New supercomputer: Gadi

In late 2019 to early 2020 the National Computational Infrastructure (NCI) will be bringing online its new supercomputer Gadi, the successor to Raijin. Full details of the new system are yet to be announced, but we do know that there will be some changes to how the new supercomputer will work compared to Raijin. CMS will put in place some new procedures to help smooth the transition to the new supercomputer over the coming months.

Files on Raijin’s /short filesystem will not be moved to the new machine, and there will most likely be a time limit instituted on how long a file can remain on Gadi’s /short space. The /g/data filesystem for long term file storage will be available unchanged on the new supercomputer.

To prepare for this change, CMS wants to help Centre researchers identify which files currently on /short need to be archived. We’re beginning to track disk usage in our NCI project accounts in greater detail using an analytics package called Grafana, which will let both the project account managers and members get a better overview of how resources at NCI are being used.

In addition to the disk changes, the new supercomputer will have the same sort of software module system as Raijin. However, older versions of the module software will be retired, with NCI only installing the most recent version of each module on the new supercomputer. Many of the software versions available on Raijin date from NCI’s previous supercomputer Vayu, and are outdated or no longer supported by developers.

To account for this some programs and libraries will need to be recompiled with new module versions. CMS will set this up for the models we support as Gadi is brought online.

More information about the new system can be found on the NCI website, here.

WRF Single Column model

Scott has been assisting Yi-Ling Hwong in converting forcing data from the GMTB single-column model to be readable by WRF. GMTB is a testbed for single-column model development, allowing WRF to run from similar input data. It will be a valuable point of comparison for Yi-Ling’s model comparison work.

To date, we have set up the initial conditions and forcing for the GMTB ‘twpice’ test case. Yi-Ling is currently looking at the results before we move on to other GMTB test cases. We are hoping other WRF users will benefit from the inclusion of WRF in the GMTB testbed.

ERA5-Land

Following the release of ERA5, ECMWF just released ERA5-Land. This is a replay of the land component of ERA5 to make it more accurate for land applications. You can find a full list of ERA5-Land features on ECMWF documentation.

Some differences from ERA5 to note are the use of a higher spatial resolution (9km vs 31km), the temporal frequency of the output is 1 hour and the data is masked for all oceans. The data is currently available from 2001 to almost the present time. The period from 1950 to 2000 will be released by batches between Australian Spring 2019 and the end of 2020. Eventually, monthly averages will also be published.

We are currently able to replicate the ERA5-Land data to NCI upon request. Feel free to ask CMS to download data you are interested in. We ask that you use the Copernicus data portal to precisely identify which variable(s) you want before putting in a request to avoid confusion about what we need to download.