On Tue, 2005-10-04 at 19:51 +0200, Arjan van de Ven wrote:
Think of it this way: if you have half your disk empty, the
filesystem
can do a proper job of finding non-fragmented space.
If only 0.0001% is free, it has almost no freedom of choice, resulting
in "you get it in whatever order some things become free".
Those are sort of extremes; there's been a bunch of research and the
outcome was that 5% free seems to be sort of the turning point in this
respect.
I suspect that research predates the Tb sized volumes, so I don't know
if it maybe is 1% on such volumes, but then again to some extend the
freedom needed will scale with the FS size
Obviously I've not done real research on the subject, but wouldn't you
expect it to scale according to the write pattern when you're under low
space pressure?
Or, to put it differently, if I've got a 100MB disk with 5MB free and
I'm trying to write a 4kB file, the fs has a similar degree of freedom
to the case of a 100GB disk with 5GB free where I'm trying to do a 4MB
write, doesn't it? Whereas obviously it has much less freedom on the
smaller disk with a 4MB write.
It seems like the necessary free space percentage varies according to
your data and IO pattern rather than according to the FS size.
(Granted, the requirements of people who have terabytes of storage
probably dictate larger files, larger IOs, and lower relative latencies
than that of people with mere tens of gigs, so I would expect some
*correlation* to filesystem size.)
--
Peter