Posted Nov 1, 2012 18:43 UTC (Thu) by ssmith32 (subscriber, #72404)
In reply to: Fedora and LVM by nix
Parent article: Fedora and LVM
I've heard this before.. but never quite understood the logic.
/home will always be on a partition with a given probability of failure.
Whether or not /var is on a separate partition does not seem like it would much affect probability of the fs metadata for /home getting corrupted.
I suppose /var may get written to more frequently (and thus increasing the chance of messing up the fs metadata), but on a desktop system, I would think the difference in the rate between which /home and /var is written is not that great.
And I haven't ever had trouble upgrading a system with a single partition (another common reason given).. installing a brand-new OS, yes.. upgrading no.
Posted Nov 1, 2012 22:31 UTC (Thu) by nix (subscriber, #2304)
[Link]
/var is written to very frequently: in particular, because wtmp is on /var, it is written to right before major changes of system state when the system is particularly likely to go wrong (suspend/resume, login, logout, reboot, shutdown, any event that triggers a log write). Thus it is likely dirty at such times, which makes it particularly prone to corruption.
Writes to /home are relatively rare by comparison (even for me, with subscriptions to dozens of mailing lists all going into nnml spools under /home).
Fedora and LVM
Posted Nov 2, 2012 0:25 UTC (Fri) by ssmith32 (subscriber, #72404)
[Link]
That does make sense :)
Much better than the hand-waving I've heard before.. ;)
Fedora and LVM
Posted Nov 2, 2012 17:31 UTC (Fri) by nix (subscriber, #2304)
[Link]
Oh it started as handwaving -- i.e. I didn't do more than simple blktrace benchmarking for a few minutes to look at the relative write rates after the initial mkfs -- but it became clear in the recent ext4 corruption fiasco that the handwaving had a basis in fact :) I corrupted /var repeatedly, and /home once or twice at most.