SoylentNews is people
SoylentNews
An ex-Google engineer is developing a new file system for Linux, with the hopes that it can offer a speedier and more advanced way of storing data on servers.
After a number of years of development, the Bcache File System (Bcachefs) "is more or less feature complete -- nothing critical should be missing," wrote project head Kent Overstreet, in an e-mail to the Linux Kernel Mailing List late Thursday.
Linux currently has plenty of working file systems, though no one file system seems to be best-suited for all uses.
The main goal of Bcachefs is to match the speed of traditional Linux file systems with the advanced scalability and reliability of newer file systems, Overstreet wrote.
Although not a sexy technology, file systems provide the interface to the operating system for storing files on a disk.
The most widely used file system among Linux users is the decades-old Gnu/Linux Extended Filesystem series of filesystems -- Ext4 being the latest release. But many organizations and users have gravitated towards other file systems, such as Btrfs or XFS, to handle very large amounts of data, or to use advanced techniques in ensuring data integrity.
This file system evolved from the work Overstreet did at Google, where he worked as a software engineer for two years from 2011 until 2013 to create caching software.
Bcachefs has all the features of a modern file system, Overstreet wrote, including checksumming to ensure data integrity, compression to save space, caching for quick response, and copy-on-write, which offers the ability for a single file to be accessed by multiple parties at once.
...
Overstreet is working on the file system on his own time, without outside funding. He is seeking other administrators and developers to test the system and even contribute to its development.
Nonetheless, the release of Bcachefs seems to have met with cautious optimism by the Linux professionals on the Hacker News online forum, though one contributor did say of Overstreet that "I hope the guy has a large stash in his bank. File systems take notoriously long to stabilize."
(Score: 2, Informative) by rleigh on Sunday August 23 2015, @11:22PM
Do I really care about those desktop things if I want a good server though? Not really. And desktop-wise it's not /that/ bad; and it's not like Linux made much forward progress over the last decade, it's all just endless churn for the sake of it. The only thing lagging that really matters is graphics support, but it's made big strides the last year or so, and continuing to make progress. It now has largely the same kernel drm and mesa as Linux. I expect I'll be on a FreeBSD desktop sooner rather than later. The only annoyance I have in practice is the GPU fan speed which is defaulting to the max.
And regarding ZFS, its origins are not that important. It wasn't "stolen", it was used under the terms of its licence. Ultimately, all that matters is how useful it is to me and what I can do with it. And really, ZFS is a *killer* feature. Linux has absolutely nothing that competes with it at present, and is unlikely to in the future unless some serious effort is invested. It's the primary reason I moved to a FreeBSD fileserver 20 months ago, and haven't looked back. [Yes, ZFS for Linux is available, but it's lagging behind in features and is not as well integrated. It's definitely better on FreeBSD.] And migrating was as simple as could be--export zpool on Linux, physically transfer disks, zpool import on FreeBSD, job done (well and, zpool upgrade to get more features not available on Linux).
Btrfs could have been "the ZFS for Linux" if some questionable design choices hadn't been made back at the start where they deviated from the ZFS design without appreciating the downsides, and having a lot of implementation bugs for so long in addition to design issues has soured many people to it repeatedly suffering from severe problems. Definitely applies to me. That some of the problems are rooted in its design is quite bad since it makes them practically unfixable. I for one hope that bcachefs does better, but it will be several years before we know if it's worth persuing, or if it has problems in its own right. To be on a par with ZFS the tools need to be well written, and just as importantly well documented. The Btrfs tools and docs were sparse at best, while any task I wanted to perform with ZFS has decent documentation and a plethora of tutorials and examples, which is really needed to be able to set up and admin effectively, as well as be able to read through up front to spec out a system.