HEP-CCE: Scalable IO for Energy and Intensity Frontier Experiments

23 Aug 2018, 07:00 → 24 Aug 2018, 21:05 US/Central

1404, 1405, 1406 & 1407 (Argonne - Bldg 240)

Peter Van Gemmeren (Argonne National Laboratory), Philippe Canal (Fermilab), Samantha Tezak (Argonne National Lab), Taylor Childers (Argonne National Laboratory)

High Energy Physics experiments continue to become more data and simulation intensive. There is an expected factor of ten gap (or more) between projected computing needs for HL-LHC experiments and projected growth of current HEP resources. In the US, High Performance Computing resources are going to be growing by more than an order of magnitude by 2021/2022 with the deployment of the US DOE's first exascale supercomputers. These resources are already becoming an important piece of the HEP computing landscape and will continue to become more important.

One challenge of using leadership computing resources is reading and writing data in scalable ways that does not lead to bottlenecks or performance penalties on the shared filesystems. This Workshop aims to bring together leading IO experts in the HEP field with experts from DOE ASCR Facilities to discuss how to move forward in the next years to make HEP software more friendly to millions of parallel threads accessing files on shared disks.

Argonne Site Map Argonne_Map_(with_closure).pdf

Thursday, 23 August

Thursday Morning Intro

Introduction to HPC IO & HEP IO

1 Brief Introduction to HPC IO

Speaker: Richard Zamora (Argonne)

2 Intro to ROOT IO

Speaker: Mr Philippe Canal (FERMILAB)

Slides 2018-HPC-IO-ROOT.pdf

10:30 Coffee Break

Thursday Afternoon Discussion

3 CMS IO Overview

Speakers: Brian Bockelman, Dr Christopher Jones (Fermilab)

Slides CMSIO.pdf

4 ATLAS IO Overview

Speaker: Dr Peter van Gemmeren (ANL)

5 NERSC Overview

Speaker: Glen Lockwood (NERSC)

6 OLCF Overview

Speaker: Dr Bronson Messer (Oak Ridge National Laboratory)

7 ALCF Overview

Speaker: Harms Kevin

12:20 Working Lunch

14:00 Coffee Break

Thursday Evening Discussion

8 Overview of HEP Activities in IO

Speaker: Jim Kowalkowski (Fermilab)

Slides ScalableIOWorkshop.pdf

9 Discussion

Dinner

We have a reservation for 15 at The Craftsman in Naperville. Please fill out the doodle sent via email if you intend to attend.

Friday, 24 August

Friday Morning Discussion

10 Discussion of ROOT Scalable IO

A discussion of how ROOT could be modified and expanded to support HPC shared IO for HEP experiments. Discussing MPI, OpenMP, TBB, etc. Interfaces required to do this.

Speaker: Mr Philippe Canal (FERMILAB)

10:30 Coffee Break

Friday Afternoon Discussion

11 Discussing Experiment Needs

This session we will discuss the needs of current and future experiments with respect to scalable IO on HPCs. This should follow on from the needs for ROOT, and discuss how experiments can then utilize extensions to take advantage of parallel IO at scale on HPCs.

Speakers: Brian Bockelman, Dr Peter van Gemmeren (ANL)

12:00 Working Lunch

Boxed lunches will be handed out during the last 30 minutes of Peter and Brian's discussion.

12 Compression

This session will be focused on how compression impacts scalable IO in HEP experiments. Most experiments store their data using compression. Can we utilize non-CPU resources to do this? What would the impact be on uncompressing on CPUs if compressed on GPUs? What tools can we use? ROOT? ZLIB? VarInt? ProMC?

Speakers: Brian Bockelman, Dr Peter van Gemmeren (ANL)