H2020 and HPC: A view from INFN

Transcription

1 H2020 and HPC: A view from INFN R. (lele) Tripiccione INFN and Universita' di Ferrara tripiccione@fe.infn.it H2020 Info Day Pisa, Feb. 5th, 2014

2 INFN Experimental and theoretical research in subnuclear, nuclear, astroparticle, complex systems physics... as well as substantial developments in related enabling technologies Some 2000 research staff/collaborators scattered at several locations all over Italy Strong ties within and outside the EU (CERN, DESY, GSI, SLAC, JLAB, FERMILAB )

3 INFN Key role and key responsibilities in well known large scale scientific experimental projects e.g., find the Higgs boson.... as well in the development of the theoretical models that explain how our Universe works...

4 . as well in the development of the theoretical models that explain how our Universe works (at all scales) which (on top of bright ideas and bright people) requires huge HPC computing resources. How huge is huge? 57

5 How huge is huge??? 0.5 Pflops x 100 d ~ 100 Tflops x 500 d In 2013: ~ 300 MCorehours on BG/Q (and counting...) 55

6 Long Term goals for HPC computing at INFN: Make sure that we have continued access to (appropriately scaled up) computing resources in HPC and make sure that we can use those resources promptly and in the most effective way. > Develop our own ideas and machines (the old way) Team up with computer centres to access state of the art systems Work hardly to use new ( nasty ) architectures as efficiently as possible Work with computing companies and share our technology background

7 INFN background in HPC: Bits of history The APE projects: several generations of application driven HPC machines for Lattice Quantum Chromodynamics Gflops ~ 20 Tflops

8 INFN background in HPC: Today Development of high bandwidth/low latency toroidal communication networks ApeNET (v5...)

9 INFN background in HPC: Today (cnt'd) Algorithmic research, design optimization and operation of large HPC codes (mainly in LQCD, LBM fluid dynamics, spin systems) Using different algorithmic strategies... Parallelizing techniques Programming methodologies. and understanding if things work correctly or why they do not... (and we even tried GPUs + ARM...)

10 A (gloomy) lesson from the past The basic bits & pieces of state of the art HPC were available in Europe ten years ago The scientific community, the funding agencies, european industry were all not able to spot the potential and an opportunity was lost

11 A (gloomy) lesson from the past The basic bits & pieces of state of the art HPC were available in Europe ten years ago The scientific community, the funding agencies, european industry were all not able to spot the potential and an opportunity was lost At the same time, on the other side of the Atlantic > Application specific LQCD computing was also being developed in the US, with a similar work style and similar funds The US project was called QCDOC......and then someone spotted the opportunity And the winner is >

12 And the winner is >

13 And the winner is > Blue Gene L (then P, then Q, then???) are now the workhorses of HPC computing (almost) all over the world. A golden example of getting the truly best of basic research for industry!

14 History likes to repeat itself. A new opportunity is opening right now: Green (HPC) computing Thanks to F. Mantovani, A. Ramirez (BSC)

15 History likes to repeat itself. A new opportunity is opening right now: Green (HPC) computing Europe has several strong assets in this area: The Arm Architecture A recent first position in the Green 500 list (Eurotech) A major company is involved in Arm manifacturing (STM) Architectural and application know how So the key question is: Are we going to be able to leverage on these strong points and have a sizeable impact within (and possibly beyond) Europe??

16 Take away lessons New opportunities are fragile: they fade away quickly, if not exploited at the right time The effort is formidable: it can only be pursued if a large enough critical mass (and support) is put together The combined effort of academia, research centers, computing centers, industry is needed The combined effort of the supply side as well as the demand side is needed Find a (reasonable) goal and go for it (with enough determination)

17 What does INFN offer?? Know how and support in the definition of a suitable green HPC architecture Architectural modeling and performance estimation Porting, validation, test, performance measurement of key heavily used HPC applications Training of top class HPC experts for scientific/engineering computing

18 What does INFN expect?? Know how on new generation HPC architectures Know how on application development/porting/tuning for new HPC architectures Improved availability of computer resources for basic science within a finite time ( H2020 ) horizon...