veyepar: XFS: Teaching an Old Dog Old Tricks

Author(s):	Dave Chinner
Location	Medium Lecture Theatre CB11.00.401
Date	jan Wed 24	Days Raw Files
Start	13:40	First Raw Start	13:12
Duration	0:45:0	Offset	0:27:57
End	14:25	Last Raw End	14:42
Chapters	00:00 0:02:36 0:32:36
Total cuts_time	46 min.

http://lca2018.linux.org.au/schedule/presentation/88/
raw-playlist raw-mp4-playlist encoded-files-playlist host archive public tweet mp4 svg png

assets release.pdf XFS_Teaching_an_Old_Dog_Old_Tricks.json
logs
Admin: episode episode list cut list raw files day marks day marks day image_files

State:

Locked:

clear this to unlock

Locked by:

user/process that locked.

Start:

initially scheduled time from master, adjusted to match reality

Duration:

length in hh:mm:ss

Name:

Video Title (shows in video search results)

Emails:

email(s) of the presenter(s)

Released:

has someone authorised pubication

Normalise:

Channelcopy:

m=mono, 01=copy left to right, 10=right to left, 00=ignore.

Thumbnail:

filename.png

Description:

markdown

XFS has been in production for more than 20 years and has been in the linux
kernel for 15 of them - it is the oldest of the current mainstream
Linux filesystems. The design of XFS was state of the art in the late 1980s,
when btrees, extents and journalling were shiny new concepts. While development
of XFS still continues, it's design and architecture is largely unchanged. many
of the tricks that XFS brought to Linux can now be found in other Linux
filesystems, too. It's the old dog of the pack.

We recently introduced a late-90s technology to XFS: shared data extents and a
copy-on-write IO path. The copy-on-write tree structure that first appeared in
the late 90s can be found in ZFS and BTRFS, but not XFS. So while XFS can now
provide some data manipulations like clones and deduplication, it doesn't provide
any of the more advanced CoW functionality like snapshots that traditional CoW
filesystems inherently provide.

Given that XFS doesn't have a copy-on-write architecture, can we use the new
functionality to provide a similar feature set to native COW filesystems? What
other old tricks do we need to pull out of the bag to make this a reality? And
how much will the result look like XFS?

In this talk I'll outline my crazy plan to add filesystem management concepts to
XFS that only copy-on-write filesystems current provide to Linux users. There
won't be anything "new" here - I'll describe the old tech that inspired the
algorithms and structures that we'll use and, more importantly, some of the
pitfalls and hard lessons learned from the original implementations of these
techniques.

With any luck, I'll present a coherent story of how we are going to bring
subvolumes, snapshots, writeable clones, send/receive remote snapshot
replication, page cache sharing and more to XFS without needing to change it's
underlying 1980s architecture.  XFS may be an old dog, but there's still some
old tricks we can teach it to keep it shiny, new and relevant to users for years
to come.

Comment:

production notes