BOF: Block I/O Enhancements

July 27, 2000
J. Corbet

A birds-of-a-feather session was held for those interested in the improvement of Linux's block I/O subsystem. It's a timely topic, of course, since it is widely understood that the block layer will see a much-needed, major overhaul in the 2.5 development series. The actual discussion was not as illuminating as one might hope, partially because a number of the people who will be involved in the block I/O rewrite were not present at OLS this year.

One interesting theme that came out of the discussion was a degree of discomfort from a number of the attendees that there is no real architectural design which describes just what the new block I/O system will look like. In particular, people from larger companies were nervous about what was going to happen, and whether the final result would meet their needs or not. They were also afraid of coding up some useful enhancement, only to find that it does not easily fit into a new architecture that materializes from elsewhere.

The long-time kernel hackers, instead, were far more relaxed. Their view was that it is better to push toward a new system in small steps, and that competing views of how things should look should be settled in the kernel development bazaar. Ideas of how to do things will change over the next year, and the best solution is just to see which solutions appear, and which of those work the best. People used to Linux kernel development are used to (and comfortable with) this mode of operation, but companies who are increasingly committing themselves to Linux remain a little nervous.

A side topic that came up was whether it is time, in 2.5, to aggressively remove support for ancient hardware. Support for ISA adapters, in particular, was mentioned. It is unlikely that ISA support will be carved out anytime soon, but it is true that the kernel eventually needs to move on, lest it bury itself under old code.

Back to OLS 2001 coverage