I/O Usage by OSF/1 Processes

From: Peter Hudson <peter_at_osm.co.uk>
Date: Mon, 5 Feb 96 15:55:14 GMT

OSF/DU Managers,

I am submitting this request on behalf of a colleague who does not (yet)
subscribe to this list. Please reply either to myself or directly to him,
marc_at_osm.co.uk. Many thanks. The request reads:

I have a question about the validity of data in the OSF/1, sorry Digital UNIX,
process accounting file (usually /var/adm/pacct). The ac_io value (see
/usr/include/sys/acct.h) seems to have a number of ranges.

The effect seems to show itself for most commands, i.e. processes. A way to
show it is to use

        acctcom -i -n pwd

(or "ls" or "mv" etc. instead of "pwd") and look at the "CHARS TRNSFRD"
column. On both machines I have tried this on, the value is either ~330k
or ~17.3M. On one, there was also a set of ~22M.

The values seem to depend, at least partially, on the display environment.
Running a command directly within an xterm usually results in a large value,
whereas a command run via, say, a telnet session, usually gives a small value.
Of course, much more processing and library code is involved with an xterm,
but even so the values seem a bit extreme.

Are the values being corrupted? They are stored in the compressed "comp_t"
format, but this seems to work for all other flavours of UNIX I have
encountered. The effect shows itself on OSF/1 3.0 and 3.2, and presumably all
other 3.*.

Can anyone shed any light on this?

                                                                        Marc

-----

Marc Houliston marc_at_osm.co.uk Open Systems Management Ltd

Tel: +44 1734 770333 Fax: +44 1734 770888
Received on Mon Feb 05 1996 - 19:36:05 NZDT

This archive was generated by hypermail 2.4.0 : Wed Nov 08 2023 - 11:53:46 NZDT