Changes between Version 12 and Version 13 of IPP_Progress_Report_20130204

Timestamp:

Feb 11, 2013, 11:22:19 AM (13 years ago)

Author:

Serge CHASTEL

Comment:

--

IPP_Progress_Report_20130204

@@ -15 +15 @@
    * Worked on an IPP framework installation dedicated for MOPS on the NEO cluster.
      * Identified non-robust middleware between pantasks_server and pcontrol at the protocol level. My pantasks_server was linked against a more recent version of readline. This version was sending an extra string to pcontrol which was not complaining or showing any error (but wasn't doing anything). Relinking against the right version of the library fixed the problem but pcontrol should be in my opinion less tolerant to that kind of problem.
-     * Nasty bug in pswarp. It fails when the number of threads is set to 1 but succeeds when it’s set to 4.
-     * Nasty bug in ppImage. It converges after a really long-time when the number of threads is set to 4 but succeeds when it’s set to 1.
+     * Nasty bug in pswarp. It sometimes fails when the number of threads is set to 1 but succeeds when it’s set to 4.
+     * Nasty bug in ppImage. It sometimes converges after a really long-time when the number of threads is set to 4 but succeeds when it’s set to 1.
      * I had to modify the {{{prepare_output}}} function in {{{chip_imfile.pl}}} deletes files that are produced at the beginning of the chip stage. The steps (I simplify a bit since there are also steps involving data control and database operations) in the chip stage are the following: 1) "Prepare output"; 2) Detrend raw files; 3) Do photometry. It is supposed to be possible to control the execution or non-execution from the recipe. In the "IPP for MOPS" context we are only interested in the third step. Unfortunately, the "Prepare output" behavior is not controlled by a recipe: files deletion occurs systematically... but those produced by step 2) and necessary for step 3) are deleted. To fix the problem, I had to modify the original {{{chip_imfile.pl}}}.
      * Using the same notations as in the previous point, {{{ppImage}}} hard-codes the 2) and 3) steps. No way to insert another step (like for instance the MOPS synthetics generation). Facade and Command Design Patterns would have helped a lot. I can tell more about this.