -130,20 +130,24 @@ def interpolate_predictions(predictions):
     args_dict = args.__dict__.copy()
     del args_dict['silent']
     del args_dict['auto']
+    del args_dict['energy']
 
     # did the user supply any of the auto detectable arguments?
-    if len(args_dict) == 0 or args.auto:
-        logger.info('No arguments where supplied, or auto mode was forced. Running auto detect on the sytem')
+    if not any(args_dict.values()) or args.auto:
+        logger.info('No arguments where supplied, or auto mode was forced. Running auto detect on the sytem.')
 
         data = auto_detect.get_cpu_info(logger)
 
         logger.info('The following data was auto detected: %s', data)
 
-        args.cpu_freq = data['freq']
-        args.cpu_threads = data['threads']
-        args.cpu_cores = data['cores']
-        args.ram = data['mem']
-        args.cpu_chips = data['chips']
+        # only overwrite not already supplied values
+        args.cpu_freq = args.cpu_freq or data['freq']
+        args.cpu_threads = args.cpu_threads or data['threads']
+        args.cpu_cores = args.cpu_cores or data['cores']
+        args.tdp = args.tdp or data['tdp']
+        args.ram = args.ram or data['mem']
+        args.cpu_make = args.cpu_make or data['make']
+        args.cpu_chips = args.cpu_chips or data['chips']
 
     # set default. We do this here and not in argparse, so we can check if anything was supplied at all
     if not args.vhost_ratio: