cvs commit: src/sys/kern kern_shutdown.c

[Scott Long]

Peter Jeremy wrote:
> On Wed, 2004-Jul-21 15:57:30 -0600, Scott Long wrote:
> 
>>Implementing a journalling filesystem would be a much more beneficial
>>use of time here.
> 
> 
> You still wind up with unwritten data in RAM, just less of it.
> 
> How much effort would be required to add journalling to UFS or UFS2?
> How big a gain does journalling give you over soft-updates?
> 

That's a very good question.  A group at RPI has been working on it for
some time, but I'm not sure how close they are to having it done.  If
you look in the commercial world, Apple, Sun, and Wasabi/NetBSD have all
done it successfully (Wasabi's isn't open source, btw).  My guess is
that it would take about 4-5 months to get it going, and then at least
8-12 months to ensure that there are no bugs and to tune performance.
Certainly not impossible, but not something that would be production
quality on short notice.  I think that you would also have to make it
othogonal to softupdates.

The gain that you get is that your filesystem recovery time drops
tremendously.   You also have a much better chance of all of the
metadata being on the disk and recoverable.  Furthermore, it opens the
door for data+metadata journalling for even more protection (at a large
cost to speed and/or buffer-cache pressure, of course).

Scott