diff --git a/systrace/convert/convert_mspti_timeline.py b/systrace/convert/convert_mspti_timeline.py
index da2c2b70847dfdaa0e3bdd541fbe741eec90f52c..e93f178a4f09e0a951da03f2b748b2f2e7a4711b 100644
--- a/systrace/convert/convert_mspti_timeline.py
+++ b/systrace/convert/convert_mspti_timeline.py
@@ -6,12 +6,14 @@ FileName：slow_node_detection.py
 Author: c00570162/congdechun
 Create Date: 2025/3/26 11:23
 Notes:
+    0726：完善通信算子连线，1 flow s ts 要小于f ts
 
 """
 import sys
 import os
 import json
 import pandas as pd
+from collections import defaultdict
 
 __all__ = ['convert_mspti_timeline']
 
@@ -30,18 +32,18 @@ OP_COLORS = {
     'HcclBatchSendRecv': "thread_state_runnable"
 }
 
-
 def create_args(row):
     return {
         "id": row["Id"],
         "comm_group": row["comm_group"],
-        "count": row["count"]
+        "count": row["count"],
+        "comm_name": row["Name"]
     }
 
-
 def split_df(df):
     """
     根据 mode 列将 DataFrame 拆分为 host 和 device 两个 DataFrame
+    0为host 1为device
     """
     df_host = df[df['SourceKind'] == 0]
     df_device = df[df['SourceKind'] == 1]
@@ -68,7 +70,6 @@ def process_df(data_df, device_id, id2name_dict: dict):
         else:
             df[['comm_op', 'comm_group', 'data_type', 'count']] = df['Name'].str.replace('comm:', '').str.split(',',
                                                                                                         expand=True)
-    df = df.drop(columns=['Name'])
     try:
         df['cat'] = "hccl"
         df['name'] = df['comm_op']
@@ -88,31 +89,114 @@ def process_df(data_df, device_id, id2name_dict: dict):
     return result
 
 
+def split_events(events: list):
+    sorted_host_events = []
+    sorted_device_events = []
+    host2device_events = []
+    for event in events:
+        if event["tid"] == "Host":
+            sorted_host_events.append(event)
+        else:
+            sorted_device_events.append(event)
+
+    if sorted_host_events and sorted_device_events:
+        if sorted_device_events[0]["ts"] > sorted_host_events[-1]["ts"]:
+            host2device_events = [sorted_host_events[-1], sorted_device_events[0]]
+
+    return sorted_host_events, sorted_device_events, host2device_events
+
+def add_outrank_flow_events(all_events: list, flow_id: int):
+    """
+        对于同组的通信算子，添加flow in和flow out事件
+    """
+    # 按ID和commname分组并收集事件
+    id_to_events = defaultdict(list)
+
+    for event in all_events:
+        if 'args' in event and 'id' in event['args']:
+            event_id = event['args']['id']
+            comm_name = event['args']['comm_name']
+            id_to_events[f"{event_id}_{comm_name}"].append(event)
+
+    # 为每个ID组创建flow事件
+    flow_events = []
+    for event_id, events in id_to_events.items():
+        # 按时间戳排序
+        sorted_events = sorted(events, key=lambda x: x['ts'])
+        # 按照tid区分device和host
+        sorted_host_events, sorted_device_events, host2device_events = split_events(sorted_events)
+        # 创建flow事件链
+        flow_id = creat_flow_events_in_list(flow_events, flow_id, sorted_host_events)
+        flow_id = creat_flow_events_in_list(flow_events, flow_id, sorted_device_events)
+        flow_id = creat_flow_events_in_list(flow_events, flow_id, host2device_events)
+    return flow_events, flow_id
+
+
+def creat_flow_events_in_list(flow_events, flow_id, sorted_events):
+    for i in range(len(sorted_events) - 1):
+        prev = sorted_events[i]
+        next_ev = sorted_events[i + 1]
+
+        # 创建flow out事件 (从当前事件指向下一个事件)
+        flow_out = {
+            'name': 'flow',
+            'cat': 'flow',
+            'ph': 's',
+            'id': flow_id,
+            'pid': prev['pid'],
+            'tid': prev['tid'],
+            # 'ts': prev['ts'] + prev['dur'],
+            'ts': prev['ts'],
+            'bp': 'e'
+        }
+
+        # 创建flow in事件 (下一个事件接收flow)
+        flow_in = {
+            'name': 'flow',
+            'cat': 'flow',
+            'ph': 'f',
+            'id': flow_id,
+            'pid': next_ev['pid'],
+            'tid': next_ev['tid'],
+            'ts': next_ev['ts'],
+            'bp': 'e'
+        }
+
+        flow_events.extend([flow_out, flow_in])
+        flow_id += 1
+    return flow_id
+
+
 def process_files(root_path, debug: bool = False):
     """
     处理指定路径下的所有 CSV 文件
     """
     csv_files = [file for file in os.listdir(root_path) if file.endswith("csv") and "device" not in file]
     all_ranks = []
+    flow_ids = 1
     for csv_file in csv_files:
         if "op_launch" in csv_file:
             continue
-        print(f"start file: {csv_file}")
+        print(f"[INFO] Start file: {csv_file}")
         csv_file_path = os.path.join(root_path, csv_file)
         df = pd.read_csv(csv_file_path)
         if debug:
             df = df.head(12)
 
         id2name_dict = df[df['Name'].notna()].set_index('Id')['Name'].to_dict()
-        # df['name'] = df.groupby('id')['name'].transform(lambda x: x.ffill().bfill())
         df_host, df_device = split_df(df)
         device_id = int(csv_file.split(".")[-2])
         host_result = process_df(df_host, device_id, id2name_dict)
+        device_result = process_df(df_device, device_id, id2name_dict)
+
         if host_result:
             all_ranks.extend(host_result)
-        device_result = process_df(df_device, device_id, id2name_dict)
         if device_result:
             all_ranks.extend(device_result)
+
+    print(f"[INFO] Generate flow for comm op.")
+    outrank_flow_events, flow_id = add_outrank_flow_events(all_ranks, flow_ids)
+    all_ranks.extend(outrank_flow_events)
     return all_ranks