diff --git a/clickbench/README.md b/clickbench/README.md
new file mode 100644
index 0000000000000..786652920a52f
--- /dev/null
+++ b/clickbench/README.md
@@ -0,0 +1,52 @@
+
+
+# Clickbench Benchmark
+
+## Introduction
+
+ is a new benchmark set provided by ClickHouse. The current result for DataFusion is from Azure's `v16s_v2` VM
+
+
+## The way to reproduce
+
+1. `git clone git@github.com:ClickHouse/ClickBench.git`
+2. cd ClickBench
+3. `bash run.sh`
+
+Note that this will output the needed result like the `result` field in . We need to manually provide `system`, `date`, ... for now, and save a new json into `results` folder. After merged by ClickBench and new html files regenerated, it'll be shown in .
+
+There're 43 queries to compute, each will be executed 3 times. each rows contains the execution time for these 3 quries. The overall output will be a 43 by 3 matrix.
+
+## To generate human readable results
+
+your can do
+
+```bash
+bash run2.sh
+```
+
+Each query will be only exeute only once, for each query it'll print the SQL expression first then output the result. (Note that this index begins with 1, the ClickBench begins with 0)
+
+## Known Issues
+
+
+1. importing parquet by `datafusion-cli` make column name case-sensitive (i.e. i need to query thoses column by double quoted string). The original `queries.sql`'s column name contains no double quotes, I manually added the double quotes for all the column name for now
+
+2. since our parquet importer doesn't support schema, I use some functions to convert the data type in queries (i.e. "EventDate" becomes "EventDate::INT::DATE"). Note that other plaform support this, so it could be added while creating table i.e.
\ No newline at end of file