File system trashing problems

[Frank Wortner]

shollen%odin.Wright.EDU at relay.cs.net (Sheila Hollenbaugh) writes:
>In v5n28, Keith F. Lynch describes file system trashing problems on a
>SUN-3/260, and quotes his Sun rep. as saying that this situation is
>"normal" when the disk partition is more than 90% full and when processes
>try to use too much memory.

"Normal?!?"  I sure hope not!  Let's hope this was just a case of
misprounciation.  "Thrashing" can occur if the filesystem gets too full.
The worst symptom in this case should be unaccepably slow performance as
the disk drive seeks wildly in an attempt to use what few free bocks
exist.  BSD "fast" filesystem implementations usually reserve the last 10%
of available space just to insure that this does not happen.  "Trashing"
should *never* occur under any circumstances---in theory. ;-)

[[ Understand what Frank means by "reserve the last 10%".  The percentage
capacity figure in the output of a "df" hides the last 10% of the disk.
When that column says "100%" full, the disk is really only 90% full.  The
kernel also prevents non-root users from writing to the disk when it is
that full.  This is why you will sometimes see disks at "106%" full.  That
means that root continued to fill it up after it reached "100%".  Although
I feel that the motivation was wrong (they did it because the so-called
"fast file system" thrashed above 90% capacity), the idea of having a
small percentage reserved for root-only access is nice.  --wnl ]]

	Frank