/sdfs

Deduplication Based Filesystem

Primary LanguageJava

sdfs

What is this?

A clustered deduplicated file system.

License

GPLv2

Requirements

System Requirements

1. x64 Linux Distribution. The application was tested and developed on ubuntu 14.0.4 and CentOS 7
2. At least 4 GB of RAM
3. Java 8 - available at https://jdk8.dev.java.net/
4. JSVC (http://commons.apache.org/proper/commons-daemon/jsvc.html)

Optional Packages

* NFS Kernel Server
* LIO - for ISCSI support

Installation

Ubuntu/Debian (Ubuntu 14.04+)

Step 1: Download the latest sdfs version
	wget http://opendedup.org/downloads/sdfs-latest.deb

Step 2: Install sdfs and dependencies
	sudo apt-get install fuse libfuse2 ssh openssh-server jsvc libxml2-utils
	sudo dpkg -i sdfs-latest.deb
	 
Step 3: Change the maximum number of open files allowed
	echo "* hard nofile 65535" >> /etc/security/limits.conf
	echo "* soft nofile 65535" >> /etc/security/limits.conf
	exit
Step 5: Log Out and Proceed to Initialization Instructions

CentOS/RedHat (Centos 7.0+)

Step 1: Download the latest sdfs version
	wget http://opendedup.org/downloads/sdfs-latest.rpm

Step 2: Install sdfs and dependencies
	yum install jsvc libxml2 java-1.8.0-openjdk
	rpm -iv --force sdfs-latest.rpm
	 
Step 3: Change the maximum number of open files allowed
	echo "* hardnofile 65535" >> /etc/security/limits.conf
	echo "* soft nofile 65535" >> /etc/security/limits.conf
	exit
Step 5: Log Out and Proceed to Initialization Instructions

Step 6: Disable the IPTables firewall

	service iptables save
	service iptables stop
	chkconfig iptables off

Step 7: Log Out and Proceed to Initialization Instructions

Initialization Instructions for Standalone Volumes

Step 1: Log into the linux system as root or use sudo

Step 2: Create the SDFS Volume. This will create a volume with 256 GB of capacity using a Variable block size.
	**Local Storage**
	sudo mkfs.sdfs --volume-name=pool0 --volume-capacity=256GB

	**AWS Storage**
	sudo mkfs.sdfs --volume-name=pool0 --volume-capacity=1TB --aws-enabled true --cloud-access-key <access-key> --cloud-secret-key <secret-key> --cloud-bucket-name <unique bucket name>
	
	**Azure Storage**
	sudo mkfs.sdfs --volume-name=pool0 --volume-capacity=1TB --azure-enabled true --cloud-access-key <access-key> --cloud-secret-key <secret-key> --cloud-bucket-name <unique bucket name>
	
	**Google Storage**
	sudo mkfs.sdfs --volume-name=pool0 --volume-capacity=1TB --google-enabled true --cloud-access-key <access-key> --cloud-secret-key <secret-key> --cloud-bucket-name <unique bucket name>
	
	
Step 3: Create a mount point on the filesystem for the volume

	sudo mkdir /media/pool0

Step 4: Mount the Volume

	sudo mount -t sdfs pool0 /media/pool0/

Set 5: Add the filesystem to fstab
	pool0           /media/pool0    sdfs    defaults                0       0

Initialization Instructions for A Multi-Node Configuration

In a mult-node cluster there are two components :

	* The Dedup Storage Engine (DSE) - This is the server/service that store all unique blocks. For redundancy, you should have at least two of these. They can live on the same 			  physical server as other components but this is not recommended for redundancy
	* The File System Service (FSS) - This is the server/service that mounts the sdfs volume. Multiple FSS services each serving its own volume can live in the same cluster.
Prerequisits :

	* Static IP address and unique name for all nodes: All nodes should have a static IP address to reduce network issues.
	* Disable host firewalls : IPTables should be disabled for testing and then enabled again if needed with open ports for multicast and udp.
	* Network Communication : Multicast communication is the default configuration. All nodes in the cluster will need to be able to talk to eachother using multicast unless 			  the Jgroups configuration is modified.Multicast communication is usually not a problem on the same subnet but can be an issue between subnets or over wans
	* Fast Low Latency Network : SDFS will work within most networks but speed will suffer is network speeds low or network latency is high. 1Gb/s or above dedicated network is 			  recommeded. 
	* Update your /etc/sysctl.conf to improve network IO performance. Below are some recommeded options :
		vm.swappiness=1
		net.core.rmem_max=254800000
		net.core.wmem_max=254800000
		net.core.rmem_default=254800000
		net.core.wmem_default=254800000
		net.core.optmem_max=25480000
		net.ipv4.tcp_timestamps=0
		net.ipv4.tcp_sack=0
		net.core.netdev_max_backlog=250000
		net.ipv4.tcp_mem=25480000 25480000 25480000
		net.ipv4.tcp_rmem=4096 87380 25480000
		net.ipv4.tcp_wmem=4096 65536 25480000
		net.ipv4.tcp_low_latency=1
		Creating and Starting a DSE

The following steps will create 2 DSEs on server1 and server2

Step 1: Create a DSE on Server1 using a 4K block size and 200GB of capacity and cluster node id of "1"

	mkdse --dse-name=sdfs --dse-capacity=200GB --cluster-node-id=1

Step 2: Edit the /etc/sdfs/jgroups.cfg.xml and add the bind_addr attribute with Server1's IP address to the <UDP> tag.

	<UDP         
	mcast_port="${jgroups.udp.mcast_port:45588}"         
	tos="8"         
	ucast_recv_buf_size="5M"       
	 ucast_send_buf_size="640K"         
	mcast_recv_buf_size="5M"       
	 mcast_send_buf_size="640K"         
	loopback="true"         
	max_bundle_size="64K"         
	bind_addr="SERVER1 IP Address"

Step 3: Start the DSE service on Server1

	startDSEService.sh -c /etc/sdfs/sdfs-dse-cfg.xml &

Step 4: Create a DSE on Server1 using a 4K block size and 200GB of capacity and cluster node id of "1"

	mkdse --dse-name=sdfs --dse-capacity=200GB --cluster-node-id=1

Step 5: Edit the /etc/sdfs/jgroups.cfg.xml and add the bind_addr attribute with Server1's IP address to the <UDP> tag.

	<UDP
	mcast_port="${jgroups.udp.mcast_port:45588}"
	tos="8"
	ucast_recv_buf_size="5M"
	ucast_send_buf_size="640K"
	mcast_recv_buf_size="5M"
	mcast_send_buf_size="640K"
	loopback="true"
	max_bundle_size="64K"
	bind_addr="SERVER1 IP Address"

Step 6: Start the DSE service on Server1

	startDSEService.sh -c /etc/sdfs/sdfs-dse-cfg.xml &

Creating and Starting a SDFS FSS - Mounting a Volume

The following steps will create 1 SDFS Volume on Server0. These are run as root.

Step 1: Create the volume configuration. The following will create a volume using 4K blocks. This must match the block size used by the DSEs

	mkfs.sdfs --volume-name=pool0 --volume-capacity=400GB --chunk-store-local false

Step 2: Create a mount point on the filesystem for the volume

	mkdir /media/pool0

Step 3: Mount the SDFS FSS (Volume)

	mount.sdfs pool0 /media/pool0

Verify its all Working

Step 1: Make sure you can see all your DSE Servers from the FSS Node. 

	On the node that you mounted the volume run sdfscli --cluster-dse-info. You should see all the nodes in the cluster. If you don't then there is probably a networking issue 			somewhere that is preventing them from talking.

Step 2: Make sure you can see all of your other volumes in the cluster.

	On any node that has a mounted volume run sdfscli --cluster-volumes. You should see all of the volumes in the cluster.

Step 3: Copy some data to the volume and make sure it all is there

	md5sum /etc/sdfs/pool0-volume-cfg.xml
	cp /etc/sdfs/pool0-volume-cfg.xml /media/pool0
	md5sum /media/pool0/pool0-volume-cfg.xml

Troubleshooting and other Notes

Running on a Multi-Node clusting on KVM guest.

By default KVM networking does not seem to allow guest to communicate over multicast. It also doesn't seem to work when bridging from a nic. From my reseach it looks like you have 		to setup a routed network from a KVM host and have all the guests on that shared network. In addition, you will want to enable multicast on the virtual nic that is shared by those 		guest. Here is is the udev code to make this happen. A reference to this issue is found here.

# cat /etc/udev/rules.d/61-virbr-querier.rules 
ACTION=="add", SUBSYSTEM=="net", RUN+="/etc/sysconfig/network-scripts/vnet_querier_enable"

# cat /etc/sysconfig/network-scripts/vnet_querier_enable 
#!/bin/sh
if [[ $INTERFACE == virbr* ]]; then
    /bin/echo 1 > /sys/devices/virtual/net/$INTERFACE/bridge/multicast_querier
fi

Testing multicast support on nodes in the cluster

The jgroups protocol includes a nice tool to verify multicast is working on all nodes. Its an echo tool and sends messages from a sender to a reciever

On the receiver run

	java -cp /usr/share/sdfs/lib/jgroups-3.4.1.Final.jar org.jgroups.tests.McastReceiverTest -mcast_addr 231.12.21.132 -port 45566

On the sender run

	java -cp /usr/share/sdfs/lib/jgroups-3.4.1.Final.jar org.jgroups.tests.McastSenderTest -mcast_addr 231.12.21.132 -port 45566

Once you have both sides running type a message on the sender and you should see it on the receiver after you press enter. You may also want to switch rolls to make sure multicast 		works both directions.

take a look at http://docs.jboss.org/jbossas/docs/Clustering_Guide/4/html/ch07s07s11.html for more detail.

Further reading:

	Take a look at the administration guide for more detail. http://www.opendedup.org/sdfs-20-administration-guide

Ask for Help

	If you still need help check out the message board here https://groups.google.com/forum/#!forum/dedupfilesystem-sdfs-user-discuss