World's largest telco releases lossless Linux filesystem

From the article:

--- "Log-structured filesystems write down all data in a continuous log-like format that is only appended to, never overwritten. The approach is said to reduce seek times, as well as minimizing the kind of data loss that occurs with conventional Linux filesystems.

For example, data loss occurs on ext3 filesystems when the system crashes during a write operation. When the system reboots, the journal notes that the write did not complete, and any partial data writes are lost." ---

Yeah. The way ext3 is always losing data is a big problem that needs to be fixed. Almost every time I come back to my FC4 box, I find stuff missing.

I'm really glad these guys have come to the rescue.

Seriously though, if the system loses power, ext3 in default "data=ordered" mode gives stronger guarantees than almost any other journalled filesystem. It guarantees that your file data won't be garbaged, though it may not be the *latest* data.

(Most journalled filesystems guarantee that the metadata and structure of the filesystem will retain their integrity, though the actual file data might get garbaged.)

In "data=journal" mode it makes the same guarantees as these guys' filesystem does.

It's nice that they're contributing, but not so nice that they are spreading misinformation about ext3.

NetBSD has had a log-structured filesystem for a while, but it has worked in only the last few releases. That LFS derived from the Berkeley "Sprite" operating system of the early 90s. How does "nilfs" differ from NetBSD's LFS, or indeed, any previous linux Log-structured filesystems (http://outflux.net/projects/lfs/ , and the apparently defunct http://www.complang.tuwien.ac.at/czezatke/lfs.html)?

A small furor erupted over some actual benchmarking of the NetBSD LFS in 1995: http://www.eecs.harvard.edu/~margo/usenix.195/

Who do you believe? All I know is that after NetBSD added Greg Ganger's "soft updates" and a few other software (not disk layout) changes, their FFS speed went way, way up. NetBSD's LFS isn't in any kind of common use, and developers have apparently abandoned the various Linux log-structured filesystems.

Note that I use Reiserfs and ext3 on my machines. I don't have a big fear of trying new filesystems.

> All I know is that after NetBSD added Greg Ganger's "soft updates" and a few other software (not disk layout) changes, their FFS speed went way, way up.

Before "soft updates", they were actually writing metadata synchronously, and casting aspersions on ext2 for not doing so. Big performance hit for that, and they *still* had to run fsck upon unplanned reboots. (In fact, they still need fsck even with soft updates, but it runs a lot faster than without.)