diff --git a/pyproject.toml b/pyproject.toml
index 57f0361..ce726a0 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "mtnlog"
-version = "1.0.3"
+version = "1.1.0"
 authors = [
     { name = "Wongkraiwich Chuenchomphu", email = "wongkraiwich@inedible.dev" },
 ]
diff --git a/src/mtnlog/__init__.py b/src/mtnlog/__init__.py
index ad8cc6e..a6b93f6 100644
--- a/src/mtnlog/__init__.py
+++ b/src/mtnlog/__init__.py
@@ -1,7 +1,7 @@
 from .json import *
 from .performance import *
 
-__version__ = '1.0.2'
+__version__ = '1.1.0'
 
 __doc__ = """Performance logger for tracking resource usage."""
 
diff --git a/src/mtnlog/performance.py b/src/mtnlog/performance.py
index b360fcc..f33ed48 100644
--- a/src/mtnlog/performance.py
+++ b/src/mtnlog/performance.py
@@ -1,24 +1,27 @@
-"""Performance logger class for logging performance metrics."""
-
 import os
 import psutil
-
 import pandas as pd
+import csv
+import logging
 from nvitop import ResourceMetricCollector, Device
+import time
+
+logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')
 
 
 class PerformanceLogger:
     """Performance logger class."""
 
     def __init__(self, log_dir, log_node):
-
         os.makedirs(log_dir, exist_ok=True)
-
         self.log_dir = log_dir
         self.log_node = log_node
         self.df = pd.DataFrame()
         self.tag = None
         self.filepath = None
+        self.writer = None
+        self.file = None
+        self.first_collect = True
         self.collector = ResourceMetricCollector(Device.cuda.all()).daemonize(
             on_collect=self.on_collect,
             interval=1.0,
@@ -27,29 +30,36 @@ def __init__(self, log_dir, log_node):
         self.start_time = None
 
     def new_res(self):
-        """Returns the directory."""
-
-        os.makedirs(f"{self.log_dir}/{self.tag}", exist_ok=True)
-
-        self.filepath = f"{self.log_dir}/{self.tag}/node-{self.log_node}.csv"
+        """Creates a new resource file directory and sets the file path."""
+        os.makedirs(self.log_dir, exist_ok=True)
+        self.filepath = f"{self.log_dir}/node-{self.log_node}.csv"
+
+        # Open the file and set up the CSV writer
+        file_exists = os.path.exists(self.filepath)
+        self.file = open(self.filepath, 'a', newline='', encoding='utf-8')
+        if file_exists:
+            self.writer = csv.DictWriter(self.file, fieldnames=self.df.columns)
+        else:
+            self.writer = None
+        logging.info("New resource file created: %s", self.filepath)
 
     def change_tag(self, tag):
-        """Changes the tag."""
+        """Changes the tag and restarts the collector if necessary."""
         if self.filepath is not None:
             self.stop()
         self.tag = tag
         self.new_res()
 
     def stop(self):
-        """Stops the collector."""
-        if not self.df.empty:
-            self.df.to_csv(self.filepath, index=False)
+        """Stops the collector and saves the collected data to a CSV file."""
+        if self.file:
+            self.file.close()
         self.df = pd.DataFrame()
 
     def get_cpu_usage_per_core(self):
         """Returns the CPU usage per core."""
         cpu_percent = psutil.cpu_percent(interval=0.1, percpu=True)
-        return {f"cpu_core_{i+1}": percent for i, percent in enumerate(cpu_percent[:self.cpu_count])}
+        return {f"cpu_core_{i+1} (%)": percent for i, percent in enumerate(cpu_percent[:self.cpu_count])}
 
     def clean_column_name(self, col):
         """Cleans the column name."""
@@ -58,24 +68,50 @@ def clean_column_name(self, col):
         return col
 
     def on_collect(self, metrics):
-        """Collects metrics."""
-
+        """Collects and processes metrics."""
         metrics['tag'] = self.tag
-
         cpu_metrics = self.get_cpu_usage_per_core()
         metrics.update(cpu_metrics)
-
         df_metrics = pd.DataFrame.from_records([metrics])
-
         df_metrics.columns = [self.clean_column_name(col) for col in df_metrics.columns]
 
         if self.df.empty:
             self.df = df_metrics
         else:
-            for col in df_metrics.columns:
-                if col not in self.df.columns:
-                    self.df[col] = None
-
+            self.df = self.df.reindex(columns=list(self.df.columns) +
+                                      [col for col in df_metrics.columns if col not in self.df.columns])
+            df_metrics = df_metrics.reindex(columns=self.df.columns, fill_value=None)
             self.df = pd.concat([self.df, df_metrics], ignore_index=True)
 
+        self._write_to_file(df_metrics)
+
         return True
+
+    def _write_to_file(self, row):
+        """Writes a row of data to the file."""
+        row_dict = row.to_dict(orient='records')[0]  # Convert DataFrame row to dictionary
+        if self.first_collect:
+            with open(self.filepath, 'w', newline='', encoding='utf-8') as f:
+                writer = csv.DictWriter(f, fieldnames=row.columns)
+                writer.writeheader()
+                writer.writerow(row_dict)
+            self.first_collect = False
+            logging.info("First collection completed with header: %s", self.filepath)
+        else:
+            with open(self.filepath, 'a', newline='', encoding='utf-8') as f:
+                writer = csv.DictWriter(f, fieldnames=self.df.columns)
+                writer.writerow(row_dict)
+                f.flush()
+            # logging.info(f"Appended row to file: {self.filepath}")
+
+    def start(self):
+        """Starts the performance logger."""
+        self.start_time = time.time()
+
+    def cleanup(self):
+        """Stops the collector and ensures all data is saved."""
+        self.stop()
+
+    def __del__(self):
+        """Destructor to ensure cleanup is called."""
+        self.cleanup()