Bostic on BSD

O'Reilly's FreeBSD DevCenter is currently running an interview with Keith Bostic, one of the principal architects of 4.4 and 4.4-lite BSD, and a co-founder of BSDI. In it he discusses the genesis of BSDI, the merits of the BSD and Linux development models, marketing open source software, and more.

Re:about this fs argument.

[Anonymous Coward]

FreeBSD, out of the box, mounts all of the filesystems with synchronous I/O. This isn't the fastest thing to do, but async is a _very_ dangerous thing to do, and isn't considered safe for mission critical deployment.

Correction: async is unsafe for UFS filesystems and their children (like FFS). It's fine on FS which were designed to run async (ext2) or which don't need to worry about it for other reasons (logging / journalling filesystems like ext3, xfs, and reiserfs for linux, or the *BSD logfs, though the latter hasn't worked in years).

The point is, you've got ext2 which is either slow and reliable or fast and easily corrupted, or FFS/Softupdates which is fast and doesn't corrupt easily.

No. ext2 does essentially the same thing as softupdates when in async mode, and has for years, spawning a kernel thread which ensures metadata integrity. Also note that even with softupdates, FFS is still noticeably slower than ext2 for most operations (though FFS is much faster than ext2 at deleting large directories).

Re:about this fs argument.

[Score Whore]

It's not a question of whether you can lose data, it's a question of how often your FS can be non-consistent, how recoverable it is, and how much of your filesystem an average panic can destroy. With sync metadata you can lose the contents of a single file that was updating when the system failed. With async you can't guarantee that status of your metadata which means that you can't automatically recover your FS. The filesystem that is invulnerable to all possible failures doesn't exist.

about this fs argument.

[oka]

I may be wrong, but isnt this 'BSD does not corrupt / Linux corrupts fs' because Linux mounts by default filesystems in async mode, BSD in sync mode?

Clarifications?

[dlc]

Anyone know what this statement means?

• Linux isn't really built around a single code base, although it is certainly more of a single code base than BSD is.

I was under the impression that the Linux code base was unified. Sure, the drivers are contributed and maintained by random folk, but Linus has the final say over what is part of Linux. This sounds like a single code base to me, as opposed to the *BSD's, where there are four code bases which have almost nothing in common except history.

Can anyone clarify this for me?

darren

Cthulhu for President! [cthulhu.org]

Re:about this fs argument.

[bugg]

This is a fairly moot argument, and I'll explain why in a moment.

FreeBSD, out of the box, mounts all of the filesystems with synchronous I/O. This isn't the fastest thing to do, but async is a _very_ dangerous thing to do, and isn't considered safe for mission critical deployment.

Now, the vast majority of BSD users use softupdates. These let safe writes to be performed, and crushing the speed of even async in many ways. Read more at http://www.mckusick.com/

The point is, you've got ext2 which is either slow and reliable or fast and easily corrupted, or FFS/Softupdates which is fast and doesn't corrupt easily.

Have a nice day.