Re: [Gnu-arch-users] Re: give us a hand with arch

[Tom Lord]

    > From: Dustin Sallings <address@hidden>

    [re: small file performance on ext*]

    >   I'm kind of reading this thread as having certain
    > optimizations of arch rely on filesystem improvements in the
    > Linux kernel.  Does this assume that the majority of arch users
    > will be running Linux, or just those who will have these
    > optimizations available?


On my BSD box, at least, small file performance is already pretty
good.   It's more like being anxious for Linux to finish catching up
in this regard.

As for the impact on arch users -- it tends to shrink over time,
regardless, as as things like the inode-signature optimization appear
and grow in range of applicability.

Speaking of file systems: people were talking about the trade-offs of
hard-linked working copies.  Can't the same aim be achieved, probably
with greater efficiency, and certainly with greater robustness, by
using shadow directories?

When shadow-directories become ubiquitous enough to start writing
portable code against, arch will be able to commit locally a bit
faster (for the cost of a `tree-lint').  If we can find ways to avoid
the tree-lint, then it can commit instantly.  Regardless, `get' (from
a local archive) can be instant.

-t