On Thu, 2002-11-07 at 09:17, George Georgalis wrote:
> I do have a few questions though. I'm not sure if blade's disks used
> xfs-1.0 or xfs-1.1; and xfs-1.2 seems eminent (I saw some xfs-1.2preX
> patches). Does the filesystem remain the same through these version
> changes or should I think about doing a mkfs.xfs each time I upgrade
> the the xfs version?
XFS on-disk format is in general not changed (remember, XFS has been
around on IRIX for some years now...) and when minor changes are made,
there is compatibility.
I'll leave some of your questions for others to answer, but...
> I've seen a few
> mentions of the Linux xfs realtime subvolume but no doc, is it ready for
> production? From what I can tell, it's a non journeled contiguous data
The realtime volume is not supported, although it is basically
functional. You need to do an ioctl to the file to mark it realtime
after it's created (but before any data is written to it) and then do
Direct I/O to the file from then on. The main difference is a more
deterministic allocator that should allocate bigger chunks at a time.
"realtime" is perhaps a bit of a misnomer. Oh, and it is journalled
just as the main data volume is.
Maybe I should just use ext2 for the media files? Would that be
> higher performance? I'm not too worried about fsck, because in the case
> of corruption I can remake the filesystem (data partition) and renew the
> data from the node server.
You'll probably just need to test in your environment, and see what
works best. Different filesystems excel at different things...
Eric Sandeen XFS for Linux http://oss.sgi.com/projects/xfs
sandeen@xxxxxxx SGI, Inc. 651-683-3102