Benefits of open source software for industry

Transcription

1 Benefits of open source software for industry Peter Råback ElmerTeam CSC IT Center for Science CFD-päivä, Hanasaari

2 Outline Free / Open Source software Open Source software for computational engineering Benefits for industry Typical use cases & examples Conclusions

3 Free / Open Source software Definition of free software Software can be used, studied, and modified without restrictions Software can be copied and redistributed in modified or unmodified form either without restriction, or with minimal restrictions only to ensure that further recipients have the same possibility. In English language the word free has two meanings Free as in free beer (suom. Ilmainen) Free as in free speach (suom. vapaa) Free software movement was idealogically rooted whereas current concept of Open Source software is more pragmatic

4 History of open source Free sharing existed long before computers (e.g. recipes) Sharing without particular license IBM s SHARE user group in 50 s and 60 s Early Internet, BBS networks 1983 GNU Project started by Richard Stallman 1985 Free software foundation (FSF) founded Non-free softare seen as unethical 1989 GPL license, 1st version 1991 Linux, Linus Torwalds put the first version on nic.funet.fi 1998 Open Source Initiative (OSI) founded by Eric Raymond & Bruce Perense OS model seen as a superior model for software development Compromise: Free and Open Source Software (FOSS)

5 Open Source Definition 1. Free Redistribution of software 2. Source code, or pointer to it, must be included 3. Derivative works must be allowed 4. May require Integrity of the author's source code 5. No discrimination against persons or groups 6. No discrimination against fields of endeavor 7. Distribution of license; no additional licenses 8. License must not be specific to a product 9. License must not restrict other software 10. License must be technology-neutral Source:

6 Free software is not Freeware (software in binary for available at no cost) No liberty to study or modify the software Shareware (try-before-you-buy software) Additionally temporal & functional limitations Software available for restricted use E.g. free academic use

7 Free software is copyrighted! To clarity the granted rights, restrictions and disclaimers Restrictions are intended To ensure author recognition To preserve future user rights Every written work is automatically copyrighted Legacy stuff from 70 s and 80 s considered public domain Software licences are used to define the details A large number of different licences available

8 License share Artistic License 6 % Others 9 % BSD % MIT 11 % GPL (2.0 or 3.0) 45 % Apache 13 % LGPL (2.1 or 3.0) 9 %

9 Main categories of licences Restrictive licences GNU, LGPL Derived work must carry the same license if published ( viral effect ) Also known as copyleft licenses Permissive licences BSD, MIT, Apache Minimal requirements on how software may be redistributed Some differences among patent rights and author integrity between the three

10 LGPL vs. GPL GNU Lesser General Public Licence (LGPL) allows linking with different licences LGPL/GPL distinction really makes a difference only for libraries In computational engineering most stand-alone software distributed under GPL license Many libraries distributed under LGPL license

11 Why is code published under Open Source? Because it must viral effect of GPL code Egocentric reasons To maximize the impact of the code - and its developers To get appreciation of your peers Improved soft funding opportunities Scientific citations larger h-index Large user base Multinational projects (EU) Increased collaboration opportunities Scientific visits, cross disciplinary collaboration etc. To make the code better Bug reports, more developers Adopted policy in the organization Policy may be enforced by funding organization (USA, France) What is the alternative in-house codes aren t really that credible any more (in science) Survival of the fittest software

12 Commercial logic of publishing under Open Source Commercial markets are heavily consolidated and require large initial investments Community does tasks that should otherwise be mainly done by the vendor Advertisement, porting, verification, contributions to the code OS software may heavily utilize other OS libraries thereby reducing the investment on development Usually money must come from services Much more difficult to create revenues that with software sales Open Source is often a strategy for the under dog No one will publish the best software in its field under Open Source!? Bad for Oracle, good for SQL MariaDB Bad for Ansys Inc., good for OpenFOAM (OpenCFD ESI Group)

13 Open Source software for Computational Engineering

14 CAD OpenCASCADE What is it? Open CASCADE is a powerful CAD/CAM/CAE kernel and development platform for 3D modeling applications. It consists of reusable C++ object libraries and a set of development tools available under OS. Modular structure (see diagram) Devolopment history EUCLID-IS CAD/CAM system 1987 Published under Open Source in 1999 as OpenCASCADE Curstomers CEA, BMW, SAMTECH, EADS, RINA, Alcatel, The only proper CAD library under Open Source?

15 CAD SALOME What is it? Free software that provides a generic platform for Pre and Post-Processing for numerical simulation. Based on a number of free software libraries Qt, OpenCASCADE, Doxygen, Python, VTK Main functions Create/modify, import/export (IGES, STEP), repair/clean CAD models Mesh CAD elements, check mesh quality, import/export mesh (MED, UNV, ASCII) Handle physical properties and quantities attached to geometrical items Perform computation using one or more external solvers (coupling) Display computation results Manage studies (creation, save, reload)

16 Meshing - Netgen What is it? An automatic 2D/3D tetrahedral mesh generator Developed mainly by Joachim Schöberl Key features Accepts input from constructive solid geometry (CSG) or boundary representation (BRep) from STL file format Connection to OpenCASCADE deals with IGES and STEP files Contains modules for mesh optimization and hierarchical mesh refinement LGPL library Netgen library is utilized by a large number of GUI projects

17 CFD - OpenFOAM No 1 CFD software under open source Features Based on C++ modules which are used to build number of solvers Uses finite volume numerics to solve systems of partial differential equations ascribed on any 3D unstructured mesh of polyhedral cells. Comes with models for fluid flows involving chemical reactions, turbulence and heat transfer Includes some rude utilities for pre- and post-processing Fully parallelizable with iterative solvers License under GPL OpenFOAM may be the best example of OS service in CE Started as a PhD project, now owned by ESI Group Many small consultancy companies and major R&D departments base their operation on OpenFOAM

18 FEM freefem++ What is it? One of the 1st free libraries (traces back to MacFEM, 1985) Developed by O. Pironneau, F. Hecht et al. A language dedicated to the finite element method that enables easy solution of Partial Differential Equations (PDE) Idea has been copied and refined (Comsol multiphysics, FEnics etc.) Mainly educational use nowadays

19 FEM library deal.ii What is it? A Finite Element Differential Equations Analysis Library A program library rather than end-user program Computational solution of partial differential equations using adaptive finite elements Uses state-of-the-art programming techniques to offer you a modern interface to the complex data structures and algorithms main aim is to enable rapid development of modern finite element codes Good demonstration of a modern approach taking use of the best available tools

20 FEM library - libmesh What is it Library for the numerical simulation of partial differential equations using arbitrary unstructured discretizations on serial and parallel platforms Provides adaptive mesh refinement computations in parallel libmesh currently supports 1D, 2D, and 3D steady and transient finite element simulations. Makes use of high-quality whenever possible: PETSc, LASPack, SLEPc, Metis, Triangle, Tetgen Active development: Univ. of Texas at Austin, Technische Universität Hamburg, Sandia National Laboratories, NASA Lyndon B. Johnson Space Center

21 FEM - Elmer What is it Multiphysical finite element software under open source Primarily targeted for end-users, but also a library Development started 1995, GPL 2005, LGPL 2012 Features GUI, Solver & Postprocessor All basic element types (1D, 2D, 3D, nodal, edge, face, p, DG) Large number of different physical equations Uses many open source libraries CAD: OpenCASCADE Meshing: Netgen, Tetgen Lin.Alg: Umfpack, MUMPS, Hypre, Lapack, Parpack Visualization: VTK

22 Numerics This area in inherently part of academic developments Many of the best products are published under Open Source Linear algebra for dense matrices Lapack Direct sparse solvers Umfpack, Mumps, Spools, Eigenvalue solvers Arpack, Parpack Ireative solvers, preconditioners Hypre Graph partitioning Metis, Scotch Collections of different tools PETSc, Trilinos

23 Visualization - VTK What Is it? Software system for 3D computer graphics, image processing, and visualization Features Consists of a C++ class library and several interpreted interface layers including Tcl/Tk, Java, and Python. VTK supports a wide variety of visualization algorithms including scalar, vector, tensor, texture, and volumetric methods Supports parallel processing Professional support provided by Kitware Inc. Proper documentation not free Supported by a number of large institutions: Los Alamos an Sandia nat.lab.

24 Visualization - Paraview What Is it? An open-source, multi-platform data analysis and visualization application Developed to analyze extremely large datasets using parallel computing Features Data exploration may be done interactive or using batch processing Can be run on laptops and supercomputers Based on VTK library

25 Visualization - VisIT What is it? Interactive parallel visualization and graphical analysis tool for viewing scientific data on Unix and PC platforms Developed by Department of Energy (DOE) Rather similar in features as Paraview

26 Qt Qt is a cross-platform complete development framework written in C++ High level of abstraction makes coding process very efficient Initially developed by Trolltech -> Nokia -> Digia Used by number of software tools in CE SALOME, Paraview, ElmerGUI,

27 Python Python is a programming language that allows for quick testing and prototyping Python bindings available in many libraries: Qt, SALOME, VTK, Paraview, PetSc, Trilinos,

28 Open source software in computational engineering Academicly rooted stuff is top notch Linear algebra, solver libraries PetSc, Trilinos, OpenFOAM, LibMesh++, CAD and mesh generation not that competitive OpenCASCADE legacy software Mesh generators netgen, tetgen, Gmsh are clearly academic Also for OpenFOAM there is development of commercial preprocessing tools Users may need to build their own workflows from the most suitable tools Also in combination with commerial software Excellent libraries for software development (Qt, python,...)

29 Why does open source dominate in science? Scientists often code but they don t have to sell their results No expectations for commercial spin-off Publication is part of scientific tradition Historically just articles, now also codes and data Many science disciplines must show transparency for the wider public E.g. research on climate change Scientific process is well compatible with the idea open source software Falsification (Popper) requires that the software may be studied and proven wrong It is difficult to falsify a black box (closed source software)

30 Open Source as scientific method Linus Torvalds Kaiken kehitystyön avoimuus on Linuxin toimintaperiaate. Käytännössä kyse on tieteellisestä metodista, jossa avoimen ja kaikille saatavissa olevan tiedon pohjalle rakennetaan uutta sekä sovelletaan sitä tietokoneohjelmiin. Open Sources: Voices from the Open Source Revolution Where scientists talk of replication, Open Source programmers talk of debugging. Where scientists talk of discovering, Open Source programmers talk of creating. Ultimately, the Open Source movement is an extension of the scientific method, because at the heart of the computer industry lies computer science. This profound similarity of open source development and scientific method may be the reason why in science OS dominates

31 Reasons to use open source software in CE free as in beer vs. free as in speech New algorithms New equations Software development Large scale comp. science Open Collaboration Parallelism License costs Free

32 Savings from license costs A common motivation for using OS software As the only reason may result to disapoinment If the software is not previously familiar the learning curve with OS software may be quite long Will the marginal utility of the work with the people doing the analysis be acceptable with OS software? Requires often more versatile IT skills Typically license cost is an issue for smaller company (or team) When the number of parallel licences grow the problem of license costs may become relevant also for bigger companies

33 Benefits of the openness of the code In collaboration all parties have access to the software Companies, universities, consultants, Open source software has more different roles May be used to attract a wider spectrum of actors Also fundamental ideas may be tested with the software Algorithms, models, Compatible with scientific method: falsification More possibilities to built tailored solutions OS codes have usually good extendability & customizability At least some control over the intellectual property Own model development does not become a hostage to vendor lock in Sometimes rules GPL licence out of question

34 Generic benefits of open source software GBDirect Benefits of Using Open Source Software 1. Reliability 2. Stability 3. Auditability 4. Cost 5. Flexibility and Freedom 6. Support and Accountability PCWorld 10 Reasons Open Source Is Good for Business 1. Security 2. Quality 3. Customizability 4. Freedom 5. Flexibility 6. Interoperability 7. Auditability 8. Support Options 9. Cost 10. Try Before You Buy

35

36 Open Source enables Open Innovation Open Innovation uses internal and external knowhow to advance the company technology In Innovation feedback is essential part in selecting the good ideas Agile software projects have ideal feedback events using fast and asynchronous Internet-based communication Code commits Issue tracking Mailing list Discussion forum Compare to the old world of committees, meetings and reports etc.

37 Elmer Infrastructure for Open Research Company B Institute C University D User / Developer / Customer Elmer As Infrastructure Elmer Courses Elmer Support GPL modules propriety modules Elmer Library HPC

38 Usability vs. Extendability Generally commercial tools are much easier to use However there is a caveat GUI (or a closed interface) can never be exhaustive making most of the things simple makes some of the things much harder In open source tools you have basically access to all data and also can often utilize well defined APIs Extending beyond current capabilities is often more realistic & faster Open source software need not to be considered fixed in terms of capabilities, you can always code new stuff

39 Evaluation of open source software There is a huge number of open source software Each software should be analyzed separately Before making large investments to the code one should study Quality is the coding of high quality Usability is the software ok for current needs Extendability can the software be easily enhanced Maturity is the software already mature Activity is the developer and user base active Continuity - is the background organization committed Support can you buy support for the code Documentation is the code well documented

40 Weaknesses of OS software in CE CAD & Meshing There is no process that would bring the best software under open source Lack of standardization Bottom-up type (Bazaar) of open source projects seem fundamentally incompatible with ISO 9001 standard One should perhaps not design buildings using OS software for the computation Big business There are no global service organization for OS software (except maybe for OpenFOAM) The information management of CAD and simulation data is becoming an integral part of the work flow in large businesses and currently OS does not have solutions for that (?)

41 Use Case: Small consultancy company For small company (<5 persons) it may be difficult to generate revenues to pay for commercial licences A small company may create its own tailored workflows Not as easily acceptable in larger companies E.g. use of open APIs and fast prototyping with python A small company does not need to be provide full set of services and can focus on the strengths of the software Organizational structure often quite constant which minimizes the risks involved in learning the software Typically the software is known prior to setting up the company Large number of small consultants for OpenFOAM, but also for other software Code Aster, Code Saturne, Elmer etc.

42 Use Case: R&D intensive small business Company develops internal technology and wants to keep the knowhow in-house Again there may be shortage of money but people are enthusiastic and have good IT skills Tailored workflows for the own technology including optimization etc. From these companies you don t hear much as they have no need to get publicity E.g. Elmer courses have attracted number of small companies throughout Europa with quite specialized simulation needs

43 Example: Open source workflow at Trafotek Cast Resin Transformer (New technology from Trafotek) CAD & meshing with SALOME using python bindings Simulation with Elmer Estimation of heat generation from magnetic losses Coupled heat and N-S equations Postprocessing with Paraview All models and simulations by Eelis Takala, Trafotek, October

44 Use Case: Company with academic collaboration Company does collaboration with the academia and wants to attract good students to solve novel problems Science favors open source Open source software guarantees access and freedom for all parties Typically the work is not part of production but more in the research end of the pipe-line

45 Example: Collaboration in electromagnetics Collaboration between CSC, Aalto, TUT, VTT, ABB, Trafotek etc. initiated to further develop Elmer as a tool for heavy electromagnetics computations. Commercial tools in parts unsatisfactory and very expensive Everybody can participate on their own strengths Development started in 2012 and has been funded by several projects First production runs already succesfully accomplished Model specification Antero Arkkio, Meshing Paavo Rasilo, Aalto Univ. Simulation Juha Ruokolainen, CSC, 2013.

46 Scalability of Whitney element AV solver for end-windings #Procs Time(s) T 2P /T P Magnetic field strength (left) and electric potential (right) of an electrical engine end-windings. Meshing M. Lyly, ABB. Simulation (Cray XC, Sisu) J. Ruokolainen, CSC, 2013.

47 Use Case: Internal simulation tool development Typically a large company that has resources to build tailored internal workflows based on OS software GPL software with viral effect may be a deal breaker Versatile solvers and libraries with well defined APIs provide ideal building blocks for internal tool development Integration with commercial or other in-house tools The solution is typically not distributed outside the company The company wants to be a forerunner and outperform commercial software The company has internal developers and users For the users the usability is made as easy as possible

48 Models for visoucs acoustics Long term collaboration between Nokia and CSC until year moratorium for OS models Development of state-of-the-art models for viscous acoustics Boundary conditions, Mika Malinen, SIAM J. Sci. Comput. 29 (2007), block precondioning, domain decomposition Simulation, Mika Malinen, CSC, 2010

49 Use Case: High performance computing Computations require so many licences that commercial software becomes too expensive It may also be that the commercial software does not scale well enough Open Source research codes often provide superior scaling (Linux & MPI) As the only reason this seldom results into action

50 Conclusions Open Source software can be An ideal tool for collaborative research fostering innovation An environment which enable the study of new things The basis for in-house simulator development Means for attracting talent and building competence Means for saving money, particularly in heavy computations Open Source software probably won t Offer sufficient complete solution for big business Be an ideal step in a rapid development cycle Be as easy to use as the commercial alternatives Offer sufficient standardization for liability-prone business For many companies commercial & open source tools are complementary The nature of the activity defines the optimal solution

51