Introduction Frédéric Derue, LPNHE Paris - Calcul ATLAS France (CAF) meeting

La page est créée Veronique Coste
 
CONTINUER À LIRE
Introduction Frédéric Derue, LPNHE Paris - Calcul ATLAS France (CAF) meeting
Introduction

                                     Frédéric Derue, LPNHE Paris

                               Calcul ATLAS France (CAF) meeting
                                 CC-IN2P3 Lyon, 9th March 2020

Calcul ATLAS France (CAF) meeting, Introduction, 9th March 2020    1
Introduction Frédéric Derue, LPNHE Paris - Calcul ATLAS France (CAF) meeting
Meetings

      ● Today :
      - CAF on 18th December [indico]
      - next CAF meeting could be after S&C, end of June/beg. of July

      ● Recent meetings :
      - LCG-FR sites, 11-13th December [indico]
      - ATLAS S&C, CERN, 10-14th February [indico]
      - WLCG DOMA meeting, 19th February [indico]
      - LCG-FR Tech, 21st February [indico]
      - LCG-FR CoDir, 6th March [indico]

      ● Next meetings :
      - LCG-FR Tech, 20th March [indico]
      - LCG-FR CoDir, 3rd April [indico]
      - HSF/WLCG workshop, 11-15th May [indico]
      - ATLAS S&C, CERN, 15-19th June [indico]
      - LCG-FR sites, 25-26th May, Strasbourg

Calcul ATLAS France (CAF) meeting, Introduction, 9th March 2020         2
Introduction Frédéric Derue, LPNHE Paris - Calcul ATLAS France (CAF) meeting
ATLAS CPU ressource usage
                                                                     HPC (as grid site)             link
     Slots of running jobs (last 3 months)
                                                     link         Cloud special (BOINC)
                                                                                                             HPC (special)

                                                                        Cloud (mainly
                        DAOD prod                                         Sim@P1)
       Full Sim

                                                                        grid                      Cpu pledge
         User Ana                                  Data reco

   ● Smooth operation
      ○ on the grid, HLT farm, T0, cloud, HPC
      ○ ~400 k jobs per day
      ○ ~66% MC simu, reco, evgen                                                                Data reco
                                                                                          Fast
   ● Analysis                                                                             Sim                       link
      ○ 10-20% share on the Grid/Cloud - not HPC                               User
                                                                                                             DAOD
   ● Large DAOD re-production                                                  Ana
                                                                                                              prod
   ● Above cpu pledge                                                          MC
                                                                               reco
                                                                                           EvGen           Full Sim

Calcul ATLAS France (CAF) meeting, Introduction, 9th March 2020                                                 3
Storage resource usage

                                       ● Storage evolution
                                         ○ usual situation with storage, using full capacity to
                                           limit; 165PB primary, 40PB
                                         ○ majority of data on disk is DAOD and AOD (recent
                                           increases in RAW/AOD from repro)
                                       ● Lifetime model
                                         ○ applied lifetime model deletion on disk in December
                                         ○ applied lifetime model deletion on tape last summer
                                         ○ space taken up again by massive DAOD production
                                         ○ much more frequent DAOD obsoletions planne
                                       ● Data movement (per day)
                                         ○ moving 1-2 PB (1.5M files) @15-20 GB/s
                                         ○ deleting 1.5 PB
Calcul ATLAS France (CAF) meeting, Introduction, 9th March 2020                              4
Network (1/2)
                 Extract of email from Eric Fédé on 13th February to LCGFR-Tech
● Evolution of LHCONE infrastructure by RENATER (early February)
  ○ La connexion Nord (depuis Paris) sur LHCONE est passée à 100Gb/s
    effectif, la liaison Sud ( depuis Lyon) restait bloquée à 40 /30 Gb/s effectif.

● Since few days
  ○ Une liaison dédiée 100Gb/s a été mise en production entre le CC et les sorties
    NORD (Paris) et SUD (Lyon) pour LHCONE
  ○ le CC utilise la sortie SUD. La connection NORD est utilisée comme backup
  ○ Les Tier 2 hors région parisienne vont par la sortie SUD de LHCONE.
  ○ d'autres changements sont prévus sur LHCOPN, LSST etc
  ○ à l'horizon de cet été nous devrions avoir une connectivité sur les réseau LHC
    importante et en adéquation avec nos ambitions.
  ○ un grand merci à l’équipe télécoms du CC et à RENATER pour avoir permis
    ces évolutions

Les cartes ne sont pas encore à jour et le nouveau lien n'apparait pas encore sur la carte
LHCONE de Renater
(https://pasillo.renater.fr/weathermap/weathermap_lhcone_france.html) . Mais en cliquant
sur les liens pour afficher les débits vous constaterez les changements dans les débits
et les échelles.
Calcul ATLAS France (CAF) meeting, Introduction, 9th March 2020                       5
Network (2/2)

● To be followed
  ○ traffic Grille entre CC/LAPP/IP2S et LPC/LPSC/CPPM passait par Paris
    puis mis sur le chamin SUD sur le routeur LHCONE de Lyon entre ces sites.
  ○ suivre la connectivité générale des sites
  ○ GRIF-LPNHE était sorti de LHCONE depuis qqs mois
    → passé à 20 Gb/s avec une sortie sur Paris (mi-février) : pb dans les transferts
       (comme vu l’été dernier)
    → config changée (19 février) pour sortie LHCONE sur Lyon et plus sur Paris
          ⇒ résoud les problèmes mais solution temporaire

Calcul ATLAS France (CAF) meeting, Introduction, 9th March 2020                  6
Other news
 ● Data management policy document at CC
   ○ document was approved by Laurent Serin (IN2P3) & Frédéric Déliot (IRFU)
     and sent to Eric (CC) on 13th February

 ● Machine Learning workshop
   ○ at the previous CAF-user meeting it was discussed about the possibility
     to get - within a year – a dedicated workshop (including hands on)
     for ATLAS France users
   ○ email (should have been) send by group leaders to their own groups
   ○ Laurent Serin discussed with few people to prepare the content
     of such workshop
   ○ which trainings have been followed by members of your groups ?

Calcul ATLAS France (CAF) meeting, Introduction, 9th March 2020                7
Web site

    ● Wiki at CC : https://atlas-france.in2p3.fr
       ○ not nice but convenient
       ○ « physics » and « news » content was never up-to-date
       ○ most of content was migrated to general web site two years ago
       ○ regular difficulties to edit pages (CC user-id/pwd not recognized)
         → ticket opened : seems to be due to an authentification by kerberos

    ● ATLAS-France general web site :
      https://collaborationatlasfrance.web.cern.ch/
       ○ exist for about two years
       ○ maintained by Claire
       ○ would require upgrade to newer version
         → no time for this for Claire
         → at minima, version will be frozen

Calcul ATLAS France (CAF) meeting, Introduction, 9th March 2020                 8
OTP
 It was reported at last ICB that
 ● FR Tier 1 Class 4 at 3.3 FTE looks small wrt other T1s
          → I said that this number has been reviewed and confirmed
 ● FR Tier 2 :
    ○ assignment looks uneven across the T2s
    ○ management view is that PIs should not be getting OTP for the administrative
      overhead of having a grid site.
      ICB agrees if the level is 50%, but does not agree at the level of 10%
         - which is the case in French sites
           → OTP numbers (including PI) to be followed by ICB
           also ICB will try to clarify the definition of tasks

 ● Shortage of some Class 2 shifters (already discussed - DAST and ADCoS)
     From ADCShifts wiki :
       - DAST : Laurent (LAL)
       - ADCoS team : Senior : Claire (LAPP), Mélissa/Sophie (LPNHE), Aresh (CC)
                        Trainee : Gregorio (LPNHE)
       - CRC : Sabine (LPSC) ⇒ do we need (or can we) more from French institutes ?
 ● Class 3
   ○ some contributions of « software » are not S&C taks
   ○ in the tables done after previous CAF, all is mixed
Calcul ATLAS France (CAF) meeting, Introduction, 9th March 2020                 9
Vous pouvez aussi lire