Документ взят из кэша поисковой машины. Адрес оригинального документа : http://www.atnf.csiro.au/people/pulsar/pmsurv/docs/epping_processing.txt Дата изменения: Fri Oct 4 11:55:20 2002 Дата индексирования: Wed Dec 26 01:37:47 2007 Кодировка: Поисковые слова: puppis

PROCESSING AT EPPING: pmsurv/docs/epping_processing.txt
Froney Crawford - July 23, 1998
FC/RNM - July 31, 1998
FC - Aug 7, 1998
RNM - Sept 21, 1998
RNM - Jan 11, 1999

0) Login as pulsar in all places.
Set up windows: 1 for tape reading on attila (tape window)
1 for the master on attila (master window)
1 for pmproc on attila (pmproc window)

1) in tape window (attila):
cd to bdhunt directory.
eliminate old stuff if it is there:
mkdir PM00YY (if bdhunt is filled with junk)
mv PM00YY_* PM00YY/.
mv digitizer.offsets PM00YY/.
mv PM00YY.pmbs PM00YY/.
(keep everything else)
mt rewi (make sure TAPE env var is set to /dev/rmt/0n on attila)
get digitizer.offsets:
sc_td -D -c 3.5 -n 1 -A PM00XX
when done, copy relevant line to "digitizer.offsets"
mt rewi

2) In pmproc window (attila):
type "pmproc" pmproc> n PM00XX (this prepares for new tape PM00XX)
In tape window (attila):
pmtape b
(make sure it starts reading data)
In master window (attila):
type ps gux and make sure no other pmchief_hunter process is
going -- if it is, kill it.
type pmchief_hunter (should start printing lines every
few secs)
In pmproc window (attila):
b SLAVENAME /full_path/name/SLAVENAME_slave/
(repeat for other slaves -- recommend 4 birdslaves
to keep up with the tape read)
Currently available directories are:
/psr5/pm/attila0/
/psr5/pm/attila1/
/psr5/pm/attila2/
/psr5/pm/attila3/
/psr9/pulsar/pm/raptor0/
/psr9/pulsar/pm/raptor1/
/psr9/pulsar/pm/raptor2/
/psr9/pulsar/pm/raptor3/
/psr9/pulsar/pm/raptor4/
/DATA/RAPTOR_3/pulsar/pm/raptor5/
/DATA/RAPTOR_3/pulsar/pm/raptor6/
/DATA/RAPTOR_6/pulsar/pm/raptor7/
/DATA/RAPTOR_11/pulsar/pm/raptor8/
/DATA/RAPTOR_11/pulsar/pm/raptor9/
/DATA/PUPPIS_2/pulsar/pm/puppis0/
/psr6/pm/tycho0/

Use pmproc to monitor processes, takes about 1 min/beam to process.

3) When birdie hunting is finished:
In pmproc window:
pmproc> t (stops tape script)
In tape window:
mt rewi (after tape script has stopped)
In pmchief_hunter window:
should exit automatically after birdie summary is done
If birdie_summary doesn't run do: %pmbird PM00XX 1 n
pmbirdslaves:
you can keep them going, if you need to stop them,
type s attila or s raptor in the pmproc window.

4) In pmproc window:
pmproc> n PM00XX (sets up new tape again for processing)
In tape window:
copy digitizer.offsets from bdhunt dir to data_pool dir.
pmtape d (make sure script starts reading data)
In master window:
make sure no other pmmaster is going (ps -fA | grep pulsar)
If it is, kill it.
pmmaster (starts printing a line every few secs)
In pmproc window:
l SLAVENAME /full_path_name/SLAVENAME_slave/
see birdslave examples for format...
each slave takes 2 GB.
Monitor processes with pmproc.
(Each beam takes about 5 hr to process.)

5) All files pertaining to a particular pointing should be eliminated
from the data_pool automatically when the pointing is done being
processed. If there is a problem with one of the beams, it will not
automatically eliminate it. It will keep the pointing around for
manual perusal. If you want it gone, edit "complete_pts.txt" in the
conductor dir and add a line with the name of the pointing you want
eliminated (e.g., PM0039_010). This should do the trick.

[NOTE: If the number of mis-processed beams becomes large enough that
this is a nuisance, perhaps the script should be changed to erase the
data beam by beam -- i.e., keep only the beams around which are bad,
not the whole pointing.]

To recover if things get in a mess:

At the pmproc prompt, exit pmtape and pmmaster.

cd to conductor dir
rm * in conductor
touch bigsister.data
touch processing_status.log
chmod 664 processing_status.log

cd to data_pool
rm PM*
rm ex* (leave digitiser.offsets)

Normally you will want to rewind the input tape
mt rewind
and (later) edit valid_pointings.txt to restart at the correct place (see below).

cd to $PM_BOOK on each processor being used and kill slave processes:
kill_slaves
On attila also do
clean_slaves
to clear all scratch directories.

Check that no pm jobs still running:
ps -fA | grep pulsar

After 5 min, check that the bigsister cron jobs are running on each processor to
be used. In the conductor directory, look for files bs_out_proc.log
ll bs*

If these don't appear, on the processor concerned
crontab -r to remove
crontab /psr/bin/sun4sol/pmcron (or crontab /psr/bin/sgi/pmcron)
Then after 5 min, check that the bs files appear.

If birdie hunting needs to be repeated:
cd to bdhunt
rm PM0039_* in bdhunt
rm birdie_files, exit_status_sc_td, candidates, checkbmtmp, checktmp
in bdhunt
cd bdhunt/PM0039
cp * ../.

cd to book dir
either rm or mv to some other place the pointing dirs you want to repeat

Start pmproc and initialise tape
>n PM0039

Start the tape reading
pmtape d (or b)
If you want to begin part way down the tape, you have one minute to edit
$PM_HOME/conductor/valid_pointings.txt to remove the files you don't want
to repeat before sc_td starts.

Start pmmaster

Launch slaves in pmproc. Check that the slaves go into WAIT state and then
TREE state when data are available.