<div dir="ltr"><div>WWIII Notes - Weds, Jan 17, 2018<br><br>Hi Everyone,<br><br>Thanks to everyone who joined the call yesterday. We discussed a few issues that have come up here at NCEP and at the UK Met Office. There are now VLab issues for all of these as a way to keep track of things. If you are interested in keeping tabs on a particular issue you can click on the “watch” on the upper right corner of any of the issues in Redmine. You will then receive emails when there are updates. <b>A description of the issues discussed yesterday: </b><br><br><b>*Bit reproducibility</b> --<a href="https://vlab.ncep.noaa.gov/redmine/issues/45652">https://vlab.ncep.noaa.gov/redmine/issues/45652</a> -- Since our conversation yesterday, I have run the matrix.base set of regtests (excluding the hybrid) on the latest version of the master twice and compared them. The results are attached in the issue. The good news is we do have some regtests that give identical results and this list will allow us to hopefully narrow our search in why we are not having bit reproducibility. If you have any suggestions on how to obtain this goal please let me know. The current thought is to look at issues of single v double precision, uninitialized variables, and to narrow down the section of codes based on the output from comparing the regtests.<br><b><br>*Wave/current interactions GSE like effects</b> -- <a href="https://vlab.ncep.noaa.gov/redmine/issues/45654">https://vlab.ncep.noaa.gov/redmine/issues/45654</a> - NCEP's operational model has shown signs of Hs decomposition that look like, but unlikely are, GSE. Figures illustrating the problem are attached in the issue. During the call we ruled out the issue being caused by a restart issue, since the problem persists for 117 hours which would have been smoothed out had it been a restart issue.<br><br><b>*Wave/current interaction in refraction - </b> <a href="https://vlab.ncep.noaa.gov/redmine/issues/45655">https://vlab.ncep.noaa.gov/redmine/issues/45655</a> - UK Met Office is seeing wave/current interaction issues in that the refraction, particularly the k-space shift, may change the spectral shape into narrowly focused beam, which the non-linear interaction term may not be able to handle properly and lead to unreasonably high waves. At present, they can cure it by simply switching off the k-shift but further investigation is needed to understand the underlying mechanism.<br><br><b>*Gint+Ice issue </b>- <a href="https://vlab.ncep.noaa.gov/redmine/issues/45653">https://vlab.ncep.noaa.gov/redmine/issues/45653</a> - when using ww3_gint to interpolate from an unstructured grid to a regular grid there are spots of high HS where there is ice. Examples of this issue are attached in the issue. One suggestion for a fix (or work around) was to try with IC2 or IC4 instead of IC0.<br><br><br></div><div><b>Master Updates: </b><br></div><div><br></div><div><b>* The master was updated today with </b><br></div><div><br><b>*We also have updates from Jian-Guo coming soon</b> who has added OpenMP capabilities for SMC grids and regular grids using the UNO scheme. From Jian-Guo: These updates are “so that the model could be run in hybrid parallel mode, that is, spatial propagation for a single spectral component could be calculated in OpenMP mode while different components are still parallelised in MPI mode. A clear advantage of this hybrid parallelisation is that memory demand on a single node could be reduced by a factor of 9 (using 9 threads per rank and 4 ranks per node for instance) so you could enlarge your model domain or increase spatial and/or spectral resolutions. However, OpenMP mode is not very efficient as our tests indicated. For instance our 9-threads spatial propagation could only reduce the elapsed time by 30% than a single thread (the MPI) case. So the hybrid mode could be quicker than the MPI mode only when they share the same number of ranks, that is, a 90-node hybrid job is only about 30% quicker than a 10-node MPI one.”<br><br><b>Next months call is currently scheduled for:<br><br>Wednesday, February 21st<br>10:30 am ET<br><br></b>Note, I might send out a doodle poll for attempting to find a better day that works for everyone. I will plan on presenting on recent coupling efforts with WW3 here at NCEP, in addition to getting updates for v6 projects and a discussion of any recent issues. <br><br>Best,<br>Jessica<br clear="all"><div><br>-- <br><div class="gmail_signature"><div dir="ltr"><div><div dir="ltr">Jessica Meixner<div>NCEP/EMC Marine Modeling and Analysis Branch </div><div>NOAA Center for Weather and Climate Prediction </div></div></div></div></div>
</div></div></div>