WebDFS

An Open Source Distributed File System that can be used behind an HTTP server (Apache, lighttpd, Hiawatha, Cherokee etc.) to implement a highly scalable DFS for storing images, etc.

WebDFS
Developer(s)Shane Hill
Stable release
0.12 / October 12, 2009 (2009-10-12)
Operating systemLinux
TypeDistributed file system
LicenseNew_BSD_license
Websitegoogle code repo

webDFS is mostly based on algorithms[1][2][3]

The algorithms come from a family of algorithms known as the RUSH family; Replication Under Scalable Hashing. If built correctly, a system built on the RUSH algorithms will have the following characteristics: (some of the text below is taken from the algorithm whitepaper)

  • Ability to map replicated objects to a scalable collection of storage servers or disks without the use of a central directory.
  • Redistributes as few objects as possible when new servers are added or existing servers are removed
  • Guarantees that no two replicas of a particular object are ever placed on the same server.
  • No central directory, clients can compute data locations in parallel, allowing thousands of clients to access objects on thousands of servers simultaneously.
  • Facilitates the distribution of multiple replicas of objects among thousands of disks. Allows individual clients to compute the location of all of the replicas of a particular object in the system algorithmically using just a list of storage servers rather than relying on a directory.
  • Easy scaling management. Scaling out is just a matter of deploying new servers and then propagating a new configuration to all of the nodes. The data will automatically and optimally be moved to accommodate the new resources. De-allocating resources is basically the same process in reverse. Simply deploy the new configuration and the data will be moved off the old resources automatically. After the data has been moved, simply take the old resources off line.
  • Easier server management. Since there is no central directory, there are no master or slaves to configure. No master or slaves means that all resources are utilized and no servers sit unused as "hot" spares or backups.
  • No single point of failure. As long as the replica to node ratio is correct, your data will be safe, redundant, and durable; able to withstand major server outages with no loss.

References
  1. http://users.soe.ucsc.edu/~elm/Papers/ipdps03.pdf
  2. http://users.soe.ucsc.edu/~elm/Papers/ipdps04.pdf
  3. http://www.ssrc.ucsc.edu/Papers/weil-sc06.pdf
This article is issued from Wikipedia. The text is licensed under Creative Commons - Attribution - Sharealike. Additional terms may apply for the media files.