A Toolkit Approach to Partially Connected Computing

Summary by Michael Stok

This presentation described a solution to the problem of effectively allowing people to work on their files both at home and at work without a cheap, fast, reliable connection between the two locations. The "occasional" connectivity problems that ISPs and the infrastructure providers suffer from are common and severe enough to make the use of a home system as a remote terminal impractical. Dan chose a toolkit approach rather than a more heavyweight solution.

The nature of the problem was particularly amenable to the development of a kit of user-level tools because there was a single user modifying a relatively small subset of files in part of the filesystem, and there was a good opportunity to sync the filesystems while Dan commuted between home and work. Because the toolkit was implemented to address a single user, there is no need for system-level locking while the systems are transferring data, and some of the things that might normally have been parameterized out into configuration files were hard-coded into the tools.

On the home end of things, the caching is effected through a modified Amd that fetches files on demand from the office server if there isn't an up-to-date local copy. To the work system, this looks like a typical NFS client, but the modified Amd is lazy about getting files. Amd is the caching tool in the toolkit.

At the end of a session, the filesystem is characterized by performing a recursive checksum on all files and checksumming the directories. The checksum of a directory can be used to tell whether the tree under that directory needs to be refreshed. The checksums are generated by the checksum tool and used by the comparison tool either on a file by file basis or in a batch mode.

Because the system was developed to employ only user level tools (the only administrative changes were to allow NFS export of the home area on the work system), some pieces of state live in the filesystem as regular files. The checksum files and files to act as deleted file markers are distinguished by naming convention only, being "invisible" to the user by virtue of being dot files. Combined with the need not to modify the system, these features may have caused the system to be described as a hack during the questions, but Dan does have a system that allows him to work that he constructed with user-level tools.

Originally published in ;login: Vol. 22, No.2, April 1997.