The ALL Load Balancing Library

 

Abstract

Scalability of parallel applications depends on a number of characteristics, among which is efficient communication, equal distribution of work or efficient data lay-out. Especially for methods based on domain decomposition, as it is standard for, e.g., molecular dynamics, dissipative particle dynamics or particle-in-cell methods, unequal load is to be expected for cases where particles are not distributed homogeneously, different costs of interaction calculations are present or heterogeneous architectures are invoked, to name a few. For these scenarios the code has to decide how to redistribute the work among processes according to a work sharing protocol or to dynamically adjust computational domains, to balance the workload. The A Load Balancing Library (ALL) developed within E-CAM at the Julich Supercomputing Center aims to provide an easy and portable way to include dynamic domain-based load balancing into particle based simulation codes. It provides several schemes to find the ideal split of the workload, from the simplest orthogonal non staggered domain decomposition, to the more fancy Voronoi mesh scheme. Within this text we provide an overview of ALL, its capabilities and current use cases, as well as where to find additional information on the library.

 

Description

Most modern parallelized (classical) particle simulation programs are based on a spatial decomposition method as an underlying parallel algorithm: different processors administrate different spatial regions of the simulation domain and keep track of those particles that are located in their respective region. Processors exchange information

  • in order to compute interactions between particles located on different processors
  • to exchange particles that have moved to a region administered by a different processor.

This implies that the workload of a given processor is very much determined by its number of particles, or, more precisely, by the number of interactions that are to be evaluated within its spatial region.

Certain systems of high physical and practical interest (e.g. condensing fluids) dynamically develop into a state where the distribution of particles becomes spatially inhomogeneous. Unless special care is being taken, this results in a substantially inhomogeneous distribution of the processors’ workload. Since the work usually has to be synchronized between the processors, the runtime is determined by the slowest processor (i.e. the one with the highest workload). In the extreme case, this means that a large fraction of the processors are idle during these waiting times. This problem becomes particularly severe if one aims at strong scaling, where the number of processors is increased at constant problem size: Every processor administrates smaller and smaller regions and therefore inhomogeneities will become more and more pronounced. This will eventually saturate the scalability of a given problem, already at a processor number that is still so small that communication overhead remains negligible.

The solution to this problem is the inclusion of dynamic load balancing techniques. These methods redistribute the workload among the processors, by lowering the load of the most busy cores and enhancing the load of the most idle ones. Fortunately, several successful techniques are known already to put this strategy into practice. Nevertheless, dynamic load balancing that is both efficient and widely applicable implies highly non-trivial coding work. Therefore it has not yet been implemented in a number of important codes. 

The A Load-Balancing Library (ALL) developed within E-CAM at the Simulation Laboratory Molecular Systems of the Juelich Supercomputing Centre, aims to provide an easy and portable way to include dynamic domain-based load balancing into particle based simulation codes. It was created in the context of an Extended Software Development Workshop (ESDW) within E-CAM (see ALL ESDW event details), where code developers of CECAM community codes were invited together with E-CAM postdocs, to work on the implementation of load balancing strategies. The goal of this activity is to increase the scalability of applications to a larger number of cores on HPC systems, for spatially inhomogeneous systems, and thus to reduce the time-to-solution of the applications .

 
Particle system before and after the load balancing. Left: equal domain sizes with bad balance; right: unequal domain sizes and good work load.
 

ALL includes several load-balancing schemes, with additional approaches currently being added. The following list gives an overview about the currently included schemes: 

  1. Tensor-Product method: For the Tensor-Product method, the work on all processes (subdomains) is reduced over the cartesian planes in the systems. This work is then equalized by adjusting the borders of the cartesian planes.
  2. Staggered Grid Method: For the staggered-grid scheme, a 3-step hierarchical approach is applied: work over the Cartesian planes is reduced before the borders of these planes are adjusted; in each of the Cartesian planes the work is reduced for each Cartesian column, these columns are then adjusted to each other to homogenise the work in each column; the work between neighbouring domains in each column is adjusted. Each adjustment is done locally with the neighbouring planes, columns or domains by adjusting the adjacent boundaries.
  3. Unstructured Mesh Method: In contrast to the Tensor-Product method and the Staggered Grid Method, the unstructured mesh method adjusts domains not by moving boundaries but vertices, i.e. corner points, of domains. For each vertex, a force, based on the differences in work of the neighboring domains, is computed and the vertex is shifted in a way to equalize the work between these neighboring domains.
  4. Voronoi Mesh Method: Similar to the topological mesh method (Unstructured Mesh Method), the Voronoi mesh method computes a force, based on work differences. In contrast to the topological mesh method, the force acts on a Voronoi point rather than a vertex, i.e. a point defining a Voronoi cell, which describes the domain. Consequently, the number of neighbors is not a conserved quantity, i.e. the topology may change over time.
  5. Histogram-based Staggered Grid Method: The histogram-based staggered-grid scheme results in the same grid as the staggered-grid scheme (see Staggered Grid Method), this scheme uses the cumulative work function in each of the three cartesian directions in order to generate this grid. Using histograms and the previously defined distribution of process domains in a cartesian grid, this scheme generates in three steps a staggered-grid result, in which the work is distributed as evenly as the resolution of the underlying histogram allows. In contrast to the other schemes this scheme depends on a global exchange of work between processes.

Use cases

ALL is being tested with the HemeLB code[1] from the Centre of Excellence CompBiomed. A recent paper describes how HemeLB’s developments in memory management and load balancing (with ALL) allow near linear scaling performance of the code on hundreds of thousands of computer codes[2]. 

ALL is implemented in the multi-GPU version of DL_MESO_DPD package (see related news item here). The intention of this integration is to allow for better performance when modelling complex systems with DL_MESO_DPD[3], like large proteins or lipid bilayers, redistributing the work load across the GPUs.

 

References

[1] D. Groen, J. Hetherington, H.B. Carver, R.W. Nash, M.O. Bernabeu, and P.V. Coveney. Analysing and modelling the performance of the HemeLB lattice-Boltzmann simulation environment. Journal of Computational Science, 4(5):412 – 422, 2013. doi: https://doi.org/10.1016/j.jocs.2013.03.002. // HemeLB URL: www.hemelb.org

[2] McCullough JWS et al. 2021 Towards blood flow in the virtual human: efficient self-coupling of HemeLB. Interface Focus 11: 20190119. doi: http://dx.doi.org/10.1098/rsfs.2019.0119 

[3] MA Seaton, RL Anderson, S Metz and W Smith, DL_MESO: highly scalable mesoscale simulations, Mol Simul 39 (10), 796–821 (2013) doi: http://dx.doi.org/10.1080/08927022.2013.772297 // https://www.scd.stfc.ac.uk/Pages/DL_MESO.aspx  

Share

December Module of the Month: Load balancing for multi-GPU DL_MESO

 

Description

This module concerns the implementation of the E-CAM Load Balancing Library (ALL) in the multi-GPU version of DL_MESO_DPD code. The intention is to allow for better performance when modelling complex systems with DL_MESO_DPD, like large proteins or lipid bilayers, redistributing the work load across the GPUs.

ALL provides several schemes to find the ideal split of the work load : Tensor-Product method, Staggered Grid Method, Unstructured Mesh Method, Voronoi Mesh Method and Histogram-based Staggered Grid Method. Due to the orthogonal domain decomposition used in DL_MESO, the Tensor-Product scheme was used, which works well for non-staggered orthogonal meshes.

Practical application

A test case was implemented (see Figure 1 a), b) and c)) that reproduces 32k water beads initially scattered along a regular structure and then slowly agglomerating towards an unique large drop confined between two parallel surfaces. The system is divided across 8 GPUs and, for the purposes of the visualisation, we restrict ourselves to 32k particles. For a larger number of particles it would not be possible to simulate the system without load-balancing, since all the particles agglomerate to a subset of the available GPUs and one or more GPUs would run out of memory having to accommodate a large number of particles. Moreover, such a strong load imbalance drastically reduces the scalability of the application.

In Figure d) we see the time history of the load imbalance for each GPU when using the ALL library. Without load balancing the system would gradually diverge from the ideal value of 12.5%. You can find a video that shows the evolution of the load-balancing for this system in another software module.

Figure 1: Load imbalance in DL_MESO with ALL library for a water drop between two surfaces. Each colour represents different domain assigned to a different GPU: a) top view, b) perspective view, c) front view, d) load balance vs time

Source code

Further details on the implementation of ALL library in DL_MESO and the source code can be found in the E-CAM software repository here.

Share

Registration open for Extended Software Development Workshop in HPC for Mesoscale Simulation

 

Few software, like DL_MESO, userMESO and LAMMPS, can currently simulate large Dissipative Particle Dynamics (DPD) simulations. In particular, DL_MESO [12] has recently been ported to multi-GPU architectures and runs efficiently up to 4096 GPUs, an effort supported by E-CAM. 

In this E-CAM Extended Software Development Workshop, the developers of the DL_MESO code themselves will provide an introduction to DPD, DL_MESO, its features and functionalities, as well as they will initiate participants to parallel programming of hybrid CPU-GPU systems. Part of the workshop will be dedicated to theory lectures and hands-on sessions on GPU architectures and OpenACC (NVidia DLI course) given by an NVidia DLI Certified Instructor, followed by the practical case of porting DL_MESO to OpenACC. 

Interested in participating? Join us on the 18-22 January for this ONLINE course. Express your motivation to attend the workshop directly through the CECAM website at https://www.cecam.org/workshop-details/8

References

[1] DL_MESO is a general purpose mesoscopic simulation package developed at Daresbury Laboratory by Dr. Michael Seaton : http://www.cse.clrc.ac.uk/ccg/software/DL_MESO/

[2] M. A. Seaton, R. L. Anderson, S.Metz, and W. Smith, “DL_MESO: highly scalable mesoscale simulations,”Molecular Simulation, vol. 39, no. 10, pp. 796–821, Sep. 2013

Share

A possible roadmap for the coarse graining and multiscale simulation community

 

A community-driven review with contributions from E-CAM “Unfolding the prospects of computational (bio)materials modeling has just been published in the Journal of Chemical Physics on the history, developments, and challenges facing coarse graining (CG) and multiscale simulation (MS)  and a set of recommendations on how the latter may be addressed. 

Continue reading…
Share

New publication is out: “Towards extreme scale dissipative particle dynamics simulations using multiple GPGPUs”

 

E-CAM researchers working at the Hartree Centre – Daresbury Laboratory have co-designed the DL_MESO Mesoscale Simulation package to run on multiple GPUs, and ran for the first time a Dissipative Particle Dynamics simulation of a very large system (1.8 billion particles) on 4096 GPUs.

 

Towards extreme scale dissipative particle dynamics simulations using multiple GPGPUs
J. Castagna, X. Guo, M. Seaton and A. O’Cais
Computer Physics Communications (2020) 107159
DOI: 10.1016/j.cpc.2020.107159 (open access)

Abstract

A multi-GPGPU development for Mesoscale Simulations using the Dissipative Particle Dynamics method is presented. This distributed GPU acceleration development is an extension of the DL_MESO package to MPI+CUDA in order to exploit the computational power of the latest NVIDIA cards on hybrid CPU–GPU architectures. Details about the extensively applicable algorithm implementation and memory coalescing data structures are presented. The key algorithms’ optimizations for the nearest-neighbour list searching of particle pairs for short range forces, exchange of data and overlapping between computation and communications are also given. We have have carried out strong and weak scaling performance analyses with up to 4096 GPUs. A two phase mixture separation test case with 1.8 billion particles has been run on the Piz Daint supercomputer from the Swiss National Supercomputer Center. With CUDA aware MPI, proper GPU affinity, communication and computation overlap optimizations for multi-GPU version, the final optimization results demonstrated more than 94% efficiency for weak scaling and more than 80% efficiency for strong scaling. As far as we know, this is the first report in the literature of DPD simulations being run on this large number of GPUs. The remaining challenges and future work are also discussed at the end of the paper.

Share

Extended Software Development Workshop: Mesoscopic simulation models and High-Performance Computing

[button url=”https://www.e-cam2020.eu/calendar/” target=”_self” color=”primary”]Back to Calendar[/button]

If you are interested in attending this event, please visit the CECAM website here.

Workshop Description

In Discrete Element Methods the equation of motion of large number of particles is numerically integrated to obtain the trajectory of each particle [1]. The collective movement of the particles very often provides the system with unpredictable complex dynamics inaccessible via any mean field approach. Such phenomenology is present for instance in a seemingly simple systems such as the hopper/silo, where intermittent flow accompanied with random clogging occurs [2]. With the development of computing power alongside that of the numerical algorithms it has become possible to simulate such scenarios involving the trajectories of millions of spherical particles for a limited simulation time. Incorporating more complex particle shapes [3] or the influence of the interstitial medium [4] rapidly decrease the accessible range of the number of particles.

Another class of computer simulations having a huge popularity among the science and engineering community is the Computational Fluid Dynamics (CFD). A tractable method for performing such simulations is the family of Lattice Boltzmann Methods (LBMs) [5]. There, instead of directly solving the strongly non-linear Navier-Stokes equations, the discrete Boltzmann equation is solved to simulate the flow of Newtonian or non-Newtonian fluids with the appropriate collision models [6,7]. The method resembles a lot the DEMs as it simulates the the streaming and collision processes across a limited number of intrinsic particles, which evince viscous flow applicable across the greater mass.

As both of the methods have gained popularity in solving engineering problems, and scientists have become more aware of finite size effects, the size and time requirements to simulate practically relevant systems using these methods have escaped beyond the capabilities of even the most modern CPUs [8,9]. Massive parallelization is thus becoming a necessity. This is naturally offered by graphics processing units (GPUs) making them an attractive alternative for running these simulations, which consist of a large number of relatively simple mathematical operations readily implemented in a GPU [8,9].

 

References

[1] P.A. Cundall and O.D.L. Strack, Geotechnique 29, 47–65 (1979).
[2] H. G. Sheldon and D. J. Durian, Granular Matter 6, 579-585 (2010).
[3] A. Khazeni, Z. Mansourpour Powder Tech. 332, 265-278 (2018).
[4] J. Koivisto, M. Korhonen, M. J. Alava, C. P. Ortiz, D. J. Durian, A. Puisto, Soft Matter 13 7657-7664 (2017).
[5] S. Succi,The lattice Boltzmann equation: for fluid dynamics and beyond. Oxford university press, (2001).
[6] L. S. Luo, W. Liao, X. Chen, Y. Peng, W. Zhang, Phys. Rev. E, 83, 056710 (2011).
[7] S. Gabbanelli, G.Drazer, J. Koplik, Phys. Rev. E, 72, 046312 (2005).
[8] N Govender, R. K. Rajamani, S. Kok, D. N. Wilke, Minerals Engin. 79, 152-168 (2015).
[9] P.R. Rinaldi, E. A. Dari, M. J. Vénere, A. Clausse, Simulation Modelling Practice and Theory, 25, 163-171 (2012).

Share

E-CAM Case Study: The development of the GC-AdResS scheme:

from smooth coupling

to a direct interface (abrupt)

Dr. Christian Krekeler, Freie Universität Berlin

Abstract

GC-AdResS is a technique  that speeds up computations without loss of accuracy for key system properties by dividing the simulation box into two or more regions having different levels of resolution, for instance a high resolution region where the molecules of the system are treated at an atomistic level of detail, and other regions where molecules are treated at a coarse grained level, and transition regions where a weighted average of the two resolutions is used. The goal of the E-CAM GC-AdResS pilot project was to eliminate  the need of a transition region so as to significantly improve  performance, and to allow much greater flexibility. For example, the  low resolution region can be a particle reservoir (ranging in detail from coarse grained  to ideal gas particles) and a high resolution atomistic region with no transition region, as was needed hitherto.  The only requirement is that the two regions can exchange particles, and that a corresponding “thermodynamic” force is computed self-consistently, which it turns out is very simple to implement.

Continue reading…
Share

A Conversation on The Fourth Industrial Revolution: Opportunities & Trends for Particle Based Simulation

 

Abstract

In the margins of a recent multiscale simulation workshop a discussion began between a prominent  pharmaceutical industry scientist, and E-CAM and EMMC regarding the unfolding Fourth Industrial Revolution and the role of particle based simulation and statistical methods there.  The impact of simulation  is predicted to become very significant.  This discussion is intended to create awareness of the general public, of how industry 4.0 is initiating in companies, and  how academic research will support that transformation.

Authors: Prof. Pietro Asinari (EMMC and Politecnico di Torino, denoted below as PA) and Dr. Donal MacKernan (E-CAM and University College Dublin, denoted below as  DM) , and a prominent  pharmaceutical industry scientist (name withheld at author’s request as  the view expressed is a personal one, denoted below as  IS)

Continue reading…
Share

Multi-GPU version of DL_MESO_DPD

This module implements the first version of the DL_MESO_DPD Mesoscale Simulation Package, with multiple NVidia Graphical Processing Units (GPUs).

In this module the main framework of a multi-GPU version of the DL_MESO_DPD code has been developed. The exchange of data between GPUs overlaps with the computation of the forces for the internal cells of each partition (a domain decomposition approach based on the MPI parallel version of DL_MESO_DPD has been followed). The current implementation is a proof of concept and relies on slow transfers of data from the GPU to the host and vice-versa. Faster implementations will be explored in future modules.

Future plans include benchmarking of the code with different data transfer implementations other than the current (trivial) GPU-host-GPU transfer mechanism. These are: of Peer To Peer communication within a node, CUDA-aware MPI, and CUDA-aware MPI with Direct Remote Memory Access (DRMA).

Practical application and exploitation of the code

Dissipative Particle Dynamics (DPD) is routinely used in an industrial context to find out the static and dynamic behaviour of soft-matter systems. Examples include colloidal dispersions, emulsions and other amphiphilic systems, polymer solutions, etc. Such materials are being produced or processed in industries like cosmetics, food, pharmaceutics, biomedicine, etc. Porting the method to GPUs is thus inherently useful in order to provide cheaper calculations.

See more information in the industry success story recently reported by E-CAM.

Software documentation and link to the source code can be found in our E-CAM software Library here.

Share

E-CAM State of the Art Workshop: CHALLENGES IN MULTIPHASE FLOWS

We would like to draw your attention to a school cum workshop on

CHALLENGES IN MULTIPHASE FLOWS

that will run on Dec 9-12, 2019, at the Monash University Prato Center,
see http://monash.it/, in Tuscany. The event is an E-CAM state-of-the-art
workshop, and its aim is to focus on computer
simulation methods for multiphase systems and their dynamics, and
their strengths and shortcomings. This is a topic that is relevant in
physics, mathematics, chemistry, and engineering, and we are trying to
bring these communities together for a fruitful exchange. At the same
time, a set of advanced lectures at the school is intended to provide
a solid foundation of background knowledge. For more information (in
particular, the list of Invited Speakers), see the

Main web site for the event

Registration is now open. Regular participants need to pay a fee of
500 Australian Dollars (roughly 300 Euros) for meals etc.; however the
first 25 students (with proven status) who register may attend for free.

DEADLINE for registration and abstract submission is September 22.

Please do not hesitate to contact the organisers (contact information on the main website for the event) if you feel you need more information beyond what is provided on the web.

The Organisers

Burkhard Duenweg, Mainz
Ravi Prakash Jagadeeshan, Melbourne
Ignacio Pagonabarraga, Lausanne

Share