Re: [ANNOUNCE] DualFS: File System with Meta-data and Data Separation

From: sfaibish
Date: Wed Feb 14 2007 - 16:43:42 EST


On Sat, 10 Feb 2007 22:06:37 -0500, Sorin Faibish <sfaibish@xxxxxxx> wrote:

Introducing DualFS

File System developers played with the idea of separation of
meta-data from data in file systems for a while. The idea was
lately revived by a small group of file system enthusiasts
from Spain (from the little known University of Murcia) and
it is called DualFS. We believe that the separation idea
will bring to Linux file systems great value.

We see DualFS as a next-generation journaling file system which
has the same consistency guaranties as traditional journaling
file systems but better performance characteristics. The new file
system puts data and meta-data on different devices (usually, two
partitions on the same disk or different disks), and manages them
in very different ways:

1. DualFS has only one copy of every meta-data block. This copy is
in the meta-data device,

2. The meta-data device is a log which is used by DualFS both to
read and to write meta-data blocks.

3. DualFS avoids an extra copy of meta-data blocks, which allow
DualFS to achieve higher performance than other journaling file
systems.

4. DualFS implements performance enhancements: meta-data prefetch,
on-line meta-data relocation and faster fsck and mkfs operations.

5. DualFS file system is suitable for use with TB and PB of storage

We have carried out different experiments which compare DualFS and
other popular Linux file systems, namely, Ext2, Ext3, XFS, JFS, and
ReiserFS. The results, both performance and management, prove the
value of the new file system design based on the separation of data
and metadata which increase performance dramatically up to 97% by
simply using an additional partition of same disk.

We have performed extensive tests using micro-benchmarks as well
as macro-benchmarks including Postmark v1.5, SpecWeb99, TPCC-uva.
We also measured performance of maintenance tasks like mkfs and
fsck which all show that DualFS performance is superior to all the
other file systems tested with performance advantage in the range
between 50-300% depending on the benchmark and the configuration.
And all this performance advantage is a direct result of the
separation of the meta-data and data.

The project started in 2000 by Juan Piernas Canovas as the primary
and almost unique contributor, with some small contributions by Toni
Cortes, and Jose M. Garcia. The project was stopped for some time. We
restarted the project last year, and after several months of updates
and tests we created a SourceForge project with the intent to share
the value of this old and yet new concept.

The DualFS code, tools and performance papers are available at:

<http://sourceforge.net/projects/dualfs>

The code requires kernel patches to 2.4.19 (oldies but goodies) and
a separate fsck code. The latest kernel we used it for is 2.6.11
and we hope with you help to port it to the latest Linux kernel.

We will present the architecture, principles and performance
characterization at the LFS07 next week.

We are very interested to get your feedback and criticism.

Sorin Faibish and Juan Piernas Canovas
--------------------------------------



-
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html





--
Using Opera's revolutionary e-mail client: http://www.opera.com/mail/
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/