Kernel Elevators Considered Inelegant (Was: Re: elevator algorithm considered irrelevant)

Craig Milo Rogers (rogers@ISI.EDU)
Thu, 19 Nov 1998 10:40:55 -0800

>Taking this one step further, you could view the entire sorting problem
>as a scheduling problem. Larry McVoy proposed priority sorting. What
>you really want to do is weigh the various factors when deciding how to
...
>Essentially, you sort requests according to a goodness() function which
>might be as simple as FIFO or ascending (one-way elevator) but it might
>also be something else.

Ideally, we boot the system off with a lean, clean, mean
block I/O scheduling policy (first-come-first-served? one-way
elevator?), then load other policies as modules (with a clean
separation between the common kernel data structures and any
policy-specific data structures). Once the basic mechanism is in
place for rapid experimentation, people can argue numbers and
methodology instead of theory and philosophy.

Craig Milo Rogers

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/