QFS version 1.2.0 with Amazon S3 support is out. Check out the blog post here for details.

Quantcast File System

Quantcast File System (QFS) is a high-performance, fault-tolerant, distributed file system developed to support MapReduce processing, or other applications reading and writing large files sequentially.

QFS servers have been tested on 64-bit CentOS 6 extensively and run on Linux variants. The QFS client tools work on OS X and Cygwin as well.

Platform	Build Status
Mac OS X
Ubuntu 14.04
CentOS 6
CentOS 7

The implementation details and features of QFS are discussed in detail in the project wiki.

Getting QFS

QFS is available through various channels.

• BigTop: QFS packages for rpm and debian based systems are available through the BigTop project.
• Binary Distributions: There are various binary distributions packaged as tarballs available on the Binary Distributions Page.
• Compile from source: QFS can be compiled on Linux variants, Mac OS X, and Cygwin. See the wiki for more information on how to build QFS yourself.

Trying QFS

Once you have aquired QFS through one of the methods above, you can take QFS for a quick test drive. Setting up a single node configuration to familiarize yourself with QFS is very easy.

1. Extract the distribution tarball.

    $ tar -xzf qfs.tgz && cd qfs
   

2. Set up a single node QFS instance. This will create a workspace in ~/qfsbase, start two chunk servers and one metaserver.

    $ ./examples/sampleservers/sample_setup.py -a install
    Binaries presence checking - OK.
    Setup directories - OK.
    Setup config files - OK.
    Started servers - OK.
   

3. Add tools binary path toPATH

    $ PATH=${PWD}/bin/tools:${PATH}
   

4. Make a temporary directory on the file system

    $ qfsshell -s localhost -p 20000 -q -- mkdir /qfs/tmp
   

5. Create a file containing "Hello World", Reed-Solomon encoded, with replication 1.

    $ echo 'Hello World' | cptoqfs -s localhost -p 20000 -S -k /qfs/tmp/helloworld -d -
   

6. Cat the file content.

    $ qfscat -s localhost -p 20000 /qfs/tmp/helloworld
   

7. Stat the file to see encoding (RS or not), replication level, and mtime.

    $ qfsshell -s localhost -p 20000 -q -- stat /qfs/tmp/helloworld
   

8. Copy the file locally to the current directory.

    $ cpfromqfs -s localhost -p 20000 -k /qfs/tmp/helloworld -d ./helloworld
   

9. Remove the file from QFS.

    $ qfsshell -s localhost -p 20000 -q -- rm /qfs/tmp/helloworld
   

10. Stop the servers.

     $ ./examples/sampleservers/sample_setup.py -a stop
    

11. Uninstall the single node instance.

     $ ./examples/sampleservers/sample_setup.py -a uninstall
    

Benchmarking QFS

A performance comparison between QFS and HDFS 1.0.2 shows QFS is faster both at reading and writing 20 TB of uncompressed data on our test system, a heterogeneous cluster with 6,500 disk drives. See more information regarding this in our wiki.

Contributing to QFS

We welcome contributions to QFS in the form of enhancement requests, patches, additional tests, bug reports, new ideas, and so on. Please submit issues using our JIRA instance and refer to the QFS code contribution policy when contributing code.

Have Questions?

Join the QFS Developer mailing list or search the archives at the Google Group.

Post comments or questions to [email protected].

License

QFS is released under the Apache 2.0 license.