# DEEP-SEA ###### tags: `EUROHPC` optimisation cycles share experience with application developers. get inputs from this group to make the website. monthly meeting with dissemination leaders of other SEA projects. Project objectives: - all on-going. Optimisation cycles --> pave the way to a more systemiatic way of co-designing application development. between the system software and applications. JUBE Benchmarking environment. + continous integration CI Gitlab, EasyBuild and Spack (future). Spack--> only for software installation. Installation to other sites. ## WP3 - Software architecture and system software malleability: Shrink & expand --> what is required at the application level? Any link with ADMIRE. T3.6 Continuous integration How does it work in the repo--> list all software/applications you want to add? integration of all DEEP-SEA components into the CI/CD infrastructure. + Spack. If not JUBE then how is it done? Automated deployment of DEEP-SEA component on the HPC. ## WP4 Node level programming Task 4.1 software stack at a glance --> very good picture. xPIC application (planning) buffering and caching Task 4.2 Omps and OpenMP optimise Energy efficiency without impacting performance OpenMP standard --> add NUMA support. BSC is part of OpenMP ARB so can influence. Dynamic load balancing. Useful for which applications? Standard need to be update so be able to say that other threads can join the "team" and increase performance. AOB Fine grained Dynamic Load balancing: OpenIFS, GROMACS OpenMP Language committee WP6 (ask again about standard) Coarse crained Load balancing Will be proposed to ECMWF and CEA Dace --> move from a proof-oc-concept tools to a DEEP-SEA stack. DaCe used in GROMACS and NEK5000 NabLab DSL-based. HPDA+HPC K2HPC Virtual Kubelet Which use case? BioInformatics workflow. Which BioInformatics workflow. Programming environment for PIM --> showcase but not mature enough to be taken up by applications. White paper on PIM (explaining why it is a viable approach). Work is continued in WP2. All tasks are on track. ## WP5 extending programming environment one task per objective. T5.1 Extension for MSA Focus on MPI. 3 different stacks. HAN: Collective communications XPMEM: inside a node. XHC Combinable with HAN Open MPI : notified RMA and their Fortran binding (not in the standard). T5.2 malleability TSMP applications. Could applications from other projects such as ADMIRE projects be used? How to use MPI session to work with malleability Work with PMIx T5.3 interoperability and composability T5.4 Resiliency T5.5 Interfaces to support modular heterogeneous systems Health check for the node. Which applications would benefit from it? Malleability --> coordinate with other EuroHPC project to accomodate different runtime. ## WP2 Measuring, Modelling, Mapping & Monitoring Memory performance. Extrae, PROFET and PAraver. Easy integration to allow tracing memory easily from any user application. Data movement s Sys-sage MSA Profiling and tracing Scalasca and Score-P Possibility to leverage boosted frequencies for heavily compute-bound workloads. export of DCDB monitoring data to LLview for job-specific monitoring For DPU; model would need to be modelled. Energy consumption. link with ADMIRE? Not much . So very different from ADMIRE. ## WP1 Applications. NEK5000 --> NEKU CFD Applications. DEEP system --> node are shutdown if not used within the last 30 minutes. DEEP-SEA and IO-SEA use the same Benchmarking framework. e.g. they use the same JUBE framework. Benchmarling and Cross-SEA collaboration (slide 164) Anything with other projects? Each application is instrumented with perfoce analysis tools Score-P or Extrae. 1.3 xPIC TSMP working with maleability Trying to implement the mallabelibity cycle. Mockup used to implement the malleability cycle. Then will be introduced in the "original" code. Collaboration with TIME-X on time parallelisation. With which applications? TSMP and malleability. Could also be used with other applications. And for instance NEK5000 (new version called NEKU) Questions: -> Complementary TSMP --> different component Indirectly with NEK 5000. JUBE synthetic benchmark could also be shared with ADMIRE. starting collaboration on Benchmarking. ## WP6 Dissemination and exploitation increase DEEP-SEA and other SEA projects 3 website projects including EuroHPC website: overview of all the projects --> should it relay what is done at project level? How do you populate it? blog: collaboration news. BSC hosted a two-day forum for advancing memory systems solutions. Open exchange and discussion between industry experts from Intel, Micron, Atos, SiPearl and other European research centres. "other European centres": who are they? Training and education. Invite other experts from other SEA and EuroHPC projects. EuroHPC Workshop on Malleability in HPC. Who does organize it? Is it viable to maintain live 3 websites? The main EuroHPC website: who is in charge of it? Who is populating it? Internal KPIs to assess the progress of the project in term of achievement. How do you measure it? Not ready yet? Collaboration Board Gave budget e.g. split budget. WP7: Coordination and Collaboration FORTH: more time reported than initially planned. Because of junior person. Very good project management Communication campaign High quality EuroHPC website Timeline well organized.