Minutes for the Einstein Toolkit, 20241219. https://docs.einsteintoolkit.org/et-docs/Main_Page#Weekly_Users_Call 9am CST
Present: Steve B, Bill G, Roland H, Maxwell R, Leo Rosa W, Zach E
Chair Steve, Minutes Bill
Last meeting of the year. No meeting on the 26th December nor on the 2nd January. Next call is 9 January 2025.
Steve, looking at MoL recently and it should work with CarpetX if someone provides linear combination. Roland wonders if tiling is a problem. Steve pushes back about traversing the schedule versus traversing the grid. Scratch variables are not grid functions in ODESolvers. Roland thinks it might well work. Steve, it would be another capability if people want it. MoL does have a linear combination routine if nothing overrides it. Roland, could make MoL use GPUs in this manner.
* Simfactory issues on Anvil [RH] Roland, Maxwell looking into this. Simfactory for the release gave 1/70th the speed on Anvil. Maxwell used his files and it gave the expected performance. So something is off on the ones in the toolkit.
Maxwell, my files are not public yet. The diff shows some simple changes, like modules which he does not expect to make a large difference, another difference is the amount of make jobs and default amount of nodes in the .ini file. Nodes is the max number of nodes the cluster has, a limit. CPU frequency is different too but is purely information. Make jobs 8 to 4 should not affect it. So seems that difference is in the modules. Roland asks about the OptionList, and expects the difference there. Roland, could be off in the runscript. Maxwell, my machine file use --mpi=pmpi2, was told to do that for some speedup, pins MPI ranks. Steve suggests changing to the other value. Roland, you need to Reconfig with scripts but will not re-compile anything, or copy by hand the runscript in config/sim .
Maxwell, will run more tests to pin down the parameter that affects the runtime.
* upcoming ET possible inclusions ** NewRadX boundary [CSC] Did not get the update into the previous release. Needs two changes, one change to CarpetX and the other to SpacetimeX. See ticket #2820.
** Z4c, high priority needs champion and reviewers [RH] Erik would like to have a co-champion and Roland can do that. Reviewers are Steve and Peter. See ticket #2748.
** BHaHAHA horizon finder [ZE] Zach, paper is being written, sections are finished. Finish the paper and up on Arxiv before the new year. So on track for the next release.
** CCE code to hook up to Spectre [RH] Cauchy Characteristic Extraction code, interpret the cauchy grid for spacetime evolution and on the null grid, and to ScriPlus that way. Deborah Ferguson and should be ready soon.
** possibly more modules for cosmology [RH] Roland, will check if Hailey has anything ready and can champion it.
** gallery examples for Canuda [CSC], Fishbone-Moncrief, IGM, GRHayL and others [RH] New gallery examples, two to contribute related to Chen and Deborah Ferguson, on axidilaton field part of Canuda but not the Toolkit. Like BNS relies on external package, Proca and Gallery thorns. It has a picture and text and estimates for runtime.
Fishbone-Moncreif postponed until those thorns have been cleaned up.
Roland, GrHayL or IGM, it would be nice to have gallery examples to make people aware of other codes in the Toolkit. Could be simple as BNS example, use with Illionis GMHD and Baikal. Zach, also do with BHaHAHA. Once McLachlan is updated with improvements, we can update.
* unanswered question on mailing list https://www.einsteintoolkit.org/tools/unanswered.php Just advertisements.
* open tickets sorted by update time https://bitbucket.org/einsteintoolkit/tickets/issues?status=open&status=...
#2837 Building Einstein Toolkit: Roland, Build trouble on a cluster in Asia. Suggest commenting out thorns you do not use and re-run make thorn. Steve, have maybe run across something similar. Does he have pkg-config installed? Zach, reads like an Adios2 compile issue. Roland, this was fixed by installing cmake on the system...look lower in the ticket. Looks like libsz is missing. Roland, installing HDF5 usually fixes this. Ticket says solved libsz problem. Trouble on OpenSuse and Ubuntu still. Needs hdfs5-devel maybe. Roland, Zypper install helps. Zach, in the Ubunto log bulding with make -j 6 and it is not finding something. Too many makes in a parallel build can expose dependency issues and think that is the problem...maybe with the ADIOS2 build system. Recommends trying -j 1...and wait. Perhaps OpenSUSE is the same/similar issue. Steve, could be something funny with ADIOS2, Steve installs it first with Spack.
#2835 add backtrace functionality from CarpetLib: Erik has reviewed part of it. Moved some of the code to CarpetX. Code same as in CarpetLib. Parameter files in test directory do not have output files, and testing segfault all end in failures...Toolkit test system should support expected failure as an outcome. This looks underway.
#1566 Update Cactus autoconf: Roland will apply it 2024-12-26.
#2807 Hydro_RNSID passes CFLAGS to LD: Roland will apply this after 2024-12-26.
#2836 add support for different data types to GPU arrays: Roland, Erik found some needed fixes. About how the Poison value is handled, set memory to poison, etc. Can discuss the proper poison value. Steve, is it NAN17? Roland, currently NANxxx, can use anything. Add doubles, floats, ints, complex8 and complex16, etc, the ones that CarpetX supports.
* tickets ready for review https://bitbucket.org/einsteintoolkit/tickets/issues?status=new&status=o...
* Any Other Business
Maxwell, discussed support for volume integrals in Kuibit on a Telegram channel. Steve uses Telegram heavily, communicate and to tell him about builds. Because it is using its own output, volume integral output is different than the standard Cactus output, so needs its own custom interpreter. Kuibit lets you do parsing flexibly. Gabrielle started by parsing headers for Kuibit. Zach, Python is a good language for this. In the Toolkit there are a lot of ways to do this. Needs to be able to interpret the strings and what are the important information. Zach, issue is the communication of the output you start with a full domain and then you have spherical volumes from that domain...support swiss cheese type topologies. Do not need to do too much in Kuibit. Maxwell, Gabrielle wanted to be able to change the configuration mid-run as people often do, so want to follow the mid-run changes in output.
From Roland, a negative about Telegram https://www.bbc.com/news/articles/cvgrvw29x4jo "The social media platform Telegram uses an algorithm that promotes extremist content, a new study shared exclusively with the BBC has revealed."
Next meeting 9 January 2025: Chair Leo Minutes Roland