Present: AGB, DA, PWD, RGB, BR, EA, GF, JF, SL, MS, SK, TW, TdV Alastair Basden, David Acreman, Peter Draper, Richard Bower, Benedict Rogers, Edo Altamura, George Fourtakas, Jonathan Frawley, Stephen Longshaw, Matthieu Schaller, Scott Kay, Tobias Weinzierl, Tom De Vuyst - General updates Manchester interviewing internal candidate next week. So hopefully will attend in 2 weeks time. - Contracts (Richard) Contacted Katie, all academic partners have indicated they're happy with the legal agreement. Issue is not heard back from the industry partners. Will send around a revised agreement that will go directly to legal people. Main goal is to get response from industry partners. If this fails, will try to get 2 agreements, one academic, and one industry which may never arrive. Should all be doable, and hopefully sorted within a month or so. Action: Alastair to try to nudge commercial partners. Action: Richard to speak to Katie just before next meeting. - Gadget4 timestepping update (DA) PPT Presentation. Elimination of asymmetric operators Options - cache particle info to reduce comms at all timesteps. Use G4 style hierarchical time integration to reduce comms within global time step, alternative time stepping scheme which is parallel in the method. - Hydro weak-scaling (Edo) Working on 3D tests. Tiling 2d init conds in layers with mean inter-particle separation. Form a cube with tiled init conds. Initial velocities don't have a component in the same layer. Most importantly, allows testing of 3d kernel for sph. Putting on proposal for subgrid modes for swift and agn feedback Driven turbulence tests. May never succeed with this for sph, but setting it up would be good. Weird behaviour over mpi - need to work on this. - RDMA updates (Peter) Not much to report. Made attempt to adapt some methods for swift. Went back to mpi 1-sided stuff, but not yet working in swift, may not be obtainable. - Individual workpackage update [all] Tobias - implemented particle management with trees, multitasked it, bug in nvidia and intel c++ implementation. Try to find out whether pthreads whether set certain barriers whether the task can swap in other tasks. Deadlocks. Hopefully something written down before Christmas. Intel have stopped development of TBB. Because new c++ standard offers tasking. Should use the c++ features, and rely on the language. Got some docker snapshots about what blocks. Similar with openMP too. Engagement. - Training (Tobias?) Organisation - epsrc proposal for training need gap on rse side in hpc context. Understanding how codes work not just using performance profiling tools. EU Popcorn profiling. - Outreach (?) Bug reports to intel/nvidia. Other groups - excalibur working groups - meant to get together to share common issues. Stephen contacting Garth. Richard will look at setting up a meeting in January for a cross-excalibur meeting. Gen-X group meeting on 25th Nov. Excalibur working group email list would be good and a web page. Action points: email around working group meetings that are known about. 2 sets of seminars. sph virtual open meeting 25th nov. quick round table. spheric seminars. monthly global seminars. - Others/AOB? AGB: Shared sph talk details with ICC. DiRAC call.