WLCG Data Steering Group

Europe/Zurich
28/R-015 (CERN)

28/R-015

CERN

3
Show room on map
Description

WLCG Data Steering Group

  • Review of Group
    • Mandate
    • Representation and Structure
    • Connection to other groups
      • HSF "Data Access and Management" WG
  • Round table on priorities
    • What strategic issues should the community be addressing, and how?
    • Use of forthcoming preGDB slot
    • Format for WLCG Workshop (mid-June)
  • Next steps

 


Reference
https://espace.cern.ch/WLCG-document-repository/Boards/MB/Minutes/MB-Minutes-160920-v1.pdf
http://indico.cern.ch/event/467577/contributions/1976036/attachments/1339953/2017450/DataMBSummary.pdf

 

Thursday 2 Feb 2017

Present

Alastair Dewhurst
Andrew Hanushevsky
Brian Bockelman
Frank Wuerthwein
Pepe Flix
Philippe Charpentier
Michel Jouvin
Paul Millar
Mario Lasnig
Ian Bird
Dirk Duellmann
Martin Barisits
Latchezar Betev
Giuseppe Lo Presti
Julia Andreeva
Oliver Keeble


    Review of Group
        Mandate
        Representation and Structure
        Connection to other groups
            HSF "Data Access and Management" WG
    Round table on priorities
        What strategic issues should the community be addressing, and how?
        Use of forthcoming preGDB slot
        Format for WLCG Workshop (mid-June)
    Next steps

##

-- Overview

IB - stress that commonalities are essential
  group should focus on strategy, should also contribute to HSF CWP
  ultimately will be part of the process for producing the next computing TDRs

FW - most useful would be exchange forum - tell each other what we're doing
  exchange for prototypes and ideas

AH -overlap of 25% to 100% with HSF WG, depending on how this group views its mandate.
  HSF wg was 25 people or so (big)
  Oriented to hi-lumi
  Trying to get a single vision
  The data access WG doc has been attached to the agenda
  https://docs.google.com/document/d/1wEsX2BegSBEZe3EXwdNR9VWNPrNEOVVTLhu-NtH-77E/edit
  Will be another meeting in May (??) for HSF

LB - Alice and LHCb have shorter timescales
  This group must also be more pragmatic to be relevant for them

No objections to the mandate (see OK's presentation) were raised.

-- Input

- Alice (LB)

Themes of interest;
  EOS and its strategic direction
  Multi-site storage, complexity reduction
  Computing Centre at CERN

DD - there is an EOS demonstrator

LB - looking for sync with other experiments

BB - need to tackle the situation that WLCG will become relatively less important at sites

FW - we have to make fewer assumptions about what people will do for us.

IB - need to be able to exploit more diverse resources
    what's the right level of abstraction?
    minimise requirements on storage providers - how?

MJ - remember caches, we will control most storage
  first priority - focus on data access paradigms
  HSF has data access thing - could simply wait for outcome

- Atlas (MB)

Main themes;
  storage evolution
  use of object stores, access patterns
  beyond x509
  going from files to events, streaming models
  increased use of tape
  network - sharing network intelligence
  SDN maybe?

more collaboration on longer term issues
  eg exploitation of object stores, also as cache and networks
  interested in multi-site storage (but not priority)

ML (joined later)
  Focus on network
  Getting the DM architecture as flexible as possible (eg to incorporate maximum capacity).


- LHCb (PC)

  4x more data after LS2 (so still less than Atlas/CMS today)
  Want to to follow trends
  On ownershiop of resources - must try to keep ownership of storage as long as possible! Specifically, disk storage. Nearline, not so much... could use commercial providers for cold storage.
  Organise storage tech for the resources that we own. Keep diversity low.
  Supports forum for storage future (for storage we control)

- CMS (BB)

  CMS has functional requirements eg protocol support, performance
     hands-off about how storage is run or managed
     May evolve CMS to a model comprising caches with various levels of reliability (eviction at T1 is rare but common at T2 for example)
     Admit to ourselves that storage is not reliable

    beyond that ... layer on top of site storage (eg AAA)
    partner with a variety of small projects

    may need action on the interfaces, e.g. to support caching operation

FW - We see US universities with more compute power than Fermi or BNL, not part of T2 system, we don't pay them, but we would like to benefit from them. For both CPU and storage. Discussion on caches as latency hiders, and adapting workflows to them.


- WLCG Operations (JA)

Shorter term stuff
 Organise storage accounting, continuation of preGDB last year
 Increased use of tape - potential fo commonality here, e.g. in monitoring
 Topology system - describing a storage object is difficult (this is happening in another group)
 Prototypes of impact on operations of these demonstrators

- Xroot (AH)

xrootd is driven by the experiments
Discussion on feasibility of getting experiments to align on a new data delivery model.  DD - caching has a good chance of alignment. CMS is happy, let's look at other cases eg Atlas

- dCache (PM)

Support for multi-VO requirements and adoption of industry standards

- France (MJ)

Work on the data model and sync with HSF

- DPM (OK)
Priorites;
  space reporting
  SRM retirement
  caching

-- Next steps

Need at least another iteration before we can start filling preGDB slots and the like. OK to schedule.

 

There are minutes attached to this event. Show them.
The agenda of this meeting is empty