Re: HUGE size of squid binary after compile

Henrik Nordstrom wrote:
>
> Yung-Zen Lai wrote:
> >
> > BUT. when I use the command "df" to check my disk.
> > I found that the capacity in my filesystem is only 86%, and i-node is 7%
> > Would you please tell me how to find errors in it?
>
> Your systems messages file may provide additional information on this.
> See /var/adm/messages or /var/log/messages.
>
> There are a couple of known causes to this:
>
> * You are using Solaris, or another system where df reports free block
> fragments and not free blocks.
>...

[Not relevant to the original query about SCO, but maybe of relevance to
people using Solaris 2.6 who haven't kept up with Sun's recommended
patches.]

Are you sure about that ...? I've never seen Solaris 2 report fragments (the
only thing I recall reporting fragments is fsck...), though by default space
is reported in 512-byte blocks rather than KB (seems to be a SysV versus BSD
distinction) - "df -k" will get KB. The free space percentages should be OK
though, giving percent of non-reserved space free.

What I *have* seen and been inconvenienced by is that with Solaris 2.6 as
originally shipped (sorted out by a later reecommended patch) it would get
totally confused about reserved space and reserve virtually none (in terms
of how it handled space allocation, the reserve *was* defined sensibly) so
that when if it said "device full" there might be no space (or just a very
few blocks) free, no space for root to borrow while sorting out the problem.

The real nuisance was that I did an installation, saw I'd got plenty of
free space in the system partitions, i.e. I'd judged the sizes about right,
and got on with setting up the system. Then discovered later (to late to
reepartition and reinstall easily) that the free space was seriously wrong
and at least one partition was overfull, eating into the root reserve, when
the patch fixed things. The other annoyance was that the patch description
talked about incorrect output from df for unmounted filesystems. That was
misleading, presumably the symptom someone originally reported (but the
patch fixed the real problem) - the filesystem was returning misleading
figures to anything that looked at them.

One other point - though not mentioned in the man page for newfs, Solaris
2.6 newfs will adjust the default reserved space so that it no longer wastes
10% unnecessarily on large partitions (apparently it's the actual amount of
reserved space rather than the proportion of total blocks that matters for
performance), and it seems happy to reduce the reserve at least as low as
1%. [The bug described above had the effect of reducing it to about 0.001%!]

                                John Line

-- 
University of Cambridge WWW manager account (usually John Line)
Send general WWW-related enquiries to webmaster@ucs.cam.ac.uk