NetBSD-Users archive

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Old Index]

Re: state or future of LFS?



>>>>> "ag" == Andreas Gustafsson <gson%gson.org@localhost> writes:

    ag> log-structured file system ... such a perfect match for solid
    ag> state disks

JFFS2 in Linux is an LFS used for NOR flash, and it's mature and works
well for them in many delivered products.  like, it is a real LFS with
a garbage collector and stuff.  It's not suitable for SSD because it
has O(n) mount time and keeps O(n) data in RAM such that it's
basically useless beyond about 0.5GB, but aside from that you may find
their experience relevant.  They use no write buffer at all for JFFS2,
but their current work (ubi/ubifs) does have a write cache.

They are interested in running above raw NOR flash (and for their
current work, raw NAND), so they are worried about wear-leveling now
and about read-disturb and scrubbing for the future, which probably
doesn't apply to SSD's with controllers.

For SSD you cannot evade the proprietary wear-levelling software in
the drive's controller, so the more interesting goal would be an
access pattern that doesn't make drives get ``fragmented'' and degrade
their performance.  I've no idea how to do that.  It may take a while
since we can't see the software inside the drive, and the vendors
won't want to give specifics about the problem unless you wring it out
of them because they want people to think about it as little as
possible.

Attachment: pgpK6ckcZpX8i.pgp
Description: PGP signature



Home | Main Index | Thread Index | Old Index