(very OT) Ideal partition schemes (history of partitioning)

Steve O'Hara-Smith steve at sohara.org
Sat Aug 29 06:21:50 UTC 2020

On Fri, 28 Aug 2020 23:08:30 -0400
Aryeh Friedman <aryeh.friedman at gmail.com> wrote:

> Also why are partitioned need at all? (both currently and historically)

	For a historical perspective I once saw a home made XENIX box with
four 8" floppy drives labelled /, /etc, /usr and /tmp. Back when drives
were small it was desirable to have more storage in the filesystem than
available on a drive, being able to combine multiple filesystems into a
single tree by mounting made it possible to do that seamlessly. Compare and
contrast with drive letters as in CP/M, MS/PC-DOS and even still lurking in
the latest Windows.

	A side benefit was that it made it possible to have different
options (read only, block sizes, ...) on different mounts and so tune
different parts of the tree for the different usage (for an extreme
example /tmp is sometimes a memory based file system these days). It was
also an enabler for seamless network filesystems from MICNET (which ran
over UUCP) to NFS by way of AFS.

	Today of course a 10TB drive is a consumer item and for most
purposes there is no need to split up the tree into multiple filesystems
just to have enough space. One consideration though is to prevent a runaway
writer from chewing up all the available disc space, having separated
filesystems limits the scope to one section of the tree (which can still be
a real PITA). Quotas is a more recent solution to the same problem, as is
the spaced reserved for root processes in UFS mounts.

	Personally I tend to use ZFS for almost everything and mount /home
(and a few other odds and ends) from my NAS which also runs a lot of jails.
This results in a lot of filesystems (especially on the NAS) coming from a
small number of pools. The way ZFS uses so many filesystems is mostly useful
for the ability to have different filesystem level settings at the mount

	For a contrasting approach there is a commercial distributed
filesystem (OneFS) where metadata of that kind (including replication
policy) is settable by directory. I won't say too much about it because it
pays my wages.

	As for disk partitioning that's just a way to pretend that one
big disk is a set of smaller ones, which is of course useful for many

Steve O'Hara-Smith <steve at sohara.org>

More information about the freebsd-questions mailing list