Annual Concurrency Forum Meeting

4 Feb 2013, 08:30 → 6 Feb 2013, 18:00 US/Central

CDF Big Conference Room (Fermilab)

The Concurrency Forum is an outcome of the "Workshop on Concurrency in the many-Cores Era" held at Fermilab in November 2011 to explore the possibility that interested High Energy Physics (HEP) institutions and projects collaborate on concurrent frameworks and applications R&D. The function of the Forum is to communicate and exchange information and results.

   The goals for the February 2013 annual meeting are the following:

- We would like to discuss progress made with each demonstrator, drawing conclusions for those that have completed their work programme and identifying topics that still need to be investigated.
- It would be nice if each collaboration (experiments, Geant4 etc.) could report on its current views on parallelism issues, such as choice of concurrency model and software technologies. This will be an opportunity for us all to discuss and see if we can converge on a common overall strategy.
- One outcome of the meeting is the possibility of launching development projects with specific deliverables that can be of benefit to the whole community. The extra day can be used by interested groups to get together and to start to discuss a programme of work.

Video Conferencing will be available through Vidyo:
Meeting Access Information: Vidyo instruction
Meeting URL: https://vidyoportal.cern.ch/flex.html?roomdirect.html&key=nQXrefkXnjKV
(Room 10667389)

Monday, 4 February

08:30 → 09:00 Session: Welcome and Meeting Goals

Conveners: Dr Pere Mato (CERN), Dr V. Daniel Elvira (Fermilab)

08:30 Welcome to Fermilab

Speaker: Dr V. Daniel Elvira (Fermilab)

Slides WelcomeConc.pptx

08:40 Meeting Goals

Speaker: Dr Pere Mato (CERN)

Slides ACFM-Goals-20130204.pdf

09:00 → 10:35 Session: Experiments Projects Plans

For each of the experiments (CMS, ATLAS, ALICE, LHCb, LC, Intensity Frontier,etc.)
an update on the requirements and priorities, performance issues, current
plans and milestones. Mainly focusing on the next 2-3 years in a wider scope
than just concurrency projects started in the experiments.

09:00 ATLAS needs and plans

Speaker: Rolf Seuster (TRIUMF)

Slides seuster.pdf

09:15 ALICE concurrency needs and plans

Speaker: Andrei Gheata (CERN)

Slides AliRoot_FNAL.pptx

09:30 CMS Multi-core Application 2013 Update

Speaker: Elizabeth Sexton-Kennedy (FNAL)

Slides ConcurrencyStatusCMS2013.pdf

09:45 Linear Collider - Concurrency Needs and Plans

Speaker: Frank Gaede (DESY)

Slides gaede_linear_collider.pdf

10:00 NOvA needs

Speaker: Dr Andrew Norman (Fermilab)

Slides parallelization_workshop.pdf parallelization_workshop.pptx

10:15 LHCb needs and plans

Speaker: Marco Clemencic (CERN)

Slides talk.pdf

10:35 → 11:00 Coffee Break

11:00 → 12:30 Session: Multi‐threaded Frameworks

This is a technical session with concentration on work that is specific to overall frameworks and moving them towards multi-core, many-core, or other high-performance parallel systems. Areas of interest are the development of multi-threaded HEP data processing frameworks, design issues such as scheduling, thread-safety of toolkits, and performance estimation. We will also discuss migration strategies for the existing algorithmic code and lessons learned from prototypes and demonstrators.

11:00 Multi-threaded Frameowrk Introductory talk

Speaker: Mr Jim Kowalkowski (Fermilab)

Slides ConcurrencyForum2013.pdf ConcurrencyForum2013.pptx

11:10 CMSSW

Speaker: Dr Christopher Jones (Fermilab)

Slides CMS_Concurrent_Workshop_02:2013.pdf

11:25 Art and the intensity frontier

Speaker: Dr Chris Green (FNAL)

Slides art-cfw2013.pdf

11:40 The CERN Concurrency Framework Project (CF4Hep)

Speaker: Dr Benedikt Hegner (CERN)

Slides CF4HepFNAL2013.pdf

11:55 Simulation Vector Prototype

Speaker: Andrei Gheata (CERN)

Slides prototype_FNAL.pptx

12:10 ATLAS Multithreading experiences

Speaker: Wim Lavrijsen (LBNL)

Slides AtlasHive_CW.pdf

12:30 → 14:00 Lunch

14:00 → 15:30 Session: Multi‐threaded Frameworks II

This is a technical session with concentration on work that is specific to overall frameworks and moving them towards multi-core, many-core, or other high-performance parallel systems. Areas of interest are the development of multi-threaded HEP data processing frameworks, design issues such as scheduling, thread-safety of toolkits, and performance estimation. We will also discuss migration strategies for the existing algorithmic code and lessons learned from prototypes and demonstrators.

14:00 Intra-module parallelism

Speaker: Mr Danilo Piparo (CERN)

Slides Intra-algorithmParallelism.pdf

14:15 Trigger integration tools

Speaker: Dr Kurt Biery (Fermilab)

Slides artdaqForConcurrencyForum.pdf

14:30 Parallelization of algorithms with Tbb in the SuperB framework

Speaker: Mr Marco Corvo (CNRS & INFN)

Slides ACFM_04_02_2013.pdf

14:45 Discussion and Conclusions

Slides

• ConcurrencyForum2013Discussion.pptx
• ConcurrencyForum2013Summary.pptx
• discussion_notes.txt

15:30 → 16:00 Coffee Break

16:00 → 17:01 Session: Libraries and Toolkits for a parallel world

Review of libraries and tookits readiness for parallel and vector applications

16:00 Introduction

Slides LibsFNAL2013.pdf

16:05 Parallelization Potential in ROOT Math libraries

Speaker: Mr Lorenzo Moneta (CERN)

Slides Parallel_Math_Concurrency_Meeting_Fermilab2013.pdf

16:21 The Future of Parallelization in ROOT

Speaker: Dr Fons Rademakers (CERN)

Slides ROOT-Concurrency-Workshop-2013.pdf

16:41 The Future of Parallelization in Geant4

Speaker: Dr John Apostolakis (CERN)

Slides Geant4-MT-Parallelization-v6.pptx

Tuesday, 5 February

09:00 → 10:30 Session: Co-­‐processors and Accelerators

Making use of GPUs and accelerators (e.g. MIC) for HEP applications. Design issues (data transfers, synchronization, framework integration, etc.). Performance evaluations. Lessons learned from prototypes and demonstrators. Each presentation should try to give answers to a number of pre-­‐defined questions on relevant topics. Examples: latencies, data transfer and pipelining; data structures design issues; performance gains and expectations; total cost of ownership; risk of vendor lock-­‐in

09:00 Introduction

Speaker: Mr Philippe Canal (FERMILAB)

Slides GPU_sessions-2013.key.pdf

09:05 GPUs in Geant4

Speaker: Soon Yung Jun (Fermilab)

Slides gpu_in_geant4.pdf

09:20 GPUs In CMS

Speaker: Thomas Hauth (CERN)

Slides Hauth_2013-03-Concurrency-FNAL.pdf

09:35 NA62 Trigger

Speaker: Felice Pantaleo (CERN)

Slides pantaleo_GPU_na62_annual_concurrency_forum.pdf

09:50 GooFit: GPU implementation of MINUIT

Speaker: Dr Rolf Andreassen (University of Cincinnati)

Slides GooFitTalk05Feb2013.pdf

10:05 Discussion and Conclusions

10:30 → 11:00 Coffee Break

11:00 → 12:30 Session: Vectorization

Techniques for using efficiently the vector instructions (SIMD) of new processors. Design issues (data structures, etc.). Performance evaluations.
Lessons learned from prototypes and demonstrators. Each presentation should try
to give answers to a number of pre-­‐defined questions on relevant topics.
Examples: data design issues, performance gains vs. effort needed, deployment
strategies, etc.

11:00 Vectorization: Introduction

Speaker: Dr John Apostolakis (CERN)

Slides Vectorization-Session-Setup-v4.pptx

11:05 Vectors: the trends in hardware

Speaker: Mr Andrzej Nowak (CERN openlab)

Slides A.Nowak-Hardware_facilities_for_vector_computing.pdf

11:15 Investigations of Vectorization in ATLAS

With focus on vector libraries and intrinsics

Speakers: Mr Joseph Boudreau (University of Pittsburgh), Roberto Vitillo (LBNL)

Slides vectorization.pdf

11:35 Autovectorisation in CMS: lessons learned, tools and techniques

Speaker: Mr Danilo Piparo (CERN)

Slides AutovectorisationInCMS.pdf

11:55 Preparing simulation for vectors

Speaker: Andrei Gheata (CERN)

Slides vectors_FNAL.pptx

12:05 Discussion: lessons and next steps

12:25 Conclusion

Speaker: Dr John Apostolakis (CERN)

12:30 → 14:00 Lunch

14:00 → 15:30 Session: Technology Watch and Evaluations

An update on the latest products, tools and technologies from vendors. Evaluations of recent hardware on HEP applications.

14:00 A brief correlation study of x86 compiler flags and performance events

Speakers: Mr Andrzej Nowak (CERN openlab), Mirela-Madalina Botezatu (CERN)

Slides A.Nowak-A_brief_correlation_study_of_x86_compiler_flags_and_performance_events.pdf

14:15 Practical results of the Intel MIC / Xeon Phi project at CERN openlab

Speakers: Mr Andrzej Nowak (CERN openlab), Julien Leduc (CERN), Sverre Jarp (CERN)

Slides A.Nowak-Practical_results_of_the_Intel_MIC-Xeon_Phi_project_at_CERN_openlab.pdf

14:30 Accelerating Science with Kepler and CUDA 5

Speaker: Jonathan Bentz (NVidia)

Slides kepler_cuda5

14:45 Programming models for Intel Xeon processors and Intel Xeon Phi coprocessors

Speaker: Scott McMillan (Intel)

Slides Programming_models_for_Intel_Xeon_processors_and_Intel_Xeon_Phi_coprocessors_-_Fermilab.pdf

15:00 ASCR and HEP

Speaker: Dr Robert Lucas (University of Southern California)

Slides FNAL130205_CS.pdf FNAL130205_CS.pptx

15:15 Performance measurement tools (for parallel application)

Speaker: Soon Yung Jun (Fermilab)

Slides performance_tools.pdf

15:30 → 16:00 Coffee Break

16:00 → 17:00 Session: Memory Optimization techniques

Other techniques besides fine-­‐grained parallelism (more than job parallelism)
aiming at the reduction of memory footprint in HEP applications (e.g. KSM, Virtualization, COW, etc.). Performance evaluations. Deployment issues. Lessons learnt from prototypes and demonstrators.

16:00 Kernel-compressed Memory and Zero Pages

Speaker: Jakob Blomer (CERN)

Slides summit.pdf

16:10 AthenaMP. Sharing memory between processes using COW

Speaker: Vakhtang Tsulaia (LBNL)

Slides AthenaMP.pdf

16:25 Managing the Memory Consumption in Parallel Gaudi (GaudiMP)

Speaker: Mrs Nathalie Rauschmayr (CERN)

Slides slides.pdf

16:45 Implications for the Memory Consumption using Geant4-MT

Speaker: Dr John Apostolakis (CERN)

Slides Geant4-MT-Memory-Consumption-v3.pptx

Wednesday, 6 February

09:00 → 10:30 Session: Meeting Summary

Summary with the main conclusions for each of the sessions. Proposals on how
work together as a community. Proposals for concrete collaboration projects.

09:00 Multi-threaded Frameworks

Speaker: Mr Jim Kowalkowski (Fermilab)

Slides

• ConcurrencyForum2013Discussion.pptx
• ConcurrencyForum2013Summary.pptx
• discussion_notes.txt

09:10 Experiments Projects Plans

Speaker: Dr Pere Mato (CERN)

Slides ACFM-Exp-Summary-20130206.pdf

09:20 Libraries and Toolkis for a parallel world

Speaker: Dr Pere Mato (CERN)

Slides workshopsummary-libs.pdf

09:30 Co-processors and Accelerators

Speaker: Mr Philippe Canal (FERMILAB)

Slides GPU_sessions-summary-2013.pdf

09:40 Vectorization

Speaker: Dr John Apostolakis (CERN)

Slides Vectorization-Session-Summary-v2.pptx

09:50 Technology Watch and Evaluations

Speaker: Dr V. Daniel Elvira (Fermilab)

Slides technology-summary-V2.pdf

10:00 Memory Optimization techniques

Speaker: Jakob Blomer (CERN)

Slides Summit.pdf

10:30 → 11:00 Coffee Break

11:00 → 12:30 Session: Informal discussions

12:30 → 14:00 Lunch

14:00 → 15:30 Session III-C

15:30 → 16:00 Coffee Break

16:00 → 17:00 Session III-D