[Techtalk] /usr Partition

[Val Henson]

On 1/19/06, Akkana Peck <akkana at shallowsky.com> wrote:
>
> But it looks like things have gotten better and sharing /home
> works a bit better now.

Yeah, that's my experience as well.  It used to be a huge pain; now
KDE and GNOME and whatever else seem to be able to coexist peacefully.

> Val, I've been wanting to ask you: I remember there used to be
> arguments for not making filesystems over a certain maximum size,
> so if you had a big disk, it was better to break it up into several
> smaller filesystems.

I am chastened to admit I don't know.  My impression is that the main
reason things were split up is because disks were darn small, and your
average UNIX server had around 6 of them.  The only other reason I can
think of (totally made up off the top of my head) is that file systems
were so delicate that you needed to repair them by hand a lot, and
having smaller file systems made that easier.  File systems used to be
incredibly fragile and repair by hand was pretty darn common.  I have
friends who used to have certain inode numbers and locations
memorized, because they had repaired them so often.

> In these days of 160G+ disks, is it kosher to have a single 100M
> filesystem?  On ext2/3 it can be a major pain because fsck takes
> forever, but what about reiser?

I'm guessing you mean one big partition?  Yeah, fsck is the main
reason, although you'll probably have to fsck everything mounted at
the time you crashed.  ext3 isn't a problem; it should just replay the
log.  Same for reiser, except that reiser is not ever really stable
and is far more complex... so I guess I'd recommend fscking reiser
file systems, uhm, every time you crash, and every 20 mounts.

> With ext3 (which I know you like
> better than reiser), how important is it to fsck often?

I honestly don't think it's important unless you see some odd
behavior, in which case you should fsck immediately.  ext3 is very
simple and stupid.  I do recommend fscking if, for example - just
TOTALLY making this up, I never ever did this nope nope nope not even
once - you give a talk, close your laptop (which does not suspend
automatically), put it inside an insulated closed bag, and let it run
out of power - sideways - while you eat lunch.  Then you should run
fsck.  Or if you're  untarring some giant thing or otherwise doing a
hell of a lot of fs writes when you suddenly kick the power cable out
of the outlet.  What I don't think you should be concerned about is a
kernel crash or software forced reboot; as long as the disk keeps
power ext3's logging stuff is likely to work like a charm.

> What about swap? What's the current guideline for how much swap you
> should have?

Last I heard the recommendation was that swap should be twice memory. 
Keep in mind that swap is basically a place to put all the program
memory that you never touch.  Swap is not useful as a place to put
memory you actually need.  You need swap mainly because the VM cannot
handle memory pressure gracefully otherwise and current kernels are
basically unusable without it.  In most cases, you could probably get
away with 512MB of swap, regardless of memory size.  I currently have
12MB of swap used on a machine with 256MB of memory.  Check yours with
"vmstat -S M" ("-S M" makes it report in MB rather than KB).

-VAL