# DCache

**Repository Path**: zhangxiao2000/DCache

## Basic Information

- **Project Name**: DCache
- **Description**: DCache: A Distributed Cache Mechanism for HDFS Based on RDMA.
- **Primary Language**: Java
- **License**: Apache-2.0
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 2
- **Forks**: 0
- **Created**: 2020-08-27
- **Last Updated**: 2022-08-02

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# Introduction

DCache: A Distributed Cache Mechanism for HDFS Based on RDMA.

The design detailed you can find from our **paper**.

Optimize the read and write process of HDFS.

Change the code base on **Hadoop2.9.0**.

Please cite our paper if you think this is useful.

X Zhang,  Liu B ,  Gou Z , et al. DCache: A Distributed Cache Mechanism for HDFS based on RDMA[C]// 2020 IEEE 22nd International Conference on High Performance Computing and Communications IEEE, 2020.


# Compile

## Dependency
1. Java
2. maven
3. protobuf
4. [JUCX](https://github.com/openucx/ucx/tree/master/bindings/java)

The [Jar of JUCX](https://gitee.com/zhangxiao2000/DCache/tree/master/hadoop-2.9.0-src/hadoop-hdfs-project/hadoop-hdfs/src/main/lib) has been included in our project.

You can refer to the community version of Hadoop and use maven to compile.

```shell
mvn package -Pdist,native -DskipTests -Dtar -Dmaven.test.skip=true
```

# Deployment

We just modify the code of read and write of HDFS. So only two Jar files changed: **hadoop-hdfs-2.9.0.jar,hadoop-hdfs-client-2.9.0.jar**. You can complile by source code or download from **Release**.

The full Hadoop you can download from [Apache](https://archive.apache.org/dist/hadoop/common/hadoop-2.9.0/).

If `${HADOOP_SRC}` is the root directory of Hadoop source code. `${HADOOP_HOME}` is the root directory of Hadoop. Then you can replace Jar files by these commands.

```shell
cp ${HADOOP_SRC}/hadoop-hdfs-project/hadoop-hdfs/target/hadoop-hdfs-2.9.0.jar ${HADOOP_HOME}/share/hadoop/hdfs/
cp ${HADOOP_SRC}/hadoop-hdfs-project/hadoop-hdfs-client/target/hadoop-hdfs-client-2.9.0.jar ${HADOOP_HOME}/share/hadoop/hdfs/
cp ${HADOOP_SRC}/hadoop-hdfs-project/hadoop-hdfs-client/target/hadoop-hdfs-client-2.9.0.jar ${HADOOP_HOME}/share/hadoop/hdfs/lib
cp ${HADOOP_SRC}/hadoop-hdfs-project/hadoop-hdfs/src/main/lib/jucx-1.7.0.jar ${HADOOP_HOME}/share/hadoop/hdfs/lib
```

If you want to use hadoop commands to start CacheNode, you need to replace these files.

```shell
cp deploy/hdfs ${HADOOP_HOME}/bin/
cp deploy/hadoop-daemon.sh ${HADOOP_HOME}/sbin/
cp deploy/hdfs-site.xml ${HADOOP_HOME}/etc/hadoop/
```

You can control whether to enable the DCache function by adjusting the parameter named **dfs.client.isSupportRDMA**(`${HADOOP_HOME}/etc/hadoop/hdfs-site.xml`)

# Usage

```shell
#start CacheNode(stop the CacheNode only by kill the process)
hadoop-daemon.sh start cachenode
#cache file
hdfs cacheadmin -addDirective -path <path> -rdma
#uncache file
hdfs cacheadmin -removeDirective -path <path> -rdma
```