newfs -m options

[Henry Spencer]

>We have several 1 gb disk drives and it does not make a whole lot of sense
>to reserve 100 mb (10%) for space threshold...

Unfortunately it does continue to make sense.  It's not a matter of having
some reserve for emergencies (in which case reducing the reserve on a big
disk would indeed make sense).  The issue is that filesystems which try to
keep the blocks of a file together, a la 4.2BSD and successors, lose
performance badly if they get quite full.  To have a reasonable chance of
finding empty blocks near the right place for a new file, the filesystem
has to have a certain "density" of empty blocks, be the filesystem one meg
or a thousand.  The constant percentage is indeed the right approach.
It's part of the price paid for performance.

Now, there are some caveats to this.  First, you have to believe that
keeping the blocks together is a win.  It definitely is on a single-user
machine; how much it helps on a timesharing system or a server is a very
good question that (as far as I know) has never been systematically looked
at.  (All those impressive 4.2BSD filesystem benchmarks were run
single-user, and multi-user comparisons seldom make any attempt to
separate the effects of better block placement from the obvious win of
bigger blocks.)  Second, you have to be using the filesystem to hold lots
of little files that come and go, rather than a few monsters that just sit
there.  And third, whether 10% is the right percentage is open to debate,
and is perhaps somewhat site-specific.  I believe people have cut it to 5%
with no major ill effects.

                                     Henry Spencer at U of Toronto Zoology
                                 uunet!attcan!utzoo!henry henry at zoo.toronto.edu