Next: Limitations of the data reduction script
Up: Interpretation of results
Previous: Structure of the output file
Determining CPU time and elapsed time usage under Unix
is relatively straightforward.
However there is no tidy and portable way of
assessing memory usage and the solution used here is
a little bit messy.
For details of the method used and a better insight into
the problems which can occur, see the inline comments in
the script reslog.
Broadly speaking however, the following comments apply:
-
The logging script will not work well for assessing applications
with a short run time or small memory footprint.
Since the purpose of these scripts is to assess resource usage
for large data reduction jobs we do not consider this a serious problem.
-
The logging script relies heavily on the details of output from
the ps command, amongst others; therefore it is
likely to be sensitive to changes in operating system and environment.
-
When very large jobs are running,
the system can behave badly,
response can be very slow,
and processes can fail unexpectedly, even ones which
do not themselves have extreme requirements.
For all the above reasons, CCDBIG is not a foolproof application,
so its results should be regarded critically;
if there are anomalous results they should be checked further
(a monitoring tool like top may be of use here)
before being believed!
Next: Limitations of the data reduction script
Up: Interpretation of results
Previous: Structure of the output file
CCDBIG: assessing CCDPACK resource usage for large data sets
Starlink System Note 69
Mark Taylor
6 July 1998
E-mail:ussc@star.rl.ac.uk