Present: Alastair, David Acreman, Richard Bower, Benedict, Rogers, Peter Draper, Tobias Weinzierl, Tom De Vuyst, Jonathan Frawley, Stephen Longshaw, George Fourtakas, Abouzied Nasar, Matthieu Schaller, Scott Kay, Edo Altamura. Action: email around working group meetings that are known about. Action: Richard Set up coordination between working groups Action: Richard Swift training day in February - workshop in Feb/Marc Action: Ali contact Olly Perks at ARM - Done Action: Ali setup NDA with networking company - Done - General updates ================= Swift traing day now a workshop in Feb/March. Stephen in touch with rolls royce/asimov project. Keen to get in touch. Simon Makintosh Smith talking about Excalibur hardware, and what we're likely to have. Need more industry speakers, have space for 4. Ben contacted motor sport company, provisionally yes, but don't want to give away trade secrets. Ben contacted national nuclear lab - keen to attend, but not sure about whether will talk. Tom - spoken to ESI - all sorts of simulation software. And will speak to someone in JLR. Carlos to talk on astrophys side. Stephen to contact EDF. And Airbus. Technical partners - probably not. Want to get date tied down, but needs clarification from speakers first. - Contracts (Richard) Required desperately. Just need Leiden now. Set up something separate with NVIDIA. - Excalibur funding call Call from Met Office. Cross cutting proposals have partly gone into this, but with some changes. Groups already doing this with some success - probably referring to us. Doesn't really reflect what we do. Surprised - old fashioned, should be additional level of task based parallelism. Seemed more along the lines of coupling different tasks. Call for disruptive tech. Parallelism is more than coupling codes. Possibly not what they're expecting. More detail will be released by EPSRC in early February. Met Office running this like a tender. Scope is 1 or more of the high prio use cases. Want applied to specific thing but with ref to other areas. e.g. swift already does, but could be applied to met office. 5th March deadline. Tobias to lead. Dave to set up a call with Beth at the Met Office (Richard, Tobias, David). Separate proposals, SPH working group = 1 proposal. Gemma will also put in a parallel in-time proposal. Approaching from opposite end. Look at the task graph, take expensive parts out. Talk to Beth about this. - Hydro weak-scaling (Edo) Investigating possibility of doing large weak scaling tests. Made initial conditions. Some runs are large, several hundred nodes. Options either a reservation, or wait until after downtime, and have it launch then. Run once with intel mpi and once with openmpi. - RDMA updates (Peter) Still working on it. Code about as fast as MPI. Digging into why not seeing improvements. Seems to hit theoretical limits. But might be some workarounds. Seems to be memory bandwidth to a single core at the moment. - GPU update (Abouzied) Understanding swift. How to use quickshed. Split gpu tasks into streams, to overlay data transfer and computation. Discussion of how to partition into GPU - Individual workpackage update [all] - Training (Tobias?) - Outreach (?) - Others/AOB? - Actions Action: email around working group meetings that are known about. Action: Richard Set up coordination between working groups - haven't found a way to do this yet. Action: Richard Swift training day in January - sponsor the next sph-sig meeting and make it exascale ambition. Get Carlos to talk on astrophys side, someone from eng side, and possibly someone from met office. Next level of ambition. From eng perspective, could try eng companies. Benedict can approach some of their sponsors. Where would we like to be in 5-10 years time. Scope out what we would imagine being able to do on an excalibur-like system. Someone to give an overview of hardware likely to come online. Perhaps Simon McIntosh Smith. Mid-late Feb. Possibly 24th. Offer to hold the training sessions after the talks, but get people to express interest in advance. Longshaw mail list to be used. Tom and Ben to talk to industry contacts. Stephen to send something to Lee Lapworth at Rolls Royce. Action: AGB prompt Olly Perks at Arm. Action: AGB to set up NDA with networking company. - General updates New PoC with networking company. - Contracts (Richard) All should have received Katie's email about the contracts. Waiting to hear from Hertfordshire. And Leiden. And LJMU. Sent just before Christmas. Commercial partners - need to hear from arm. AGB to prompt Olly Perks. Katie pushing for 15th Jan. - Hydro weak-scaling (Edo) - RDMA updates (Peter) Runs about as fast as MPI. Struggling to understand what has gone wrong. All enhancements in place, but still somewhat slower than expected. Some possibilities, problem is probably in swift. Suggestion - add a global dependency on all the timing tasks. - GPU update (Abouzied) Nothing to update. - Individual workpackage update [all] David - hoping to have a student working on this for a masters project, but he has got a job instead. So need to get back to doing stuff. Blended delivery of a module this term, so will have to delay until later. Thing to look at is caching of mpi call, grav-only calc, cache bits of the tree. Fast running particles don't do the long range calc. Helps for gravity but not hydro. Richard - Rudigar Pakemore has solved asynchronous timesteps. But not for hydro. Can do this in arepo, but not exact. Need to be able to write down the hamiltonian of the system. Dave - different ways of thinking about this. eg split into linear and non-linear operator. - Training (Tobias?) Starting perf anal workshop - >60 people signed up. Jonathan leading swift in this. - Outreach (?) Action - Content for website - - Others/AOB?