cRIO-9068 stream to USB mass storage device

[BradM]

In its default mode, ext3 will be a transactional system (wherein the journal contains transactions that include metadata as well as file data).

Also, you are likely thinking of the fsck times, in journaled filesystems a recovery after incomplete writes will simply replay the journal, taking a fraction of a second. Many desktop distros will force a fsck if it detects that the filesystem has not been cleanly unmounted, however I believe the udev listing for mounting new devices does not actually contain the checks to kick off a fsck prior to mounting the filesystem.

[QFang]

My understanding is that the normal/default way to mount an ext3 is NOT to include the data as it makes for potentially very large journals.  Also, replaying the journal is not a "fraction of a second" operation according to e.g. the marketing slides, however I reckognize the neutrality of that article is not assured, there is still some good information there.

Further, reading on the ext3 wiki, it seems a lot of the details of an ext3 mount really depends on how it is mounted, though the lack of CRC's in the journal data seems to indicate this system is NOT safe for use-cases where power could be interrupted because the journal could end up with data in a bad state, thus "play back" corrupt data.  See http://en.wikipedia.org/wiki/Ext3#No_checksumming_in_journal

In another discussion (here) it's stated that ext3 does not handle folders with 5k+ files (unless it has an additional extension) and a link from that discussion points to a comparison showing various performance tests where ext3 shows a couple of strengths and several weaknesses (relatie to the other tested systems).

If I'm missunderstanding something with regards to the journal checksum issue that is described in the wiki link above, please let me know.  This, more than the potential performance issues, concerncs me the most, followed by the possible boot-time and run-time performance issues.

Given the NI RT distro, has anyone tried to extend for ext4 (which at least adds the checksum to the journal, along with some other improvements)?  -Perhaps this is relatively straight forward?

Thanks for participating in the discussion.

[QFang]

I'm sorry for the hi-jack, I've started my own discussion here

Please follow it if you can!

[metux]

Also, replaying the journal is not a "fraction of a second" operation according to e.g. the marketing slides,

These numbers are just wrong. We've measured ext3 and ext4 carefully in big datacenter environments (petabyte scale clusters). 

Oh, btw, if you wanna have a transactional filesystem w/ COW use bttrfs.

In another discussion (here) it's stated that ext3 does not handle folders with 5k+ files

Such huge directories are pretty unusual - even on big spooling systems (eg. mail servers w/ a million of users), so it wasn't optimized for that in older versions, but fixed later. (yes, as some data structures needed a change, this was implemented at an extra feature, that has to be enabled explicitly). I doubt that anybody needs 5+ files per dir in an embedded system.

OTOH, ext3 was superseded by ext4, which was superseded by bttrfs.