Shrinking Filesystems... is it important?
dcm at plato.austin.ibm.com
dcm at plato.austin.ibm.com
Thu Jun 13 00:02:33 AEST 1991
In article <1991Jun12.060212.8037 at ux1.cso.uiuc.edu> shair at ux1.cso.uiuc.edu (Bob Shair) writes:
>While we're talking about dynamically shrinking filesystems, and whether
>AIX V3 might someday provide that facility, let me ask a purely
>hypothetical question... What's it worth?
It's worth alot. What happens if you accidentally extend /usr
too far, using up your entire disk, then /tmp fills up? How do
you take those extra blocks from /usr and use them to extend /tmp?
Right now you backup /usr, reboot off your maintenance diskettes,
shrink /usr's logical volume, mkfs /usr, restore, then extend /tmp.
Big Waste of Time(tm).
Dynamically extendible filesystems is a tremendous win, however
it loses some of its usefulness unless you can reclaim blocks
and use them to extend other filesystems.
>If we want dynamically shrinkable filesystems, surely the developers
>will add something like additional backwards chains... I don't know
>what (I've no connection with development). Whatever it is, it will
>take space on disk, take additional instructions to process, update
>or skip over, and end up slowing disk performance to some extent.
I doubt that it will affect the structure of the filesystem. Back
in the early days of AIXV3, I was actually working on an algorithm
for shrink fs. It *was* working for the easy cases, and almost
working for harder ones. Then they turn around and change the data
structures... Sigh. I never had a chance to go back and rewrite it.
It's not that hard. The hardest part is error recovery (what happens
if the system crashes in the middle of the operation?). I don't think
it requires any changes in the filesystem itself.
>We've gotten along in Unix for some time now without this feature.
True. However, we need shrink to make extend more usable.
Craig
--
Craig Miller Internet: dcm at aixwiz.austin.ibm.com
IBM Austin Vnet: tkg007 at ausvmq
AIXV3 Change Team (level3) IBM internal: dcm at littleguy.austin.ibm.com
"Another satisfied customer..."
More information about the Comp.unix.aix
mailing list