Summary of November GDB, November 14th, 2018 (CERN)

Agenda

Agenda

Introduction

Next GDBs

• March 2019 at CERN, usual week: not moving, book early
• April 2019: during ISGC, one week earlier than usual (April 3d)

Pre-GDB: authn/authz in December

Next WLCG/HSF/OSF workshop: JLab decided

• Full week of March 18, 2019
• Joined with OSG too

XDC Project Report - Oliver Keeble

slides

WLCG core requirements Also long tail

Caching Data lifecycle Policy driven management QoS

RUCIO being brought in - not original partner

Producing interesting/useful developments to core to WLCG tools

Discussion

Simone: Comment - several solutions described in order to benefit from them we’d need them implemented more widely than the storage solutions in XDC - how will they appear elsewhere. For example if we came to rely on messaging events Q: XDC 14 months but the DOMA activities go longer. Is there a strategy to go beyond A: All the work goes upstream in the products involved. Then it falls in to ‘usual’ WLCG support. This stuff no different from other things WLCG needs.

Stephane: Q: Can we get anything concrete for experiment support now

A: You may not need it for your scenario - read only cache group

Network Virtualisation WG report - Marian Babick

slides

Discussion

Ian C Q: You mention a steep learning curve A: That will continue for some time. Networking & compute people need to talk. May be able to bridge in time. Some work helping people start - youtube tutorial - data centre side should improve - we are still early on. End to end transfers may continue to be

Maarten: Comment - keep in mind you can always make things overly complex. Should keep eye out for low hanging fruit, not always try to do everything that is possible. Experiment & IT people will have to maintain & debug. Marian: in line with identifying concrete use cases

DOMA Third Party Copy WG report - Alessandra Forti

slides

Deployment timeline for sites to provide one non-GridFTP protocol that supports third party copies.

Xrootd and HTTP working groups have different approaches to credentials, HTTP with macarons and xrootd with proxy delegation.

HTTP works with most implementations as released.

Xrootd waits for release 4.9, and also release of specifications to implement in dCache at a later point in time.

Discussion

Q: Are there any performance comparisons?

A: No, not yet

Q: What is the TPC wg relation to the AAI wg with regards to which tokens to use etc, and what implications these ongoing work has for eachother.

A: The conclusion was that the tokens in the TPC wg are low level storage element implementation stuff, and that the WLCG AAI used for the TCP wg is the current production x509 system.

CERN SSO Project - Paolo Tedesco & Hannah Short

slides

Current systems are kerberos and web SSO, group management is problematic. In addition WLCG has a x509 federation.

Future direction is likely some kind of token based authenticaion, probably based on KeyCloak.

The project is in initial phases, hope to deploy alpha versions of services during next year.

Atlas Data Carousel report - Xin Zhao

slides

The project is a study to see how/if ATLAS can use site tape in a more active mode to implement a data carousel system to reduce the need for disk space.

The tests have been to run tests restoring a reasonable large dataset from tape sites to establish baseline numbers.

The first phase has numbers from 0.3-3 GB/s, with a wide range on how many tape drives used (2 - 36). Some sites have worryingly large numbers for number of tape remounts.

ATLAS finds the results better than expected in aggregate throughput. Sites have found it very useful for finding places to tune, fix, etc.

Multiple future improvement work identified for tape software and sites.

After some improvements being done, ATLAS would like to do a near production environment where data gets tranfers to both T1s and T2s, all in parallel at a random time.

-- IanCollier - 2018-11-26