A C++ based MPI-enabled Tasking Framework to Efficiently Parallelize Fast Multipole Methods for Molecular Dynamics

Haensel, David

Dissertation / PhD Thesis

FZJ-2018-06944

A C++ based MPI-enabled Tasking Framework to Efficiently Parallelize Fast Multipole Methods for Molecular Dynamics

Haensel, D. (Corresponding author)FZJ*

2018

xiii, 121 (2018) = Dissertation, TU Dresden, 2018

Please use a persistent id in citations: http://hdl.handle.net/2128/20273 urn:nbn:de:bsz:14-qucosa2-314122 urn:nbn:de:bsz:14-qucosa2-314122

Abstract: Today’s supercomputers gain their performance through a rapidly increasingnumber of cores per node. To tackle issues arising from those developmentsnew parallelization approaches guided by modern software engineering are in-evitable. The concept of task-based parallelization is a promising candidate toovercome many of those challenges. However, for latency-critical applications,like molecular dynamics, available tasking frameworks introduce considerableoverheads. In this work a lightweight task engine for latency-critical applica-tions is proposed. The main contributions of this thesis are a static data-flowdispatcher, a type-driven priority scheduler and an extension for communication-enabled tasks. The dispatcher allows a user-configurable mapping of algorithmicdependencies in the task-engine at compile-time. Resolving these dependenciesat compile-time reduces the run-time overhead. The scheduler enables the pri-oritized execution of a critical path of an algorithm. Additionally, the prioritiesare deduced from the task type at compile-time as well. Furthermore, the afore-mentioned task engine supports inter-node communication via message passing.The provided communication interface drastically simplifies the user interface ofinter-node communication without introducing additional performance penalties.Thisisonlypossiblebydistinguishingtwodeveloperroles–thelibrarydeveloperand the algorithm developer. All proposed components follow a strict guideline toincrease the maintainability for library developers and the usability for algorithmdevelopers. To reach this goal a high level of abstraction and encapsulation isrequired in the software stack. As proof of concept the communication-enabledtask engine is utilized to parallelize the FMM for molecular dynamics.

Note: Dissertation, TU Dresden, 2018

Contributing Institute(s):