Missing disk space

[Steve Mawer]

I have a situation on RT 6150 AIX V2.2.1 in which tracts of disk space go
missing, and I'm completely baffled as to where it's going.

I have a few (well lots of, actually,) user programs which connect via
named pipes to a central server process.  This central process connects
via another named pipe to a process which talks via a TCP/IP socket to its
equivalent on another machine.  After the connections have been running
for a while, there appear to be many blocks of filespace which cannot be
accounted for.

Everything happens in a /u partition mounted on /dev/hd1.  On startup,
`du /u' reports 31660 blocks used, while `df' reports 35260 free out of
67060, ie 31800 blocks used, a discrepancy of 140.  I don't expect the
two numbers to be *exactly* the same, but at least they're close :-).

After connecting 36 user processes and leaving it running for two days,
`du /u' reports 38808 blocks used, and `df' reports 23848 free, therefore
43212 blocks used.  This is a discrepancy of 4404 blocks.  I first
suspected the named pipes, and found that although they typically have
a size of 0 and therefore show no blocks to `du', the inode has up to
eight 2K blocks allocated.  If I shut down 1 user connection, the free
space increases by the number of blocks allocated to that named pipe.

However, if I add all the blocks in all the named pipes, I only get 344.
The only way I can get back the rest is by stopping all my processes.

Can anyone give me some pointers as to where the disk blocks are being
used, as I'm trying to support a site where disk space is tight already,
and restarting the system to recover space is unacceptable.

-- 
Steve C. Mawer        <scm at datlog.co.uk> or < {backbone}!ukc!datlog!scm >
                       Voice:  +44 1 863 0383 (x2153)