"Theodore Y. Ts'o" <tytso@MIT.EDU> writes:
> Adding journalling isn't so simple. One, you need ordered writes, which
> is something the low-level block device interface doesn't provide.
> Secondly, you have to worry about performance, which is hard to do while
> still having the ordering semantics.
How important do you consider a good journaling fs for linux? It
seems to me like journalling would be most important for something
like a huge ftp site.... but ftp.freebsd.org is now 106GB of disk
space. It takes me at least 1 minute to fsck a 1.5 GB partition.
That means to fsck this disk farm at ftp.freebsd.org after a crash,
it's going to take around an _hour_, which seems unacceptable by any
standard.
I don't think Ted mentioned once that he did not think it was
important. In fact journaling support for ext2 is in fact in the
works, but the guts of the I/O and vfs subsystems must be revamped to
have the necessary facilities such an extention would need. This is
what is being worked on currently.
sunsite.ms.mff.cuni.cz has over 110GB of storage and more will be
added, it is a moderately fast SparcLinux box, and it is estimated
that it would take not one but several hours to fsck everything.
Believe me, it is known that journaling or something similar is
needed.
But somehow they get away with this. Maybe they mount most of the
disks read-only? Perhaps I'll write and ask.
You will get an answer probably like "FreeBSD is so stable it never
crashes" or that they do in fact use the journaling filesystems
written for the BSD vfs layer, the latter I highly doubt as it is slow
as shit...
---------------------------------------------////
Yow! 11.26 MB/s remote host TCP bandwidth & ////
199 usec remote TCP latency over 100Mb/s ////
ethernet. Beat that! ////
-----------------------------------------////__________ o
David S. Miller, davem@caip.rutgers.edu /_____________/ / // /_/ ><