The Coda Distributed File System

Carnegie Mellon University has developed an exciting file system. Mr. Braam, one of the developers, tells us all about it.
Coda in Action

Coda is in constant active use at CMU. Several dozen clients use it for development work (of Coda), as a general purpose file system and for specific disconnected applications. The following two scenarios have exploited the features of Coda very successfully. We have purchased a number of licenses for Wabi and Windows software. Wabi allows people to run MS PowerPoint. We have stored Wabi and Windows 3.1 including MS Office in Coda and it is shared by our clients. Of course .ini files with preferences are particular to a given user, but most libraries and applications are not. Through hoarding we continue to use the software on disconnected laptop computers for presentations. This is frequently done at conferences.

Over the years of its use we have not lost user data. Sometimes disks in our servers have failed, but since all of our volumes are replicated, we replaced the disk with an empty one and asked the resolution mechanism to update the repaired server. All one needs to do for this is to type ls -lR in the affected file tree when the new disk is in place. The absence of the file on the repaired server will be noticed, and resolution will transport the files from the good servers to the newly repaired one.

There are a number of compelling future applications where Coda could provide significant benefits.

  1. FTP mirror sites should be Coda clients. As an example, let's take, which has many mirrors. Each mirror activates a Perl script, which walks the entire tree at Red Hat to see what has been updated and fetches it—regardless of whether it is needed at the mirror. Contrast this with Red Hat storing their ftp area in Coda. Mirror sites should all become Coda clients too, but only Red Hat would have write permission. When Red Hat updates a package, the Coda servers notify the mirror sites that the file has changed. The mirror sites will fetch this package, but only the next time someone tries to fetch this package.

  2. WWW replication servers should be Coda clients. Many ISPs are struggling with a few WWW replication servers. They have too much access to use just a single http server. Using NFS to share the documents to be served has proven problematic due to performance problems, so manual copying of files to the individual servers is frequently done. Coda could come to the rescue since each server could be a Coda client and hold the data in its cache. This provides access at local disk speeds. Combine this with clients of the ISP who update their web information off-line and we have a good application for mobile clients too.

  3. Network computers could exploit Coda as a cache to dramatically improve performance. Updates to the network computer would automatically be made as they become available on servers, and for the most part the computer would operate without network traffic, even after restarts.

Our current efforts are mostly to improve the quality of Coda. The rough edges, which inevitably come with research systems, are slowly being smoothed out. Write-back caching will be added in order for Coda to operate much faster. The disconnected operation is an extreme form of write-back caching, and we are leveraging these mechanisms for write-back caching during connected operation. Kerberos support is being added. The networking protocols supporting Coda are making this easily possible. We would like to have cells which will allow clients to connect to more than a single Coda cluster simultaneously. Further ports will hopefully allow many systems to use Coda.

Getting Coda

Coda is available by FTP from You will find RPM packages for Linux as well as tar files of the source. Kernel support for Coda will come with the Linux 2.2 kernels. On the WWW site, you will find additional resources such as mailing lists, manuals and research papers.

Peter adores his wife Anne, and together they love Alaska with its mountains, wildlife and a halfway acceptable population density. Nothing is better than having a moose on your porch there or camping on a not too scary glacier. Until March 1997 Peter was a faculty member in the Mathematical Institute at Oxford. In the summer of 1995 Peter became president of Stelias Computing Inc. which assembled the InfoMagic Workgroup Server. Dabblings in Mach and the GNU Hurd evolved into porting Coda to Linux. E-mails about this with Satya, the visionary leader of the Coda and Odyssey projects, led to a visit to Carnegie Mellon University in late 1996 and eventually to him joining the Computer Science faculty. He is now leading the Coda project. He can be reached at