DKRZ projects: Issueshttps://swprojects.dkrz.de/redmine/https://swprojects.dkrz.de/redmine/redmine/favicon.ico?17095821032014-08-06T16:29:45ZDKRZ projects
Redmine YAXT - Feature #333 (Resolved): Caching of communicatorshttps://swprojects.dkrz.de/redmine/issues/3332014-08-06T16:29:45ZThomas Jahnsjahns@dkrz.de
<p>YAXT currently creates communicators internally to provide isolation from other parts of the system. This can be potentially costly when it introduces additional synchronization. For this reason it seems sensible to cache previously created communicators instead of destroying them immediately.</p>
<p>An alternative scheme requires managing tags in the library more closely and is potentially less resource intensive (depending on how costly a communicator is).</p> ScalES-PPM - Bug #297 (Assigned): Fix i4 assumptions in multiple routineshttps://swprojects.dkrz.de/redmine/issues/2972012-01-02T16:02:29ZThomas Jahnsjahns@dkrz.de
<p>In the Fortran 90 part of the library, to always suit default kind INTEGER arguments, many generics only have specific implementations with default INTEGER arguments, because an i8 implementation wasn't needed so far. Unfortunately for bridging the gap to C several of these routines also assume default kind equals i4. Thus it would be preferable to have implementations for both i4 and i8 resulting in generics which work on any setting of default integer kind.</p> ScalES-PPM - Feature #296 (Assigned): Multi-threaded repartitioning routinehttps://swprojects.dkrz.de/redmine/issues/2962012-01-02T13:20:29ZThomas Jahnsjahns@dkrz.de
<p>The multi-process variants of the repartition-by-swapping routines are a blue-print of how to construct comparable multi-threaded variants.</p>
<p>Additionally in a hybrid environment, it might also help to have a multi-threaded, multi-process variant.</p> ScalES-PPM - Task #295 (New): Apply optimizations to repartitioning routineshttps://swprojects.dkrz.de/redmine/issues/2952012-01-02T13:17:43ZThomas Jahnsjahns@dkrz.de
<p>The repartition-by-swapping routines can benefit from the following improvements:</p>
<ul>
<li>repartition in parallel by pairing up lowest weight with highest weight part, second-lowest weight with second-highest weight part etc.</li>
<li>sort the larger part weights (size n) to achieve O(log n) lookup</li>
</ul>
<p>This issue is mostly for documenting the progress of an on-going implementation: the above is done for the variant for multi-process and 4 byte integer weights (mp_i4) the following are required for completion:</p>
<ul>
<li>mp_dp</li>
<li>mt_i4</li>
<li>mt_dp</li>
</ul>
<p>Where the multi-threaded variants have yet to be written.</p> ScalES-PPM - Feature #294 (New): Add OpenMP multithreaded index construction routinehttps://swprojects.dkrz.de/redmine/issues/2942012-01-02T13:11:31ZThomas Jahnsjahns@dkrz.de
<p>Index construction can take huge amounts of time where the mask is large and not layed out in unit-step.</p> ScalES-PPM - Bug #286 (New): Fix test for FPU precision correction codehttps://swprojects.dkrz.de/redmine/issues/2862011-09-19T15:48:49ZThomas Jahnsjahns@dkrz.de
<p>Currently the test wether FPU precision correction is always invoked even if, because of SSE or otherwise correct handling of 32/64bit quantities in the FPU, no correction is necessary.</p>
<p>configure.ac should be improved with a test for wether correction is necessary.</p> ScalES-PPM - Documentation #283 (New): Account for preprocessor directiveshttps://swprojects.dkrz.de/redmine/issues/2832011-09-19T11:16:06ZThomas Jahnsjahns@dkrz.de
<p>doxygen is currently unaware of preprocessor defines in Fortran code. configure should substitute this in the doxygen configuration file.</p> ScalES-PPM - Documentation #282 (New): Add introductory page to unit-documentationhttps://swprojects.dkrz.de/redmine/issues/2822011-09-19T11:14:43ZThomas Jahnsjahns@dkrz.de
<p>The introductory page generated by doxygen is currently empty. It should contain or link to some of the documentation already present on this web-site or in the README files.</p> Quick Enqueue - Feature #276 (Assigned): Option argument --wall-time-limit should parse stringhttps://swprojects.dkrz.de/redmine/issues/2762011-06-03T16:20:50ZThomas Jahnsjahns@dkrz.de
<p>The time string should conform to the queuing system requirements.</p> Quick Enqueue - Feature #269 (Assigned): Handle llsubmit error/status output.https://swprojects.dkrz.de/redmine/issues/2692011-06-01T16:29:41ZThomas Jahnsjahns@dkrz.de
<p>Currently error and status are not handled.</p>
<ol>
<li>Error output should be presented to the user with some harmless comments stripped.</li>
<li>In case the job submission takes longer than some adjustable timeout, progress should be reported to the console in some way.</li>
</ol>
<p>In case 2. it would probably be wise to always write to /dev/tty.</p> Quick Enqueue - Feature #267 (Assigned): Adjust query frequency.https://swprojects.dkrz.de/redmine/issues/2672011-06-01T16:23:16ZThomas Jahnsjahns@dkrz.de
<p>The program does currently only query output appended to the log files once per second. This should be adjusted to the inverse of the recently occurred frequency of updates to the files.</p>
<p>I.e. if the file saw 5 lines added in the last second the the sleep time should be adjusted to 0.2 seconds.</p> ScalES-PPM - Task #260 (Resolved): Investigate wether the library contains symbols not conforming...https://swprojects.dkrz.de/redmine/issues/2602011-03-24T16:24:45ZThomas Jahnsjahns@dkrz.de
<p>The result is yes, and it will need some work to fix, mostly in the hands of Thomas and Florian.</p> ScalES-PPM - Task #258 (Assigned): Produce release-ready version of ScalES-PPMhttps://swprojects.dkrz.de/redmine/issues/2582011-03-24T16:22:41ZThomas Jahnsjahns@dkrz.de
<p>To distribute a version of ScalES-PPM instead of only giving repository access the following needs to be done:</p>
<ul>
<li>decide on first public version number, a doodle has been setup <a class="external" href="https://www.dkrz.de/redmine/doodles/1">https://www.dkrz.de/redmine/doodles/1</a></li>
<li>investigate wether the library contains symbols not conforming to the PPM-prefix requirement</li>
<li>check wether make dist-check works</li>
<li>deposit a .tar.gz on web-site</li>
</ul>
<p>This will be then made available in an upcoming MPIOM version.</p> ScalES-PPM - Feature #220 (Assigned): Add distributed summation routineshttps://swprojects.dkrz.de/redmine/issues/2202010-12-08T10:52:31ZThomas Jahnsjahns@dkrz.de
<p>Parallel summation of floating-point values is not associative and thus gives different results depending on the execution sequence which changes with MPI implementation and decomposition of the summed array.</p>
<p>Therefore a method to give identical results for any decomposition should be implemented, three alternatives are currently evaluated:</p>
<ol>
<li>summation of double double precision values (fast, but only limited guarantees against cancellation)</li>
<li>summation of wordsize integers derived from scaling of the FP values to a pre-established exponent (slow on some platforms where fp to integer conversion is slow, does not address cancellation or overflow)</li>
<li>summation of bignum integers derived from conversion of the FP values (slow, but addresses both overflow of intermediates and cancellation on any exponent range)</li>
</ol>
<p>The first alternative is already implemented in a prototype and only needs porting to the ScalES-PPM.</p> getrusage - Support #8 (Assigned): Add memory use sampling for Linux /proc interfacehttps://swprojects.dkrz.de/redmine/issues/82009-06-16T08:42:39ZThomas Jahnsjahns@dkrz.de
<p>Linux does not support the maxrss field of getrusage, but can keep track of memory use by enumeration of memory mappings in /proc/pid/maps. libproc of procps package <a class="external" href="http://procps.sf.net/">http://procps.sf.net/</a> should hold the necessary code for interpretation.</p>