Skip to Content.
Sympa Menu

usatlas-hllhc-computing-l - Re: [Usatlas-hllhc-computing-l] First meeting of the distributed training working group

usatlas-hllhc-computing-l AT lists.bnl.gov

Subject: US ATLAS HL-LHC computing discussion

List archive

Chronological Thread  
  • From: Torre Wenaus <wenaus AT gmail.com>
  • To: Rob Gardner <rwg AT uchicago.edu>
  • Cc: usatlas-hllhc-computing-l AT lists.bnl.gov
  • Subject: Re: [Usatlas-hllhc-computing-l] First meeting of the distributed training working group
  • Date: Fri, 10 Aug 2018 15:44:13 +0200

Oops, as Doug just noticed, that time conflicts with our new event service/HPC ops meeting.
So we will meet Tue Aug 14 at noon eastern. Sorry for the noise.
  Torre


On Fri, Aug 10, 2018 at 3:26 PM Torre Wenaus <wenaus AT gmail.com> wrote:
We’ll have the first meeting of the distributed training WG on Wed Aug 15 10am eastern time. Thanks to all who responded to the poll. Agenda to follow, suggestions appreciated. We’ll use Vidyo (the ATLAS/CERN standard videoconferencing tool) unless that presents a problem for someone.
  Abid & Torre


On Fri, Aug 10, 2018 at 3:21 PM Torre Wenaus <wenaus AT gmail.com> wrote:
Hi Rob,
Very interesting, thanks! It does sound relevant. For a future meeting you're able to attend...
  Torre


On Fri, Aug 10, 2018 at 1:53 PM Robert Gardner <rwg AT uchicago.edu> wrote:
Hi Torre

Unfortunately I’m on ‘staycation’ until Aug 22.  I keep missing these important meetings!  I did want to mention potentially relevant work we did supporting  this year’s CoDaS-HEP training event at Princeton (http://codas-hep.org).  We (Ilija, Benedikt, Lincoln) built a portal and backend to scale out to CHASE-CI — gpu resources on the Pacific Research Platform:  http://codas.slateci.net/.  There were some lessons learned there about sign-ups (using institutional identity management) and Kubernetes scheduling (we had 60 JupyterLabs running each attached to its own GPU).  We plan to create a version of this to support the new ATLAS analytics platform (for ML to Elasticsearch for ADC analytics), in time for the Oct S&C week.   It looks like this might be a little outside the scope here but perhaps parts could be re-purposed for potential user frontends.

Cheers,
- Rob



On Aug 7, 2018, at 9:42 AM, Torre Wenaus <wenaus AT gmail.com> wrote:

Hi,
We are planning a first meeting of the distributed training working group, one of the working groups defined at last month’s US ATLAS / CSI workshop at BNL. If you’re interested in attending please fill in the doodle:
The distributed training WG is to examine the scaling out of ML training across distributed/parallel resources in order to minimize the turnaround time on network tuning and ML studies. The technical approaches to be looked at include those discussed in the workshop; cf. the talks of Abid (Horovod), Alexei (PanDA), and Amir.  It was agreed at the workshop to define concrete objectives for the WG by the end of September, so that will be the main topic of this meeting.
  Abid & Torre

_______________________________________________
Usatlas-hllhc-computing-l mailing list
Usatlas-hllhc-computing-l AT lists.bnl.gov
https://lists.bnl.gov/mailman/listinfo/usatlas-hllhc-computing-l




Archive powered by MHonArc 2.6.24.

Top of Page