=Paper=
{{Paper
|id=Vol-1787/530-534-paper-92
|storemode=property
|title=Jupyter extension for creating CAD designs and their sub-sequent analysis by the finite element method
|pdfUrl=https://ceur-ws.org/Vol-1787/530-534-paper-92.pdf
|volume=Vol-1787
|authors=Oleg Iakushkin,Anna Kondratiuk,Olga Sedova,Valery Grishkin
}}
==Jupyter extension for creating CAD designs and their sub-sequent analysis by the finite element method==
https://ceur-ws.org/Vol-1787/530-534-paper-92.pdf
Jupyter extension for creating CAD designs and their
subsequent analysis by the finite element method
O. Iakushkin, A. Kondratiuk, O. Sedova, V. Grishkin
Saint Petersburg State University, 7/9, Universitetskaya emb., Saint Petersburg, 199034, Russia
E-mail: o.yakushkin@spbu.ru
Creating designs in CAD and performing their stress-strain analysis are complex computational
tasks. Their successful solution depends on a number of prerequisites: availability of large computa-
tional power; comprehensive knowledge of physical and mathematical computing; and solid skills of
programming and working in a variety of separate software products that are not integrated to each
other directly.
The paper presents a system aimed at CAD models development and verification from the ground
up. The system integrates geometry construction, mesh model creation and deformation analysis into a
uniform computing environment operating as a SaaS solution. It is based exclusively on open source
software and allows to use the Python programming language and SALOME, GMSH, FEniCS and
ParaView libraries. The system’s architecture and certain issues of working with libraries are dis-
cussed. The paper also presents a browser-based tool for CAD design creation and analysis, which tool
is the front end of the software product we created.
Keywords: CAD, Cloud, Services, IPython
The work was supported by RFBR research project № 16-07-011113 and SPBU projects 0.37.155.2014 and
9.37.157.2014
© 2016 Iakushkin O. O., Kondratiuk A. N., Sedova O. S., Grishkin V. M.
530
Introduction
There are three major reasons prompting a researcher to develop his own FEM solver to analyse
the state of solids. Firstly, an attempt to build a product analogous to SolidWorks or ANSYS. We will
not dwell upon this option, since it is very unlikely to become a success story. Secondly, the software
mentioned might not meet the specific features of a particular problem, impose demanding hardware
requirements, or exceed the budget due to high licensing fees. Thirdly, the choice might be driven by
discrepancy between solutions obtained analytically and through automation [Bogdanov 2015,
Тазиева 2016]. The third case results in software development which can verify a solution and thereby
enhance the quality of research. Therefore, this case appears to be the most well-grounded, both theo-
retically and practically.
New applications analysing the state of solids will specialise in a particular sphere, e.g. in stress
and strain analysis of pipelines under season-specific geological conditions. They will also be distin-
guished by the use of innovative computing technology: cloud resources and coprocessors. These will
allow to optimise the running costs of the software developed. The use of available open libraries on
computational mechanics of a deformed body is also important for such software development. This
paper is focused on the use of libraries providing solution algorithms for solid-state analysis problems
in the form of a cloud service.
The software solutions of both research and commercial nature become increasingly complex and
resource-intensive. The growing number of such solutions adopt the cloud service model of distribu-
tion. This trend is combined by the hardware specialisation: there are systems with GPGPU; with co-
processors, e.g. Xeon Phi and FPGA with OpenCL support; or with specific networking functionality.
The system components often need to be developed using programming languages: С++, С#, .Net,
Python, etc. Furthermore, such components might impose additional system requirements. Cloud plat-
forms allow to create computing networks employing virtual heterogeneous compute nodes. These
nodes run various operation systems and applications.
Today there is no shortage in FEM solvers based on open-source libraries. We have studied some
of FEM solvers most cited in academic literature. They are used both in open-source and commercial
applications: PETSc, FENiCS, and Trilinos. These widely used libraries enable work with mesh ob-
jects, triangulation, and solution of non-linear variational problems. They are also thoroughly de-
scribed in numerous academic works and technical manuals. The libraries are available under free
LGPL licences, which allows their application in both open- and closed-source software. The support
is exercised by such reputable organisations as Argonne National Laboratory, Sandia National Labora-
tories, Simula Research Laboratory, and University of Cambridge.
The libraries allow to develop software using GPU and FPGA coprocessors functionality and to
apply MPI in order to distribute workload between compute nodes. The libraries have a remarkable
feature: the parallel work of CPUs is accomplished through MPI, not streams. This reveals storing ad-
herence of the developers to high-performance computing. The coprocessors can be managed through
a number of interfaces, i.e. OpenCL and CUDA. Coprocessors are mostly employed in solvers dealing
with systems of equations, not with geometrical problems. They are capable of enhancing the compu-
tational speed up to one order of magnitude [Iakushkin 2015, Iakushkin 2014]. The libraries were test-
ed on hyper-elasticity problems. The experiment indicated they can load all cores of CPUs within an
MPI cluster to the full capacity.
The cloud computing is primarily distinguished by its flexibility, which means a large number of
compute nodes can be activated on demand. After the task is completed, the nodes can be deactivated.
Cloud services providers, such as Microsoft Azure and Amazon EC2, do not allow to activate an MPI
cluster in one click: each allocated node must be configured manually. However, the libraries under
discussion require high-performance cluster systems, which challenges their use within a cloud plat-
form. The question arises: how can we activate a cluster of the needed capacity and to launch a solu-
tion of a specific problem? The development of a service enabling a number of concurrent queries is
531
even more problematic. The difficulty pertains to the monitoring of clusters activation and deactiva-
tion; controlling their settings; and storing the results yielded. In other words, any service which em-
ploys one of the libraries mentioned should be mainly focused on solving problems through FEM,
while every single problem might be time-consuming and require the entire capacity of the system.
This is a radical departure from the philosophy of cloud computing, i.e. the on-demand increase of
computational capacity.
The goal of the study was as follows: to automatically activate and set up new nodes; and to unite
them into an MPI cluster. We have built a number of control programmes for Microsoft Azure cloud
application platform which are capable of attaining this goal. In particular, they allow to execute a
programme, which is provided by a user, on the computing resource. After the programme is executed,
the programme complex returns the computation result and deallocates the cloud resources provi-
sioned for the task. The programme complex was tested on Microsoft Azure platform with the finite
element libraries mentioned.
The complex allows to activate the required number of concurrently running MPI clusters in or-
der to solve a particular problem: e.g. problem-solving by multiple solvers or performing computation
with different parameters simultaneously. The integrated solution presented is distinguished by the
possibility of parallel high-performance computing and flexible use of computing resources mobilised
only when needed.
Problem statement
The goal of the study was to automatically deploy and set up new computing nodes and to con-
nect them into a single MPI cluster. We developed a set of control programs to solve these tasks on
Microsoft Azure cloud platform. Specifically, the programs allow to run a user-provided computing
application on the allocated resources. After the application is completed, the software returns the re-
sult to the user and releases the allocated cloud resources. The software complex was tested on Mi-
crosoft Azure using the FEM libraries described in this paper.
Our Implementation
The architecture of software complex is shown in Picture 1. It allows to simultaneously deploy
the required number of MPI clusters to solve tasks. For instance, one task can be solved by different
solvers or the calculation can be performed in parallel using different parameters. The solution de-
scribed here has an important distinguishing feature: it can run high-performance computing tasks in
parallel by using the flexibility of cloud systems and allocating computing resources only when need-
ed.
The described system is aimed at development and verification of CAD models from scratch. It
integrates geometry creation, mesh modelling and deformation analysis into a single computing sys-
tem of “system as a service” type. The system is based solely on open source software and allows to
use the Python language as well as SALOME, GMSH, FEniCS and ParaView libraries to solve the
assigned tasks. The paper describes the system’s architecture and provides an analysis of selected is-
sues related to libraries. It also illustrates the results produced by the developed software—the tool for
creating and analyzing CAD models in a web browser (Picture 2).
532
Picture 1: screenshot of a visualization system demonstrating an object after force was applied to it
Picture 2: Screenshot of a visualization system demonstrating an object after force was applied to it.
533
References
Bogdanov A.V., Degtyarev A.B., Khramushin V.N. High performance computations on hybrid systems:
will "grand challenges" be solved? // Computer Research and Modeling. — 2015. — Vol. 7,
No. 3. — P. 429438 (in Russian).
Iakushkin, Oleg, and Valery Grishkin. "Unification of control in P2P communication middleware:
Towards complex messaging patterns." PROCEEDINGS OF THE INTERNATIONAL CON-
FERENCE ON NUMERICAL ANALYSIS AND APPLIED MATHEMATICS 2014 (ICNAAM-
2014). Vol. 1648. No. 1. AIP Publishing, 2015.
Iakushkin O.O., Degtyarev A.B., Shvemberger S.V. Decomposition of the modeling task of some ob-
jects of archeological research for processing in a distributed computer system // Computer Re-
search and Modeling. — 2014. Vol 7, No. 3. – P. 533-537 (in Russian).
Tazieva L.M., Iakushkin O.O. Creation problems of three-dimensional models and textures from pho-
tographs of the object / Control Processes and Stability 2016, Vol. 3(19), № 1, PP. 527-533 (in
Russian).
534