The INFN activities in the frame of the national strategy 1 - - PowerPoint PPT Presentation

▶

Jul 29, 2023 622 likes •812 views

The INFN activities in the frame of the national strategy 1 Tommaso Boccali INFN Pisa A bit of History 2 INFN is the Italian National Research Institute coordinating (and funding) activities on Particle, Astro-particle, Nuclear,

SLIDE 1

The INFN activities in the frame of the national strategy

Tommaso Boccali – INFN Pisa

SLIDE 2

A bit of History …

 INFN is the Italian National Research Institute coordinating (and funding) activities on Particle, Astro-particle, Nuclear, theoretical and Applied Physics  While not primary INFN goal, (scientific) computing has grown as a necessary tool for the research field  INFN has participated or seeded many Italian activities in computing, in the last decades

 The harmonization of an Italian Research Network  The growth of an organic set of High Throughput Computing (HTC) sites  R&D activities in High Performance Computing (HPC)  The harmonization of access to computing resources via the GRID Middleware, and now Indigo-DataCloud

 What next for us?

SLIDE 3

INFN facilities on the territory

 INFN is mostly unique among Italian research institutes (and not only): strong decentralization over the territory, basically have presence

 Wherever there is a (sizeable) Physics Department

 INFN and Academic personnel door to door in the same building

 3 national laboratories  Specialized centers

 Among which, you should already know well: CNAF

SLIDE 4

The Network – GARR (Italy’s NREN…)

 GARR was born in the ’80s, in an attempt to harmonize scientific networks

 Most of the personnel was formerly INFN, and still is …

 Various technological steps since then, now with GARR-X dedicated fibers between INFN centers (and not only), at multiple 10 Gbps, multiple 100 Gbps soon

 More that 15.000 km of GARR owned fibers

 ~9.000 Km of backbone  ~6.000 Km of access links

 About 1000 user sites interconnected

 Among which also schools, hospitals, …

 > 1 Tbps aggregated access capacity  > 2 Tbps total backbone capacity  2x100 Gbps IP capacity to GÉANT  Cross border fibers with ARNES (Slovenia), SWITCH (Switzerland).  > 100 Gbps to General Internet and Internet Exchanges in Italy  NOC and engineering are in-house, in Rome.

SLIDE 5

INFN Scientific computing facilities

 (~) 1990-2000: each INFN facility had a small sized computing center, handling everything from mail servers to the first scientific computing farms  2000+: consolidation

n the WLCG

hierarchy, with one Tier-1 center and 9 Tier- 2 centers – we are still there, with evolutions  In WLCG, Italy around 10% of the total installation  2020+: transition to new models?

PADOVA/LEGNARO Tot Cores: 5200 (55 kHS06) Disk Space: 3000 TB Netw connectivity: 20 Gb/s CNAF/BOLOGNA Tot Cores: 21250 (221 kHS06) Disk Space: 22765 TB Tape Space: 42000 TB Netw connectivity: 80 Gb/s TORINO Tot Cores: 2500 (27 kHS06) Disk Space: 2500 TB Netw connectivity: 10 Gb/s MILANO Tot Cores: 2448 (23 kHS06) Disk Space: 1850 TB Netw connectivity: 10 Gb/s PISA Tot Cores: 12000 (125 kHS06) Disk Space: 2000 TB Netw connectivity: 20 Gb/s FRASCATI Tot Cores: 2000 (20 kHS06) Disk Space: 1350 TB Netw connectivity: 10 Gb/s ROMA Tot Cores: 3172 (32 kHS06) Disk Space: 2160 TB Netw connectivity: 10 Gb/s NAPOLI (INFN and UNINA) Tot Cores: 8440 (69 kHS06) Disk Space: 2805 TB Netw connectivity: 20 Gb/s BARI (INFN and UNIBA) Tot Cores: 13000 (130 kHS06) Disk Space: 5000 TB Netw connectivity: 20 Gb/s CATANIA Tot Cores: 3000 (30 kHS06) Disk Space: 1500 TB Netw connectivity: 20 Gb/s COSENZA Tot Cores: 3500 (35 kHS06) Disk Space: 900 TB Netw connectivity: 10 Gb/s

SLIDE 6

INFN is not only HTC!

 While INFN does not deploy large scale HPC centers (like PRACE), it has a long history of HPC R&D and operation

 APE project: late 80s (APE) to ~2005 (ApeNext): in-house developed machines (mostly) for Lattice QCD  ExaNeSt: solutions for Fast Interconnect for HPC  Human Brain Project: Wavescales (brain during sleep / anesthesia)  Imaging in Medical Physics with Big Data algorithms

At the times, APE competitive withthe most powerful machines for LQCD

SLIDE 7

EU Projects (past / present / future)

 Long history of participation in GRID projects

 EDG, EGEE, EGI, EMI, WLCG…

 Evolution from the GRID:

 EGI_Engage  INDIGO-DataCloud  IPCEI-HPC-BDA  EOSCpilot

 Ongoing Projects

 HNSciCloud  West-Life  EGI_Engage  INDIGO-DataCloud  EOSCpilot (INFRADEV-04)  ExaNeST  …

 Under evaluation:

 EOSC-HUB (EINFRA-12)  DEEP-HybridDataCloud (EINFRA-21)  XDC (EINFRA-21)  ICARUS (INFRAIA-02)  SCALE-UP Open DataCloud (ICT-16)

SLIDE 8

Collaboration with other Italian Research Institutes

 INAF (Astrophysics):

 MoU signed for extensive collaboration. Many projects in common (CTA a clear example of Computing-demanding experiment)  Attempt to form a common infrastructure, sharing computing centers

 ASI (Italian Space Agency)

 Mirror Copernicus Project + realization of a national infrastructure for satellite data analysis

 Bari, CNAF Cloud Infrastructures, evaluating Indigo Tools

 CINECA (PRACE Tier-0, consortium of 6 Research Institutes + 70 Universities)

 Realize common infrastructure, with resource sharing and co-location  INFN is acquiring a sizeable share in the upgrade of CINECA HPC Marconi system (already now at #12

n Top500), for its Theoretical Physics use cases

 INFN is planning to acquire a fraction of 2018+ CPU HTC resources @ CINECA, while maintaining the storage in house  In general, CNAF+CINECA, given also the physical vicinity, constitute an example of HPC/HTC integration with beneficial consequences for the whole Italian research system

 CNR,…

SLIDE 9

What is happening around us?

 INFN is part of a large scientific ecosystem, and participate the never-ending process of defining the future modalities of access and provisioning of scientific computing  In Italy, our Tier-1 is close (~10km) to a HPC PRACE Tier-0 (CINECA), and collaboration and integration of facilities is becoming more and more important

 Planning for co-located resources, Tb-level direct connection

 Other Italian HTC realities (although smaller) have resource deployments; attempts of national level integration ongoing

 GARR-X Progress, ENEA, ReCaS, INAF, …  Formerly integrated via Italian Grid Initiative (IGI) – a new attempt at concertation

 WLCG (++?)

 CERN started a process via the Scientific Computing Forum for the evolution of LHC computing for 2025+

 Evolution towards fewer big sites O(10 MW)  Reduction of operation costs  Open to HPC by initial design  Open to commercial procurement by initial design

 HEP Software Foundation

 Preparing a Community (driven) White Paper to serve as baseline for High Energy Physics Scientific Computing in the next decade(s)

 At EU level, incentive towards a common infrastructure research / industry / administration

SLIDE 10

EOSC and IPCEI

 European Open Science Cloud (EOSC)

 Goals:

 Aims to give Europe a global lead in scientific data infrastructures, to ensure that European scientists reap the full benefits of data-driven science.  Develop a trusted, open environment for the scientific community for storing, sharing and re- using scientific data and results  Start by federating existing scientific data infrastructures, today scattered across disciplines and Member States

 EOSCpilot the first related project (INFRADEV-04), aiming to:

 Design and trial a stakeholder-driven governance framework  Contribute to the development of European open science policy and best practice;  Develop demonstrators of integrated services and infrastructures in a number of scientific domains, showcasing interoperability and its benefits;  Engage with a broad range of stakeholders, crossing borders and communities, to build trust and skills

 European Data infrastructure

 Goals:

 Deploy the underpinning super-computing capacity, the fast connectivity and the high-capacity cloud solutions they need

 IPCEI: An Important Project of Common European Interest on HPC and Big Data Enabled Applications

 Signed by Luxembourg, France Italy and Spain (2016)

 Italy as a “system” participates with “The Italian Data Infrastructure” - lead by INFN

SLIDE 11

An example of current/future directions

GPFS TSM

Compute nodes Compute nodes Compute nodes Compute nodes

Tasks

Compute nodes Compute nodes Compute nodes Compute nodes Compute nodes Compute nodes Compute nodes Compute nodes Compute nodes Compute nodes Compute nodes Compute nodes Compute nodes Compute nodes Compute nodes Compute nodes

~700 Km, 20 Gbit/s dedicated L2 link (in production) ~10 Km, O(Tbit/s) dedicated physical link (planned) GPFS TSM

Logical View

SLIDE 12

An example of current/future directions

GPFS TSM

Tasks ~1000 Km, 20 Gbit/s dedicated L2 link (in production) ~10 Km, O(Tbit/s) dedicated physical link (planned)

Compute nodes Compute nodes Compute nodes Compute nodes

In (small) prod In prod For 2018 Under test … (EGI Federated Cloud?)

SLIDE 13

The next decade - challenges for INFN (and the field)

 Some expected heavy computation efforts

 LHC: HL-LHC starting 2026+  CTA: In Operation in the early 2020s  SKA: In operations since somewhere in 2020s  Human Brain Project  …

SLIDE 14

Infrastructure for the next decade - possible plans

 Need to go beyond simple evolution – unless O(10x) larger funding sources identified  New strategies and long-term planning needed

 Infrastructure level: how many sites? How big? Which “Middleware”? Which human effort? In-house or Commercial?  Data is our asset, CPUs can come and go: build a trusted infrastructure for data, allow ephemeral models for CPU access

 in-house, “friendly opportunistic” (other sciences), commercial contracts… all needs to be on equal ground and equally supported by the Middleware  INDIGO a step in this direction, more needed

 Follow opportunities whenever available, even for short time

SLIDE 15

Infrastructure for the next decade - possible plans #2

 INFN has started activities/grants/projects with Commercial Cloud Providers and in general resource owners  R&D is the key to close the gap

 Abandon serial programming, embrace more performing architectures and tools

 GPU, FPGA, Custom ASIC  Big data tools (Map&Reduce, Deep Learning, smart networks …)

 …and training / manpower must follow!

 INFN just opened 12 positions dedicated to Scientific Computing R&D, more hopefully to come!

SLIDE 16

R&D In Technology the COSA project

 Computing On Soc Architecture

 Explore low power architectures (CPU+GPU + eventually FPGA) for scientific computing  Benchmark & port code, from

 HEP: MC code, reconstruction code  Parallel applications @ GPUs

 LQCD, astrophysics, spin-glass, …

 Biology  Deep learning and image recognition

 Developments on fast node interconnect, for use of these systems for HPC (APENet+ derived) 16

SLIDE 17

Conclusions

 Even if INFN primary goal is by no means Computer Sciences, its research activities have become more and more entangled with large scale computing  Next decade challenges cannot be confronted with via a simple evolutionary model; INFN feels deeply involved in the design of next-generation scientific computing model and infrastructure – with all the ecosystem participation!

 Both at the national level, partnering with other Italian Research realities, and international level

 Being agile in resource provisioning, and able to exploit even short term opportunities is a must given future challenges; we need agile software stacks which allow for seamless utilization of “all we can get”

 INDIGO-DataCloud has been a step towards a less closed model!

 R&D is paramount, at wide scale:

 Architectures  (BigData) Tools  Advanced Networks  Novel Data Center solutions  Next Generation Middleware(s)