A new era in AFS development and user capability has dawned. Long held as a mainstay of shared file systems, AFS (an enterprise file system with a data management model) is a prime open source candidate because so many AFS clients depend heavily on source code access and manipulation. Until recently, source licenses for AFS remained under lock and key and bound those clients who bought source licenses to strict confidentiality. But on August 15, 2000, IBM announced the open sourcing of AFS under the IBM Public License (IPL). The source code will be made available this month (September 2000) in the developerWorks Open source zone.
AFS has been commercially available from IBM's Transarc Lab for over 10 years. Files stored in AFS are accessed through user installations, which include commercial, government, and university clients, many of whom are active members of the open source community. Opening the code should facilitate faster and more efficient development of AFS files through increased cooperation between users. The open sourcing applies only to AFS; DFS is not being open sourced.
IBM hopes that opening AFS will speed the pace of innovation and make it easier for users to collaborate in the area of enterprise file sharing. Once the code is opened, clients will be able to jointly work on current projects, many of which stand to benefit immensely from such cooperation. Ports previously unsupported by Transarc and AFS may also begin to see development and significant growth. The University of Michigan, for example, did disconnected work (in which files not connected to the network, user profiles, etc. can be accessed) years ago that can now be publicly added to AFS. Other major contributors and clients who have been excited about the move include Carnegie Mellon, Intel, Morgan Stanley, and United Airlines.
Developers in the client pool for AFS are already active members in the open source community and will appreciate and understand how to work with an open source product. These developers will most likely comprise the primary developer pool for the newly formed Open AFS. Corporate employees, IBM personnel, members of universities and research labs, and commercial customers of AFS will be able to join together in a shared marketplace of ideas for AFS use and development. Because of the overlap between AFS clients and Linux, interest has also been expressed in bringing AFS into the Linux community. Rumor has it that an old Mac port and a Linux port are already waiting in the wings.
The AFS 3.6 code base will be forked to form Open AFS and IBM AFS, because core elements of the AFS code cannot, for legal reasons, be open sourced. IBM is legally bound, for example, to keep Dev (the encryption technology) closed, because the government will not allow this technology to be shipped out to certain countries. Other elements of the code are also under similar licenses that do not allow their release, and neither the IPL or the GPL are broad enough to provide for sub-licensing these or publishing them separately, were the entire AFS code to be released.
Open AFS will be open sourced under the IBM Public License (IPL). An advisory board (six to eight members) will interface with the community and accept/reject changes to the official version of Open AFS. This board will include one IBM technical person, one IBM business person, and roughly six members of the AFS community. A core group (currently 5 members) for the Advisory Board is being appointed, including Laura Stentz of IBM and an IBM technical person, Peter Honeyman (director of the CIT center at University of Michigan), a representative from Carnegie Mellon, where the AFS project began its early development in the IBM-funded Information Technology Center, and a representative from MIT, which has also done a lot of work on AFS. The board will meet to finalize the process for accepting new members, code contributions, and so on as the project evolves.
A commercially supported AFS (under separate development) will continue to be available through IBM. By offering a commercial product alongside Open AFS, IBM hopes to provide a wider range of choices and flexibility on AFS deployment in individual environments. This will provide product reliability and stability for long-standing AFS users who cannot, for managerial and practical logistics, make the switch to Open AFS. Although it is expected that all users will participate in and benefit from the opening of AFS code, many large corporations rely heavily on an existing infrastructure that depends on a continued commercial relationship with IBM and Transarc support.
It is expected that many smaller companies previously priced out of the client base for AFS, university labs, and research centers will be the primary users of Open AFS as their sole resource. These groups can also expect to see a substantial influx of funding from larger more commercial users of AFS, who will have a vested interest in seeing the product grow and flourish in the open source community. There is also a chance that as Open AFS development widens, IBM will consider expanding the current AFS ports.
Because IBM will not control the development of Open AFS, the company will not offer support services for Open AFS. IBM/Transarc will continue to sell, maintain, port (to new versions of currently-supported operating systems), support, and provide minor enhancements to IBM AFS. Although there is no guarantee that IBM and Open AFS will be compatible once separate development begins, the company will clearly have a vested interest in making sure that the commercial version supports the open version. Although the two products will officially diverge in code, IBM will be represented on the Open AFS board, and will contribute to Open AFS development through funding and lab research. Open AFS servers will continue to be able to access IBM AFS servers, and the Open AFS committee will make strong efforts to maintain compatibility.
AFS (and Transarc) began development at Carnegie Mellon's Information Technology Center in the mid 80s, and AFS was launched on campus in 1985. In 1989 Transarc spun off from Carnegie Mellon and began proprietary development. In May of 1989 Transarc began to sell commercial support for AFS. IBM bought the Transarc lab in 1994, and completed integration in 1999.
Along with AFS, Transarc works on its sister product DFS. The lab collaborated in the early 90s with the Open Software Foundation on the DCE distributing environment initiative. They also do development on Edgeserver and WebSphere Application Server, Enterprise Edition.
Transarc will no longer be officially staffed with Open AFS developers, although many of the lab's members have expressed interest in private contributions and will work on Open AFS in their spare time. Contributions are also expected from the IBM Almaden research lab, many of the developers that work on ADS and DFS, and pockets of expertise within the IBM developer community.
- Check out OpenAFS for sources, binaries, and documentation.
- Compare AFS to arla, another limited open source AFS.
- Check out the Coda File System from Carnegie Mellon University.
- Learn about xFS, the serverless network file service from the University of California at Berkeley.
- Learn about XFS, a high-performance journaling file system from sgi.
- Visit the Carnegie Mellon Computer Science Department, where AFS and Transarc were born.
-
LCS, the MIT Laboratory for Computer Science, continues to be a major sponsor and contributor to AFS Open.
Maya Stodte is a contributing writer and editor for developerWorks, and freelance writer and researcher. She can be reached at mstodte@pop.rcn.com.
Comments (Undergoing maintenance)





