VIP Competence Centre

Monday 15 Mar 2021, 16:00 → 17:00 Europe/Amsterdam

VIP Competence Centre

Participants

• Sorina
• Axel
• Enol
• Andrea

Discussion

VIP is already up and running and available through the EOSC Portal

 

It relays heavily on DIRAC for managing the computing resources (submitting jobs), although these resources are either HTC or managed manually (VM started directly and then added to DIRAC as a CE). Capacity needs are variable, depends a lot on the users coming, at least for the time being there is no need to expand capacity beyond what’s already available with the SLA. GPUs needs expected to increase, but need a  better way of handling them (avoid the current manual setup)

 

Currently using FranceGrilles DIRAC instance which should be merged with the EGI one 

 

Storage is also managed via DIRAC but completely independent from the computing part. VIP is also looking into alternative storage systems that can be used besides DIRAC. First candidate is iRods which is well supported at the FranceGrilles level. 

 

Check-in integration is postponed to 2022, also interested in making it easier to manage the VO (with Comanage?)

 

Software applications included in VIP are curated by VIP and made available as containers via CVMFS. Execution of containers hasn’t been as smooth as expected, with several incompatibilities (SIF vs OCI images, different versions of singularity). Better documentation and support for containers in HTC is needed.

 

FAIR:

• Uses Boutiques (https://boutiques.github.io/) for providing complete description of the application, that can be considered quite FAIR already.
• Data is more tricky and need to be carefully thought. Probably needs also support from the DIRAC point of view. Improving the FAIRness of DIRAC would clearly benefit VIP.
• DIRAC is planning to integrate with RUCIO which could help in this area. Still not there, but change should be quite transparent to clients

 

Next action: create Competence Centre with members:

• VIP
• DIRAC: focus on transition to new instance, dynamic allocation of resources (GPU, Cloud) to meet demand from VIP, storage
• CVMFS: support for containers
• Check-in: support for integration (lower priority at the moment)