* [LSF/MM ATTEND] Filesystems, Btrfs, topics from Oracle
@ 2014-01-15 3:02 Liu Bo
0 siblings, 0 replies; only message in thread
From: Liu Bo @ 2014-01-15 3:02 UTC (permalink / raw)
To: lsf-pc@lists.linux-foundation.org, linux-fsdevel@vger.kernel.org
Hi everyone,
I'd like to attend the LSF/MM conference. I've been doing various file system
works for the last 3 years, most of that focuses on btrfs.
* I built up btrfs's error handling framework which works similar to what
ext4_std_error does.
* I placed much energy on btrfs's fsync performance and improve it to be
an acceptable number.
* I got snapshot-aware defrag project to work which addresses serious
defragment problems among multiply snapshots.
* I added nocow and compress control on per file basis to btrfs.
* I wrote btrfs inband dedup.
---
TOPICs that I want to raise for disscusion,
1. 'NO SPACE', ENOSPC
Btrfs needs to reserve metadata space for updating its core structure Btree,
sinc we cannot predict how much we'll use precisely, now btrfs has to be
pessimestic and reserve metadata space as the worst case for each update.
As it's always 'the worst case', it's inevitable to waste metadata space due to
the current design.
So far we've made some progress on this, with overcommit stuff, it works
better, but with dedup enabled, both space pressure and scalability become big
troubles again.
Also, our algorithm of global_rsv calculation(reservation for core trees like
tree-roots tree, space allocation extent tree) is no more fit in the case of
dedup.
So I'd like to talk with people about some potential solutions and the progress
and analysis I made.
2. Memory shrinker for btrfs
Memory can be pressure but btrfs just uses too much, it'd better we get a
reasonable shrinker for this.
3. Multiple blocksize for dedupe and Bloom filter
For dedupe, we currently only implement fixed blocksize, ie 4K, 8K, etc, but in
practice use, multiple blocksize for dedup will be more flexible.
And bloom filter is a good case for finding if the extent is an unique one,
but it's not easy to do 'delete' operations in bloom filter system because much
update can be invovled. So I'd like to share views about these with people to
see how far we can get, I believe that if we get a fairly good bloom filter,
other subsystems can get a number of benefits as well.
4. Oracle request for btrfs
As oracle's linux has been widely used in customers, there are some requested
features for btrfs, I'd like to share these to see if it's worth making
short-term or long-term plans somehow.
thanks,
-liubo
^ permalink raw reply [flat|nested] only message in thread
only message in thread, other threads:[~2014-01-15 3:02 UTC | newest]
Thread overview: (only message) (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2014-01-15 3:02 [LSF/MM ATTEND] Filesystems, Btrfs, topics from Oracle Liu Bo
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).