Summary of March GDB, April 11, 2018

Agenda

https://indico.cern.ch/event/651352/

Introduction - I. Collier

presentation

• Mattias: the NorduGrid workshop mentioned in the slides is actually the ARC workshop.

Naples Workshop Summary - I. Collier

presentation

WLCG Strategy & Discussion - I. Bird

presentation

• Mattias: it will not only be necessary to port software to new CPU architectures, but also to new libraries and compilers.
• Romain: do not underestimate the enormous cost of identity management: it should be taken into account early on.
  • Ian B.: even if it is not very visible in the strategy document, we explained the need of a transition to auth mechanisms used elsewhere and there are submitted projects to prototype work on this respect. The document will also get updates in the future.

Naples - Common Data Management - next steps - S. Campana

presentation

• Markus E.: will these working groups take over planning and development from e.g. the dCache project?
  • Ian B.: the dCache team is fully involved in these discussions since the beginning and Patrick agrees with these plans.
• Simone: the replacement of GridFTP with HTTP and WebDAV could be accomplished in the short term between the GDB and WLCG operations coordination, as it is mostly a deployment and configuration issue.
• Andreas: if Rucio is mentioned as a "very promising candidate" as a common solution, does this mean that it has been already decided to adopt it as such?
  • Simone: It is up to CMS to make a decision, which is in the process of evaluating Rucio, as it was presented at the Rucio workshop.
• Markus E.: what are the plans for the experiment DDM systems? Should not the architecture be defined first?
  • Simone: experiments are part of these working groups. This concerns in particular the end-to-end integration working group. We will need some R&D and defining an architecture will be one of the first steps, but understanding the components is a prerequisite.
  • Torre: in this context we will also address the tight relationship between DM and WM
• Alessandro: how can we optimise information sharing among the different working groups? Wouldn't it be better to have a single integration group and a single technology group each with different activities rather then many groups?
  • Ian B.: we know what are the tasks that need to be done, but the organisational details still need to be fully defined
• Maria: during the ongoing O&C week, CMS is discussing how to get organised. We will refine the DM plans as we get input from the experiments.
• Oxana: do you expect commitment from sites for setting up prototypes (without disturbing normal work)?
  • Simone: yes, we will need resources, hardware and people for the prototypes, initially from contingencies at facilities.
  • Ian C.: we have a good record of sites contributing to common work without breaking things, and not all sites will be asked to contribute and at all times.
• Maria: we would like to kick-off these activities in May
• Latchezar: do you plan to run these WGs in parallel to existing structures, or to absorb those? Too many working groups are difficult to manage for smaller collaborations. It should be in the mandate to see how to supersede current activities when there is an overlap.
  • Ian B., Ian C.: we should use current groups whenever possible and create new ones only when needed.

Scientific Computing Forum report - H. Meinhard

presentation

• Latchezar: I see a big overlap between the SCF and the previously discussed groups
  • Ian B.: the SCF is a discussion forum mainly intended for the funding agencies with no decision making involved.

Naples - Performance & Cost Modelling - next steps

presentation

Ian C.: is there enough effort available? Markus: network experts would be welcome. Also from the physics side, we would need someone with some knowledge about HL-LHC parameters. Site managers from small size sites would also be welcome, there is already representation from large sites.

Naples - Analysis Facilities & Use Cases - next steps

presentation

Ian C.: do you see implications for the infrastructures particularly on aspects we are not providing now? Eduardo: SWAN and the infrastructure needed for scaling out interactive analysis is probably the main one. Some functionality for peeking on sparse data is another example. Final remark: we need a set of topical meetings to discuss what different people are doing in the are of end user analysis.

Naples - Workload Management - next steps

presentation

Ian C.: is the pre-gdb a possible place where discussing this? Needs to be favourable also to the US timezone, so afternoon. Simone/Torre/Ian/Ale: there are concrete topics in the DMWM area that we should start talking in working groups (or activities, or tasks): usage of high latency media, usage of caches and workflows in a data lake environment with CPUs non co-located with storage.

Alessandro DG: resource provisioning , aka submit pilots (wrappers) to sites: reality is that we do have more than one per experiment solution to do very similar things. Can we consolidate? (e.g. FTS is the one interacting with storages, can't we have something similar to compute?). To be followed up.

Network Virtualisation WG

presentation

Torre: making SDN End-2-End is a challenge, should we consider making it work at the level of the ³lake² perimeter rather than across the full mesh? Maria: yes, probably this a good approach to start with. Shawn: we can try to be close N2N trying to show advantages for a selected set of sites. Marian: the situation is very different depending on the R&E network provider. Some provide and share more and some less.

-- AndreaSciaba - 2018-04-11