Workshop: Federated Infrastructures und cloud computing: Organisation and Preparation of BMBF Proposal - DESY Indico
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
Workshop:
Federated Infrastructures und cloud
computing: Organisation and Preparation of
BMBF Proposal
4 September 2020
KAT Andreas Haungs, KIT
KIT – The Research University in the Helmholtz Association www.kit.edu
Initiative for a Large-scale cosmic
structure: fields and Gravitational waves Nuclear
(national / global) objects Astrophysics
Analysis & Data
Ultra-high energy
Center cosmic rays
in Astroparticle
Physics Galactic
cosmic rays
Astroparticle Physics =
Understanding the
Multi-Messenger gamma
neutrino astronomy
Universe
mass
Dark Universe
needs an search for Dark
experiment-overarching Matter annihilation
(computing) platform!
(there is no CERN or FAIR or
ESO) neutrino
search for Dark
Matter scattering astronomy
J.Blümer
2 A. Haungs, September 2020
Analysis and Data Center in Astroparticle Physics
Simulations Real-time Education
Data Open Data
Analysis & Methods analysis in Data
availability access archive
development center Science
Data availability: Open access:
All researchers of the individual experiments or facilities It is necessary to make the scientific data available also to
require quick and easy access to the relevant data. the interested public: public data for public money!
Analysis: Education in data science:
Fast access to the generally distributed data from Not only data analysis itself, but also the efficient use of
measurements and simulations is required. Corresponding central data and computing infrastructures requires special
computing capacities should also be available. training.
Simulations and methods development: Data archive:
Researchers need an environment for simulations and the The valuable scientific data and metadata must be
development of new methods (machine learning). preserved and remain interpretable for later use (data
Real-time analysis center: preservation).
The multi-messenger ansatz requires a framework to
develop and apply methods for joint data stream analysis.
3 A. Haungs, September 2020
Status Infrastructures in Astroparticle Physics
Computing:
• (Co-use of) Institutional resources (partly WLCG resources)
• GridKa: Tier1-centre in the world wide LHC Computing Grid (e.g. Auger@GridKa)
• Experiment-oriented resources (e.g. CTA@DESY)
• Co-use of facility infrastructures (e.g. IceCube at DESY)
• Moderate use of HPC cluster (Gauß Alliance)
Resarch Data Management:
• KCDC: KASCADE Cosmic ray Data Centre (data access)
• VISPA: to analyze data (Learning Deep Learning)
• GAVO (German Astrophysical Virtual Observatory)
• CERN Open Data Portal (not yet used by APP)
4 A. Haungs, September 2020Federated Infrastructures for Astroparticle Physics (…in Germany)
• Starting position
o more and more complex experiments and research facilities
o rapidly increasing digitization levels and therefore growing data volumes of the instruments
o sophisticated simulation and data analyses
o request of combination of data from different facilities (Multi-Messenger APP)
considerably growing needs of the scientific community for an efficient Information and
Communication Technology (ICT) infrastructure
• A scientific (ICT) infrastructure for data-intensive research requires
o large Storage, fast Network, high Computing Power
• The future computing model for Astroparticle Physics will have many similarities and synergies with
the HEP (HL-LHC) activities
• Such an ICT infrastructure for Astroparticle Physics should be seen in the context of broader
national research data infrastructures and at the international level, e.g. in the context of European
cloud initiatives
• Such a common virtual ICT infrastructure should be connected to experiment-specific infrastructures
and should foster the inclusion of commercial resources.
requires in Germany a dedicated (federated) infrastructure
5 A. Haungs, September 2020Assessment of the demand for federated resources in computing of APP:
- To WLCG system projected requests of German share of computing requests of the ErUM-Pro projects
in addition to usage of institutional resources.
- Projected requests for 2028: factor ~8 for CPU, ~5 Disk and ~10 Tape, factor ~20 for GPU (mainly due to
ET)
- Theory: The current needs are met by federal or state-operated supercomputer centers (Jülich SC, Leibniz Center Munich, HLRN,
etc.); not clear if this is possible for 2028.
Request in 2021 GERDA / Multi-
Auger IceCube CTA* ET KATRIN LEGEND DARWIN Messenger Theorie Summe
CPU [CPU-years] 500 500 500 0 500 n/a 0 100 0 2100
GPU [GPU-years] 40 200 0 0 0 n/a 0 50 0 290
Disk [PB] 0.8 1 0.5 0 n/a n/a 0 0.2 0 2.5
Tape [PB] 3 0 0 0 n/a n/a 0 0 0 3
Projected for 2028 GERDA / Multi-
Auger IceCube CTA ET KATRIN LEGEND DARWIN Messenger Theorie Summe
CPU [CPU-years] 800 2000 1000 5000 600 n/a 2500 1000 1000 13900
GPU [GPU-years] 70 400 0 5000 400 n/a 13 500 300 6670
Disk [PB] 1.5 2 3 2 n/a n/a 1.9 2 0.2 12.6
Tape [PB] 5 10 10 0 n/a n/a 1.1 4 0 30
6 A. Haungs, September 2020Draft of a text as result of this assessment:
Deutsch:
Der Bedarf an Computing-Ressourcen für die Astroteilchenphysik in Deutschland wird in den nächsten Jahren beträchtlich
zunehmen. Im Jahre 2020 erfolgt das Computing für die deutschen Leuchtturm-Experimente (Auger, CTA, IceCube, ET, KATRIN,
Gerda/Legend, DARWIN, Multi-Messenger, Theorie) im Wesentlichen über institutionelle, experimentspezifische oder wie bei der
Theorie über föderierte Supercomputer- Ressourcen und nur zu einem kleinen Teil über den deutschen WLCG-Verbund. Eine
Abschätzung des Bedarfes für 2021 für den deutschen ‘fair-share’ Anteil am Computing der internationalen Experimente ergab eine
Summe von 2.000 CPU-Jahren, 300-GPU-Jahren, 2.5 PB Plattenplatz und 3 TB Tape Kapazität, die im Wesentlichen bereits über das
WLCG (Tier-1 und Tier-2) abgedeckt sind. Eine Projektion in das Jahr 2028 ergab einen gesteigerten Bedarf von ca. Faktor 8 in
CPU-Jahren, Faktor 20 in GPU-Jahren, Faktor 5 in Plattenplatz und Faktor 10 in Tapes.
English:
The demand for computing resources for astroparticle physics in Germany will increase considerably in the coming years. In 2020,
the computing for the German flagship experiments (Auger, CTA, IceCube, ET, KATRIN, Gerda/Legend, DARWIN, Multi-Messenger,
Theory) will mainly be carried out via institutional, experiment-specific or, as in the case of theory, federated supercomputer
resources and only to a small extent via the German WLCG network. An estimation of the 2021 requirements for the German fair-
share of the computing of the international experiments resulted in a sum of 2,000 CPU years, 300 GPU years, 2.5 PB disk space
and 3 TB tape capacity, which are already largely covered by the WLCG (Tier-1 and Tier-2). A projection into the year 2028 showed
an increased demand of about factor 8 in CPU years, factor 20 in GPU years, factor 5 in disk space and factor 10 in tape capacity.
7 A. Haungs, September 2020You can also read
