Commit 010e0cca authored by hardy's avatar hardy

bring ulsch and dlsch stats for gnb

parent 95108806
"""
To create graphs and pickle from runtime statistics in L1,MAC,RRC,PDCP files
"""
import subprocess
import time
import shlex
import re
import sys
import matplotlib.pyplot as plt
import pickle
import matplotlib.pyplot as plt
import numpy as np
import os
def collect(d, node_type):
if node_type=='enb':
cmd='cat L1_stats.log MAC_stats.log PDCP_stats.log RRC_stats.log'
else: #'gnb'
cmd='cat nrL1_stats.log nrMAC_stats.log nrPDCP_stats.log nrRRC_stats.log'
process=subprocess.Popen(shlex.split(cmd), stdout=subprocess.PIPE)
output = process.stdout.readlines()
for l in output:
tmp=l.decode("utf-8")
result=re.match(rf'^.*\bPHR\b ([0-9]+).+\bbler\b ([0-9]+\.[0-9]+).+\bmcsoff\b ([0-9]+).+\bmcs\b ([0-9]+)',tmp)
if result is not None:
d['PHR'].append(int(result.group(1)))
d['bler'].append(float(result.group(2)))
d['mcsoff'].append(int(result.group(3)))
d['mcs'].append(int(result.group(4)))
def graph(d, node_type):
figure, axis = plt.subplots(4, 1,figsize=(10, 10))
major_ticks = np.arange(0, len(d['PHR'])+1, 1)
axis[0].set_xticks(major_ticks)
axis[0].set_xticklabels([])
axis[0].plot(d['PHR'],marker='o')
axis[0].set_xlabel('time')
axis[0].set_ylabel('PHR')
axis[0].set_title("PHR")
major_ticks = np.arange(0, len(d['bler'])+1, 1)
axis[1].set_xticks(major_ticks)
axis[1].set_xticklabels([])
axis[1].plot(d['bler'],marker='o')
axis[1].set_xlabel('time')
axis[1].set_ylabel('bler')
axis[1].set_title("bler")
major_ticks = np.arange(0, len(d['mcsoff'])+1, 1)
axis[2].set_xticks(major_ticks)
axis[2].set_xticklabels([])
axis[2].plot(d['mcsoff'],marker='o')
axis[2].set_xlabel('time')
axis[2].set_ylabel('mcsoff')
axis[2].set_title("mcsoff")
major_ticks = np.arange(0, len(d['mcs'])+1, 1)
axis[3].set_xticks(major_ticks)
axis[3].set_xticklabels([])
axis[3].plot(d['mcs'],marker='o')
axis[3].set_xlabel('time')
axis[3].set_ylabel('mcs')
axis[3].set_title("mcs")
plt.tight_layout()
# Combine all the operations and display
plt.savefig(node_type+'_stats_monitor.png')
plt.show()
if __name__ == "__main__":
import yaml
class StatMonitor():
def __init__(self,):
with open('stats_monitor_conf.yaml','r') as file:
self.d = yaml.load(file)
for node in self.d:
for metric in self.d[node]:
self.d[node][metric]=[]
def process_gnb (self,node_type,output):
for line in output:
tmp=line.decode("utf-8")
result=re.match(r'^.*\bdlsch_rounds\b ([0-9]+)\/([0-9]+).*\bdlsch_errors\b ([0-9]+)',tmp)
if result is not None:
self.d[node_type]['dlsch_err'].append(int(result.group(3)))
percentage=float(result.group(2))/float(result.group(1))
self.d[node_type]['dlsch_err_perc_round_1'].append(percentage)
result=re.match(r'^.*\bulsch_rounds\b ([0-9]+)\/([0-9]+).*\bulsch_errors\b ([0-9]+)',tmp)
if result is not None:
self.d[node_type]['ulsch_err'].append(int(result.group(3)))
percentage=float(result.group(2))/float(result.group(1))
self.d[node_type]['ulsch_err_perc_round_1'].append(percentage)
def process_enb (self,node_type,output):
for line in output:
tmp=line.decode("utf-8")
result=re.match(r'^.*\bPHR\b ([0-9]+).+\bbler\b ([0-9]+\.[0-9]+).+\bmcsoff\b ([0-9]+).+\bmcs\b ([0-9]+)',tmp)
if result is not None:
self.d[node_type]['PHR'].append(int(result.group(1)))
self.d[node_type]['bler'].append(float(result.group(2)))
self.d[node_type]['mcsoff'].append(int(result.group(3)))
self.d[node_type]['mcs'].append(int(result.group(4)))
def collect(self,node_type):
if node_type=='enb':
cmd='cat L1_stats.log MAC_stats.log PDCP_stats.log RRC_stats.log'
else: #'gnb'
cmd='cat nrL1_stats.log nrMAC_stats.log nrPDCP_stats.log nrRRC_stats.log'
process=subprocess.Popen(shlex.split(cmd), stdout=subprocess.PIPE)
output = process.stdout.readlines()
if node_type=='enb':
self.process_enb(node_type,output)
else: #'gnb'
self.process_gnb(node_type,output)
def graph(self,node_type):
col = 1
figure, axis = plt.subplots(len(self.d[node_type]), col ,figsize=(10, 10))
i=0
for metric in self.d[node_type]:
major_ticks = np.arange(0, len(self.d[node_type][metric])+1, 1)
axis[i].set_xticks(major_ticks)
axis[i].set_xticklabels([])
axis[i].plot(self.d[node_type][metric],marker='o')
axis[i].set_xlabel('time')
axis[i].set_ylabel(metric)
axis[i].set_title(metric)
i+=1
plt.tight_layout()
# Combine all the operations and display
plt.savefig(node_type+'_stats_monitor.png')
plt.show()
node_type = sys.argv[1]#enb or gnb
d={}
d['PHR']=[]
d['bler']=[]
d['mcsoff']=[]
d['mcs']=[]
if __name__ == "__main__":
node = sys.argv[1]#enb or gnb
mon=StatMonitor()
cmd='ps aux | grep modem | grep -v grep'
process=subprocess.Popen(cmd, shell=True, stdout=subprocess.PIPE)
#collecting stats when modem process is stopped
CMD='ps aux | grep mode | grep -v grep'
process=subprocess.Popen(CMD, shell=True, stdout=subprocess.PIPE)
output = process.stdout.readlines()
while len(output)!=0 :
collect(d, node_type)
process=subprocess.Popen(cmd, shell=True, stdout=subprocess.PIPE)
mon.collect(node)
process=subprocess.Popen(CMD, shell=True, stdout=subprocess.PIPE)
output = process.stdout.readlines()
time.sleep(1)
print('process stopped')
with open(node_type+'_stats_monitor.pickle', 'wb') as handle:
pickle.dump(d, handle, protocol=pickle.HIGHEST_PROTOCOL)
graph(d, node_type)
print('Process stopped')
with open(node+'_stats_monitor.pickle', 'wb') as handle:
pickle.dump(mon.d, handle, protocol=pickle.HIGHEST_PROTOCOL)
mon.graph(node)
import subprocess
import time
import shlex
import re
import sys
import matplotlib.pyplot as plt
import pickle
import numpy as np
import os
def collect(d, node_type):
if node_type=='enb':
cmd='cat L1_stats.log MAC_stats.log PDCP_stats.log RRC_stats.log'
else: #'gnb'
cmd='cat nrL1_stats.log nrMAC_stats.log nrPDCP_stats.log nrRRC_stats.log'
process=subprocess.Popen(shlex.split(cmd), stdout=subprocess.PIPE)
output = process.stdout.readlines()
for l in output:
tmp=l.decode("utf-8")
result=re.match(rf'^.*\bPHR\b ([0-9]+).+\bbler\b ([0-9]+\.[0-9]+).+\bmcsoff\b ([0-9]+).+\bmcs\b ([0-9]+)',tmp)
if result is not None:
d['PHR'].append(int(result.group(1)))
d['bler'].append(float(result.group(2)))
d['mcsoff'].append(int(result.group(3)))
d['mcs'].append(int(result.group(4)))
def graph(d, node_type):
figure, axis = plt.subplots(4, 1,figsize=(10, 10))
major_ticks = np.arange(0, len(d['PHR'])+1, 1)
axis[0].set_xticks(major_ticks)
axis[0].set_xticklabels([])
axis[0].plot(d['PHR'],marker='o')
axis[0].set_xlabel('time')
axis[0].set_ylabel('PHR')
axis[0].set_title("PHR")
major_ticks = np.arange(0, len(d['bler'])+1, 1)
axis[1].set_xticks(major_ticks)
axis[1].set_xticklabels([])
axis[1].plot(d['bler'],marker='o')
axis[1].set_xlabel('time')
axis[1].set_ylabel('bler')
axis[1].set_title("bler")
major_ticks = np.arange(0, len(d['mcsoff'])+1, 1)
axis[2].set_xticks(major_ticks)
axis[2].set_xticklabels([])
axis[2].plot(d['mcsoff'],marker='o')
axis[2].set_xlabel('time')
axis[2].set_ylabel('mcsoff')
axis[2].set_title("mcsoff")
major_ticks = np.arange(0, len(d['mcs'])+1, 1)
axis[3].set_xticks(major_ticks)
axis[3].set_xticklabels([])
axis[3].plot(d['mcs'],marker='o')
axis[3].set_xlabel('time')
axis[3].set_ylabel('mcs')
axis[3].set_title("mcs")
plt.tight_layout()
# Combine all the operations and display
plt.savefig(node_type+'_stats_monitor.png')
plt.show()
if __name__ == "__main__":
node_type = sys.argv[1]#enb or gnb
d={}
d['PHR']=[]
d['bler']=[]
d['mcsoff']=[]
d['mcs']=[]
cmd='ps aux | grep modem | grep -v grep'
process=subprocess.Popen(cmd, shell=True, stdout=subprocess.PIPE)
output = process.stdout.readlines()
while len(output)!=0 :
collect(d, node_type)
process=subprocess.Popen(cmd, shell=True, stdout=subprocess.PIPE)
output = process.stdout.readlines()
time.sleep(1)
print('process stopped')
with open(node_type+'_stats_monitor.pickle', 'wb') as handle:
pickle.dump(d, handle, protocol=pickle.HIGHEST_PROTOCOL)
graph(d, node_type)
......@@ -5,7 +5,7 @@ enb :
mcs:
gnb :
PHR:
bler:
mcsoff:
mcs:
\ No newline at end of file
dlsch_err:
dlsch_err_perc_round_1:
ulsch_err:
ulsch_err_perc_round_1:
\ No newline at end of file
......@@ -22,19 +22,22 @@ class StatMonitor():
self.d[node][metric]=[]
def process_enb (self,node_type,output):
def process_gnb (self,node_type,output):
for line in output:
tmp=line.decode("utf-8")
result=re.match(r'^.*\bPHR\b ([0-9]+).+\bbler\b ([0-9]+\.[0-9]+).+\bmcsoff\b ([0-9]+).+\bmcs\b ([0-9]+)',tmp)
result=re.match(r'^.*\bdlsch_rounds\b ([0-9]+)\/([0-9]+).*\bdlsch_errors\b ([0-9]+)',tmp)
if result is not None:
self.d[node_type]['PHR'].append(int(result.group(1)))
self.d[node_type]['bler'].append(float(result.group(2)))
self.d[node_type]['mcsoff'].append(int(result.group(3)))
self.d[node_type]['mcs'].append(int(result.group(4)))
self.d[node_type]['dlsch_err'].append(int(result.group(3)))
percentage=float(result.group(2))/float(result.group(1))
self.d[node_type]['dlsch_err_perc_round_1'].append(percentage)
result=re.match(r'^.*\bulsch_rounds\b ([0-9]+)\/([0-9]+).*\bulsch_errors\b ([0-9]+)',tmp)
if result is not None:
self.d[node_type]['ulsch_err'].append(int(result.group(3)))
percentage=float(result.group(2))/float(result.group(1))
self.d[node_type]['ulsch_err_perc_round_1'].append(percentage)
def process_gnb (self,node_type,output):
def process_enb (self,node_type,output):
for line in output:
tmp=line.decode("utf-8")
result=re.match(r'^.*\bPHR\b ([0-9]+).+\bbler\b ([0-9]+\.[0-9]+).+\bmcsoff\b ([0-9]+).+\bmcs\b ([0-9]+)',tmp)
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment