Experiences with ext3? (cont.)

[Duckula]

(extending [webmasterworld.com...] )

Kerneltrap has posted benchmarks (for the future version) and some discussion on the topic: Linux: Benchmarking Filesystems In 2.6.0-test2 [kerneltrap.org]

Journalled filesytems have a runtime cost, and you're paying that all the time.

If you're going 200 days between crashes on a disk-intensive box then using a journalling fs to save 30 minutes at reboot time just doesn't stack up:

you've lost much, much more time than that across the 200 days.

It all depends on what the machine is doing and what your max downtime requirements are.

[Nick_W]

Looks like ext3 came out fairly well. I can't understand why ext2 wasn't included though? (i know it's not a journaling system and there was some talking of using it as a 'baseline')....

I had a powercut last night, ext3 handles that very, very well...

<added>You added to your post as I replied, now I look like I don't have a clue ;) - That's actually mostly true. But, from a Linux Desktop users point of view I do find ext3 to be very reliable and stable. As far as servers go, well, I'll leave that to the more tech minded Linux users....</added>

Nick

[waitman]

Hello

If you only have remote access to the box, life may not be pleasant fixing an ext2 filesystem if the box loses power.

Perhaps I am mistaken, however in the past I have had to get in front of the machine, boot into single user mode and fix the filesystem "manually" when I used ext2. Losing power on an ext2 system can also royally screw your database tables.

I haven't had any trouble with performance with ext3, however I honestly haven't paid much attention to it. It makes sense that you pay something of a performance penalty.

Take care,

[drbrain]

That's because by default, ext2 uses an async mount. Using any filesystem with an async mount is like trying to build a card house on a folding table with a leg that keeps collapsing. While it'll be ok so long as you don't bump the table, when you do it may all go *poof*.

This is why a journaled system, or even better, UFS with softupdates is preferred. With UFS, you get the benefits of having your metadata always being in a consistent state and short fsck times, without wasting space on a journal.

In FreeBSD 5.x, you even get a background fsck, so your system can come up immediately and fsck in the background while your system is operational.

[peterdaly]

We have terrabye volumes (more or less.) Downtime (days) caused by an fsck is much more of a problem then the slower real-time performace.

It all depends on what you are using the system/partition for. Use the right tool for the job.

-Pete