Re: journaling filesystem
Victor Yodaiken (yodaiken@chelm.cs.nmt.edu)
Sun, 15 Jun 1997 14:52:21 -0600
On Jun 15, 9:30pm, Michael Neuffer wrote:
Subject: Re: journaling filesystem
>On Sun, 15 Jun 1997, geerten kuiper wrote:
>> Tandem Computers, known for their high availability, do not have a
>> journaling fs. Audit trail is generated at a higher level (where files
>> consist of data rather than just a stream of bytes), is handled mostly by
>> user space processes, and is stored in normal files.
...
>The need for a JFS goes down dramatically if you can't bring down the
>machine, but we are talking here about machines that you don't even think
>about below $100k (for the smallest possible configuration of course).
IMHO, journaling FS's are perhaps not the right answer for
Linux. A unix FS can be made fault-tolerant by simply introducing
an inode lookup table in the FS. All changes can be done on
free blocks and committed only whenthe FS is in a consistent
state. There is an old paper by Sam Glazer and David Arnow that
describes this scheme in some detail, but it's really quite simple ---
much simpler than a full journaling FS and I think less dependent
on some of the dubious assumptions that are made in journal FS
designs.