follow comments

doc/fluid/howto/optimization/timeline_cn.md

 # timeline工具简介
 
+## <span id="local">本地使用</span>
+
 1. 在训练的主循环外加上`profiler.start_profiler(...)`和`profiler.stop_profiler(...)`。运行之后，代码会在`/tmp/profile`目录下生成一个profile的记录文件。
 
 	**提示：**
@@ -30,3 +32,22 @@ python Paddle/tools/timeline.py --profile_path=/tmp/profile --timeline_path=time
 1. 结果如下图所示，可以放到来查看timetime的细节信息。
 
 	![chrome timeline](./timeline.jpeg)
+
+## 分布式使用
+一般来说，分布式的训练程序都会有两种程序：pserver和trainer。我们提供了把pserver和trainer的profile日志用timeline来显示的方式。 
+ 
+1. trainer打开方式与[本地使用](#local)部分的第1步相同
+
+1. pserver可以通过加两个环境变量打开profile，例如：
+```
+FLAGS_rpc_server_profile_period=10 FLAGS_rpc_server_profile_path=./tmp/pserver python train.py
+```
+
+3. 把pserver和trainer的profile文件生成一个timeline文件，例如：  
+```
+python /paddle/tools/timeline.py
+    --profile_path trainer0=local_profile_10_pass0_0,trainer1=local_profile_10_pass0_1,pserver0=./pserver_0,pserver1=./pserver_1
+    --timeline_path ./dist.timeline
+```
+
+4. 在chrome中加载dist.timeline文件，方法和[本地使用](#local)第4步相同。

doc/fluid/howto/optimization/timeline_en.md

-# how to use timeline tool to do profile
+# How to use timeline tool to do profile
+
+## <span id="local">Local</span>
 
 1. Add `profiler.start_profiler(...)`和`profiler.stop_profiler(...)` to the main training loop. After run, the code will generate a profile record file `/tmp/profile`. **Warning**: Please do not run too many batches when use profiler to record timeline information, for the profile record will grow with the batch number.
 
@@ -18,7 +20,6 @@
 1. Run `python paddle/tools/timeline.py` to process `/tmp/profile`, it will generate another
 file `/tmp/timeline` by default. You can change the path by cmd parameter, please take a look at
 [timeline.py](https://github.com/PaddlePaddle/Paddle/blob/develop/tools/timeline.py) for details.
-
 ```python
 python Paddle/tools/timeline.py --profile_path=/tmp/profile --timeline_path=timeline
 ```
@@ -31,3 +32,22 @@ python Paddle/tools/timeline.py --profile_path=/tmp/profile --timeline_path=time
 
 
 	![chrome timeline](./timeline.jpeg)
+	
+## Distributed
+This tool can support distributed train programs(pserver and trainer) too.
+
+1. Open traniner profiler just like how to use in [local](#local).
+
+1. Open pserver profiler: add some enviroment variables, eg:
+```
+FLAGS_rpc_server_profile_period=10 FLAGS_rpc_server_profile_path=./tmp/pserver python train.py
+```
+
+1. Merge pservers' and trainers' profiler file, eg:
+```
+python /paddle/tools/timeline.py
+    --profile_path trainer0=local_profile_10_pass0_0,trainer1=local_profile_10_pass0_1,pserver0=./pserver_0,pserver1=./pserver_1
+    --timeline_path ./dist.timeline
+```
+ 
+1. Load `dist.timeline` in chrome://tracing

doc/fluid/howto/performance/profiler.md

@@ -95,3 +95,22 @@ struct RecordEvent {
   }
 };
 ```
+
+### Report sample
+
+```
+Event                                             Calls       Total       Min.        Max.        Ave.        Ratio.      
+thread101::deserial                               1410        392.302     0.032768    14.1058     0.278228    0.00117247  
+thread100::GetRPC                                 11          2951.13     7.60675     1426.75     268.284     0.00882     
+thread100::serial                                 14          75.3212     0.07584     36.2135     5.38009     0.000225112 
+thread100::SendRPC                                14          13.9494     0.003072    3.97517     0.996389    4.16905e-05 
+thread99::GetRPC                                  15          3012.62     2.79062     1426.61     200.841     0.00900378  
+... 
+thread0::matmul_grad                              1480        3674.28     0.375808    181.608     2.48262     0.0109813   
+thread0::matmul                                   1480        3365.82     0.196608    172.256     2.2742      0.0100594   
+thread0::mul_grad                                 3840        3167.39     0.411648    3.33824     0.82484     0.00946633  
+thread0::fetch_barrier                            5           3082.82     354.385     1617.88     616.564     0.00921359  
+thread0::dropout                                  2480        3014.05     0.201728    6.76454     1.21534     0.00900807  
+```
+
+Note: profiler can merge the same operator's time which runs multiple times in the same thread.
\ No newline at end of file