diff --git a/quicksync/Readme.md b/quicksync/Readme.md
new file mode 100644
index 0000000..f6a23ce
--- /dev/null
+++ b/quicksync/Readme.md
@@ -0,0 +1,46 @@
+# Fast replay
+
+Binance chain team provide a snapshot data of all history blocks and state for downloading. Full nodes can replay blocks locally without verification after downloading the data.
+
+## Before start
+Some suggestions before using the tool:
+- Local publisher is a better choice than Kafka publisher. The local file of the local publisher can compress and rotate according to the config  `localMaxSize` and `localMaxAge` of ${chain-home}/config/app.toml, so the size of the local file won't be an issue, and the data can be reused. The binance chain client takes around 7ms to publish a Kafka message when the Kafka cluster contains 3 nodes. Because the `ack` of Kafka needs all isr to confirm, which may become the bottleneck. If Kafka is essential, then a single node Kafka will improve the performance.
+- As it use memory DB to store state, a VM with large memory will help.
+
+## Step
+- Download snapshot from the [url](https://s3.ap-northeast-1.amazonaws.com/dex-bin.bnbstatic.com/chain-analisis-download-data/server-bnbchaind-disaster-node-gaiad-data.zip?AWSAccessKeyId=AKIAYINE6SBQLLLS7OXI&Signature=lKVPVfQSxVewd2AZNaw81TgGWqs%3D&Expires=1610338195) provided by binance team and unzip it into the ${chain-home}/data. You can download by `nohup curl -s ${the download url} > data.zip &`.  
+- Download new binary from current [page](https://github.com/binance-chain/node-binary/blob/quicksync/quicksync/bnbchaind). 
+- Modify the config `with_app_stat = true` of ${chain-home}/config/config.toml into `with_app_stat = false`.
+- Modify all the config `indexer = "kv"` of ${chain-home}/config/config.toml into `indexer = "null"`. 
+- Modify the config `state_sync_reactor = true` of ${chain-home}/config/config.toml into `state_sync_reactor = false`.
+- Backup the state DB by `mv ${chain-home}/data/application.db ${chain-home}/data/application.db_backup`. This operation only need do once.
+- Start node by `nohup ./bnbchaind start --iavl-mock  true   --home ${your-chain-home}  &`.  
+- Each time to replay from genesis again, please do `rm -rf ${chain-home}/data/application.db ${chain-home}/data/cs.wal ` before starting the binary.
+
+## What to do after finish replaying blocks 
+Once the replay finish, the process will stop automatically.
+
+- Replace the binary of the official one.
+- Recover the state DB by `mv ${chain-home}/data/application.db_backup ${chain-home}/data/application.db`. 
+- Modify the config `with_app_stat = false` of ${chain-home}/config/config.toml into `with_app_stat = true`.
+- start node by `nohup ./bnbchaind start --home ${your-chain-home} &`
+
+
+## Difference between new binary and official one:
+- Merkle tree implement is replaced by memory DB.
+- No root hash verification, no basic verification for tx, no signature verification.
+- No snapshot at breath block.
+- No replay of validatorSet change.
+
+## Test result:
+- **Hardware**: 8core,32G linux
+- **Main-net**
+    - **Settings**: local publisher with all topic, replay 17927253 blocks, the number of tx is 3686353
+    - **Result**: It takes 574 minutes, the the average speed is 520 blocks/seconds.  
+- **QA-net**
+  - **Settings**: local publisher with all topic, replay 2000000 blocks, the number of tx is 2676645
+  - **Result**: It takes 54 minutes, the average speed is 617 block/seconds. When there are 50 txs in one block, the speed is about 140 blocks/second. If the block is empty, the speed is about 625 blocks/second.
+
+
+
+
diff --git a/quicksync/bnbchaind b/quicksync/bnbchaind
new file mode 100755
index 0000000..f5fd885
--- /dev/null
+++ b/quicksync/bnbchaind
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:576e565eb45e095715b7074cca2d57f6457eae0bb33e61ef418203246bf15597
+size 46515632