Alastair, Tobias, Stephen, Richard, Matthieu, David, George, Peter, Ben, Tom, Scott, Edo. - Actions Action: email around working group meetings that are known about. Action: Richard to chase Katie about response to ARM. - General updates Matthieu - ARM - unless there is a feeling that next machines will be ARM, wait and see. - CiSE article (Richard) Special edition of Computing in Science and Engineering (CiSE). Invited to submit an article. Helps us, something to refer to and point people to. Special issue on the ExCALIBUR project. Input from Peter and Abouzied. Tobias to put in something about separation of concerns. - June 8th workshop (Richard) A workshop to compare progress between groups. See whether working groups could combine phase B proposals. Can't be shifted. Brief status summary, 1 slide. Intended for teams to meet and contribute in breakouts. 9:30-1pm. Spheric is also on. Matthieu can join for a bit. Can't see how a useful phase B proposal could be combined with another working group. - challenge problem (May 28th) (Richard) Meant to set a challenge problem / microbenchmarking app, which can be used to benchmark the exascale systems. Similar to the swift benchmarking repo with reframe etc. Can use this. Concentrate on weak scaling. GPU port not ready. For challenge 2, put some eng resource into GPU time, or something more abstract. Difficulty is identifying a correct case for this. Ben/George to think about what can be suggested. Are they happy with coupling? Some physics on GPU, some on CPU. A problem we can do today on largest computers, how can we scale this. Richard to ask Andy Turner about extending Edo work. A real calculation or something EPSRC fund high end consortia, funding computational projects. Is a whole community on turbulence and turbulent reacting flows. This could be a way of linking into existing activity. Significant link between outputs from excalibur and these HECs. If looking for a reason for a generic case around turbulence, could link back to this. Mesoscale workgroup could also be useful here. Can put in a challenge-2 problem along these lines. - Training (Tobias?) http://www.peano-framework.org/index.php/workshops/2021-performance-analysis-workshop-from-analysis-to-insight/ This Thursday is all about correctness tools - tool demos and presentations about MUST. Assesses correctness of MPI codes. Archer tool checks for threading errors. Also tool about memory access correctness. Will help with scaling. - Cross cutting proposals (met office) - Tobias, task parallelism Learnt that big has been successful, together with Hartree. Start date 1st June. Another series of workshops. Work on tools that allow tracking of parallelisation character. Proposing to introduce pragmas or an API to generate abstract footprint and a simulator. Second pillar of the grant will push people to a data-driven paralllelisation approach. 3rd pillar looks are features in task-based which are missing. e.g. interaction of taskbased and distributed parallelism. Temporary moving of tasks to nodes. Starts 1st June, for 3 years. Looks at task inefficiency patterns. 0.3 FTE for 3 years. Hartree 0.2 FTE for 3 years. - Dave - time parallelism) Includes weather/climate and fusion. Community building and outreach for time parallelism. KO 1st June. Identify opportunities for Knowledge Exchange. Trying out different time-parallel algorithms. - Code coupling (Ben/Stephen) PI response for 1st stage due this week, ongoing. - Graphics (Dave) Student project. Released to students this week. - Future paradigms (RGB/DiRAC) Should be hearing. - Quantum computing (RGB) Should be hearing. - GPU update (Abouzied) Good progress implementing loops. On leave this week. - RDMA/MPI updates (Peter) Playing around with reducing data sent across fabric. Version of swift that only exchanges parts of gravity particles that it needs to. Probably going faster than it used to. Some overheads of moving data around the machine. Seeing 10-20% improvement for some steps where particles had already been transferred. Will re-run everything. - Hydro weak-scaling (Edo) Repeating some of the weak scaling. Thinking about geometry. Gizmo paper driven turbulence test. How to set up initial conditions from this. Driven turbulence - need code to drive in Fourier space. - other workpackage update [all] - Others/AOB? Next meeting: 12pm, 1st June (Tuesday)