Advanced Computing Laboratory (Group Schenk)
 Research
 People
 Projects
 Software
 Book
 Publications
 Teaching
 Student Projects
 Open Positions
 Contact
Research
The research of the Advanced Computing Laboratory is centered around the topic of multicore and manycore algorithms for computational science simulations on emerging highperformance computing (HPC) architectures. To this end, the research connects several relevant subfields of computer science with the needs of computational science and HPC. Typically, we drive research towards extremescale simulations in computational algorithms, application software, programming, and software tools. We are currently involved in several HPC and computational science research and simulation projects that develop methods and applications targeted at the next generation of petaflop/exaflop architectures.
Interdisciplinary cooperation is a key to the work of this group, which functions as a link between various branches of computer science, computing technology, and application areas ranging from applied mathematics, to various branches of the engineering and natural sciences.
Competence Areas
People
 Aryan Eftekhari (PhD Student)
 Radim Janalík (PhD Student)
 Juraj Kardos (PhD Student)
 Thomas Koester (MSc Student, Double Degree Program with FAU Erlangen)
 Drosos Kourounis (Senior Researcher, Lecturer)
 Dimosthenis Pasadakis
 Olaf Schenk (Full Professor)
 Fabio Verbosio (PhD Student)
Projects
Our group develops advanced computational algorithms and mathematical software targeted at extremescale simulations. We are currently involved in HPC and computational science research projects that develop methods and applications targeted at the next generation of petaflop/exaflop architectures. Our research is externally supported by
 European Research Commission (EC)
 Commission for Technology and Innovation (CTI)
 Platform for Advanced Scientific Computing (PASC)
 Swiss National Supercomputing Center
 Swiss National Science Foundation (SNSF)
and by various industrial collaborations. Listed below is a selection of our core research projects.
Research Projects of the Advanced Computing Laboratory
For details, please see our publications.
Software
PARDISO
The package PARDISO is a threadsafe, highperformance, robust, memory efficient and easy to use software for solving large sparse symmetric and unsymmetric linear systems of equations on sharedmemory and distributedmemory multiprocessors.
Map of PARDISO users
BELTISTOS
The package BELTISTOS is a new interior point (IP) software that has been extensively employed for the solution of OPF and MPOPF problems. The BELTISTOS algorithm is demonstrated to provide several orders of magnitude faster solution times than standard optimization software, such as IPOPT, MIPS, and KNITRO, using significantly less memory. BELTISTOS was cocreated by Dr. Drosos Kourounis and Prof. Olaf Schenk in 2017. It is an especially versatile nonlinear solver, using stateoftheart algorithmic solver options such as a new release of PARDISO to accommodate varied objective and constraint nonlinearities in continuous variables. It is designed for largescale problems with up to millions of thousands of variables.
Book
Combinatorial Scientific Computing
Uwe Naumann, Olaf Schenk (Editors)
Publisher: Chapman and Hall/CRC (Dec 15, 2011)
ISBN10: 1439827354, ISBN13: 9781439827352
Combinatorial Scientific Computing explores the latest research on creating algorithms and software tools to solve key combinatorial problems on largescale highperformance computing architectures. It includes contributions from international researchers who are pioneers in designing software and applications for highperformance computing systems. The book offers a stateoftheart overview of the latest research, tool development, and applications. It focuses on load balancing and parallelization on highperformance computers, largescale optimization, algorithmic differentiation of numerical simulation code, sparse matrix software tools, and combinatorial challenges and applications in largescale social networks. The authors unify these seemingly disparate areas through a common set of abstractions and algorithms based on combinatorics, graphs, and hypergraphs. Combinatorial algorithms have long played a crucial enabling role in scientific and engineering computations and their importance continues to grow with the demands of new applications and advanced architectures. By addressing current challenges in the field, this volume sets the stage for the accelerated development and deployment of fundamental enabling technologies in highperformance scientific.
Publications
2018
Refeered Journal Articles
 E. Agullo, P. Arbenz, L. Giraud, O. Schenk
Guest editorial: Special Issue on Parallel Matrix Algorithms and Applications (PMAA’16)
Parallel Computing, pp. 135136, February 2018, DOI: 10.1016/j.parco.2018.01.003
Refeered Conference Articles
 S. Scheidegger, D. Mikushin, F, Kuebler, O. Schenk
Rethinking largescale economic modeling for efficiency: optimizations for GPU and Xeon Phi clusters,
in Proceedings of the 32nd IEEE International Parallel & Distributed Processing Symposium (IPDPS'18), May 2125, 2018, (acceptance rate: 24.5%, 116/461)  C. O Malley, L. Roald, D. Kourounis, O. Schenk, G. Hug
Security Assessment in GasElectric Networks,
in IEEE Xplore Proceedings of the 20th Power Systems Computation Conference. IEEE Xplore Proceedings. PSCC 2018, 20th Power Systems Computation Conference. Dublin, Ireland. June 1115, 2018  O. Conor, G. Hug, D. Kourounis, O. Schenk
Finite Volume Methods for Transient Modeling of Gas Pipelines
in IEEE Proceedings of the 5th IEEE International Energy Conference. 5th IEEE International Energy Conference. Limassol, Cyprus. 37 Jun, 2018
2017
Refeered Journal Articles
 D. Kourounis, A. Fuchs, O. Schenk
Towards the next generation of multiperiod optimal power flow solver
December 2017, IEEE Transactions on Power Systems , DOI: 10.1109/TPWRS.2017.2789187  F. Verbosio, A. De Coninck, D. Kourounis, O. Schenk
Enhancing the Scalability of Selected Inversion Factorization Algorithms in Genomic Prediction
September 2017, Journal of Computational Science, DOI: 10.1016/j.jocs.2017.08.013  M. Rietmann, M. Grote, D. Peter, O. Schenk
Newmark Local Time Stepping on High Performance Computing Architectures
Volume 334, pp. 308–326, April 2017, Journal of Computational Physics, DOI: 10.1016/j.jcp.2016.11.012  C. Lengauer, M. Bolten, R. Falgout, O. Schenk, X. Zhou, L. Zhao
Guest editorial: Special Issue on Advanced StencilCode Engineering
Journal on Concurrency and Computation: Practice and Experience, 2017, Volume 29, Issue 18, DOI: 10.1002/cpe.4142
Refeered Conference Articles
 J. Bloch, O. Schenk
Selected inversion as key to a stable Langevin evolution across the QCD phase boundary
35th International Symposium on Lattice Field Theory (Lattice 2017), DOI: https://arxiv.org/abs/1707.08874  A. Eftekhari, O. Schenk, S. Scheidegger
Parallelized Dimensional Decomposition for Dynamic Stochastic Economic Models
in Proceedings of the ACM Platform for Advanced Scientific Computing Conference, PASC’17, pages 38:1–38:11. June 2017, DOI: 3093172.3093234
2016
Refeered Journal Articles
 A. De Coninck, B. Baets, D. Kourounis, F. Verbosio, O. Schenk, S. Maenhout, J. Fostier
Needles: LargeScale Genomic Prediction with Markerbyenvironment Interaction
January 2016, Journal of Genetics, DOI: 10.1534/genetics.115.179887  D. Kourounis, O. Schenk
Constraint Handling for GradientBased Optimization of Compositional Reservoir Flow, Journal of Computational Geosciences
October 2015, Volume 16(5), pp 1109–1122, DOI: 10.1007/s10596 01595245  P. Arbenz, L. Grigori, R. Krause, O. Schenk
Guest editorial: Special Issue on Parallel Matrix Algorithms and Applications (PMAA’14, Part 2)
Parallel Computing, pp. 135136, August 2016, DOI: 10.1016/j.parco.2016.08.003
Refeered Conference Articles
 L. Riha, T. Brzobohaty, A. Markopoulos, T. Kozubek, O. Schenk, W. Vanroose
Efficient Implementation of FETI Solver for Multi and ManyCore Architectures using Schur Complements
September 2015, Proceedings of the International Conference on High Performance Computing in Science and Engineering, HPCSE2015, Lecture Notes in Computer Science (LNCS), Vol: 9611, Springer, 2016, DOI:10.1007/9783319403618_6
2015
Refeered Journal Articles
 J. Brumm, D. Mikushin, S. Scheidegger, O. Schenk
Scalable HighDimensional Dynamic Stochastic Economic Modeling
Journal of Computational Science, 2015, DOI: 10.1016/j.jocs.2015.07.004  D. Kourounis, O. Schenk
Constraint Handling for GradientBased Optimization of Compositional Reservoir Flow
Journal of Computational Geosciences, October 2015, Volume 16(5), pp 1109–1122, DOI: 10.1007/s1059601595245  C. Lengauer, M. Bolten, R. D. Falgout, O. Schenk, 15161 Abstracts Collection
Advanced StencilCode Engineering
15161, Dagstuhl Seminar Proceedings, pp. 56–75, Schloss Dagstuhl  LeibnizZentrum für Informatik, Germany, 2015. DOI: 10.4230/DagRep.5.4.56
Refeered Conference Articles
 M. Rietmann, M.J. Grote, D. Peter, O. Schenk, B. Ucar
Loadbalanced local time stepping for large scale wave propagation
in Proceedings of the 29th IEEE International Parallel&Distributed Processing Symposium, IPDPS’15 (acceptance rate: 21.8%, 108/496), DOI:10.1109/IPDPS.2015.10  A. De Coninc, D. Kourounis, F. Verbosio, O. Schenk, B. De Baets, S. Maenhout, J. Fostier
Towards Parallel Largescale Genomic Prediction by Coupling Sparse and Dense Matrix Algebra
in Proceedings of the 23rd Euromicro International Conference on Parallel, Distributed, and NetworkBased Processing,2015, DOI: http://dx.doi.org/10.1109/PDP.2015.94
2014
Refeered Journal Articles
 C. Petra, O. Schenk, M. Anitescu
Realtime Stochastic Optimization of Complex Energy Systems on High Performance Computers
IEEE Computing in Science & Engineering  Leadership Computing (Volume:16, Issue: 5), pp. 32  42, DOI: 10.1109/MCSE.2014.53  C. Bekas, A. Grama, O. Schenk, Y. Saad
Parallel Matrix Algorithms (editorial)
Parallel Computing,Volume 40, Issue 7, July 2014, Pages 159–160,http://dx.doi.org/10.1016/j.parco.2014.06.001  D. Mikushin, N. Likhogrud, E. Z. Zhang, C. Bergström
KERNELGEN – the design and implementation of a next generation compiler platform for accelerating numerical models on GPUs
IPDPSW '14, Proceedings of the 2014 IEEE International Parallel & Distributed Processing Symposium Workshops, pp. 10111020, http://dl.acm.org/citation.cfm?id=2672916  M. J. Grote, J. Huber, D. Kourounis, O. Schenk
Inexact InteriorPoint Method for PDEConstrained Nonlinear Optimization
SIAM J. Sci. Comput. 363 (2014),pp. A1251A1276, http://dx.doi.org/10.1137/130921283  C. Petra, O. Schenk, M.Lubin, K. Gärtner
An augmented incomplete factorization approach for computing the Schur complement in stochastic optimization
SIAM J. Sci. Comput 362 (2014), pp. C139C162, http://dx.doi.org/10.1137/130908737  G. Kollias, M. Sathe, O. Schenk, A. Grama
Fast Parallel Algorithms for Graph Similarity and Matching
Journal of Parallel and Distributed Computing, Volume 75, Issue 5, May 2014, pp. 2400–2410, http://dx.doi.org/10.1016/j.jpdc.2013.12.010  D. Kourounis, L.J. Durlofsky, J. D. Jansen, and K. Aziz
Adjoint formulation and constraint handling for gradientbased optimization of compositional reservoir flow
Journal of Computational Geosciences, pp.121, 2014. http://dx.doi.org/10.1007/s1059601393858  P. Arbenz, L. Grigori, R. Krause, O. Schenk
Guest editorial: Special Issue on Parallel Matrix Algorithms and Applications (PMAA’14, Part 1)
Parallel Computing, pp. 99100 (2015) , DOI: 10.1016/j.parco.2015.10.004
Refeered Conference Abstracts
 P. Sanan, S. Schnepp, D. May, O Schenk
Composite solvers for linear saddle point problems arising from the incompressible Stokes equations with highly heterogeneous viscosity structure
AGU Fall Meeting, San Francisco, Dec. 1519, 2014
2013
Refeered Journal Articles
 S. Wagner, M. Sathe, O. Schenk
Optimization for Process Plans in Sheet Metal Forming
The International Journal of Advanced Manufacturing Technology, Springer, DOI: 10.1007/s0017001355157, Dec. 2013.  P. Basini,T. NissenMeyer, L. Boschi, E. Casarotti, J. Verbeke, O. Schenk, D. Giardini
The influence of nonuniform ambient noise on crustal tomography in Europe
Accepted in Journal of Geochemistry, Geophysics, Geosystems (Gcubed)  M. Luisier, O. Schenk
GateStack Engineering in ntype UltraScaled Si Nanowire FieldEffect Transistors
IEEE Transactions on Electron Devices, vol. 60, no 10, pp. 33253329, Oct 2013.
Refeered Conference Articles
 A. Kuzmin, M. Luisier, O. Schenk
Fast Methods for Computing Selected Elements of the Green's Function in Massively Parallel Nanoelectronic Device Simulations
EuroPar 2013 Conference,August 2630, Accepted, in press.  L. Gaudio, M. J. Grote, O. Schenk
Interior point method for timedependent inverse problems
11th International Conference on Mathematical and Numerical Aspects of Wave, WAVE2013, June 37, 2013, Accepted, in press.
2012
Refeered Journal Articles
 M. Sathe, O. Schenk, H. Burkhart
An AuctionBased Weighted Matching Implementation on Massively Parallel Architectures
Parallel Computing 38 (2012), pp. 595614, http://dx.doi.org/10.1016/j.parco.2012.09.001  F. Curtis, J. Huber, O. Schenk, A. Wächter
A Note on the Implementation of an InteriorPoint Algorithm for Nonlinear Optimization with Inexact Step Computations
Mathematical Programming B, pp. 119 (2012), Springer Berlin / Heidelberg, doi: 10.1007/s1010701205574.
Refeered Conference Articles
 M. Rietmann, O. Schenk, P. Messmer, T. NissenMeyer, D. Peter, P. Basini, D. Komatitsch, J. Tromp, L. Boschi, D. Giardini
Forward and Adjoint Simulations of Seismic Wave Propagation on Emerging LargeScale GPU Architectures
ACM/IEEE Supercomputing 2012.  M. Christen, O. Schenk, Y. Cui
PATUS: Parallel AutoTuned Stencils For Scalable Earthquake Simulation Codes
ACM/IEEE Supercomputing 2012.  H. Burkhart, M. Sathe, M. Christen, M. Rietmann, O. Schenk
Run, Stencil, Run, HPC Productivity Studies in the Classroom
6th Conference on Partitioned Global Address Space Programming Models, October 1012, 2012, Santa Barbara, USA.  M. Christen, O. Schenk
PATUS: A Code Generation and AutoTuning Framework For Parallel Stencil Computations
Second International Workshop on Advances in High Performance Computational Earth Sciences: Applications and Frameworks (IHPCES) in conjunction with the International Conference on Computational Science (ICCS 2012), June 46, 2012, Omaha, Nebraska, USA.
2011
Book and Book Chapters
 Combinatorial Scientific Computing
Uwe Naumann, Olaf Schenk (Editor)
Publisher: Chapman and Hall/CRC, ISBN10: 1439827354, ISBN13: 9781439827352  J. Huber, U. Naumann, O. Schenk, A. Wächter
Algorithmic Differentiation and Nonlinear Optimization for an Inverse Medium Problem
chapter in Combinatorial Scientific Computing by U. Nauman and O. Schenk (Editors), pp. 203232, book in the Computational Science series from Chapman and Hall/CRC, ISBN10: 1439827354, ISBN13: 9781439827352.  O. Schenk, M. Sathe, B. Ucar, A. Sameh
Towards A Scalable Hybrid Linear Solver Based On Combinatorial Algorithms
chapter in Combinatorial Scientific Computing by U. Nauman and O. Schenk (Editors), pp. 96127, book in the Computational Science series from Chapman and Hall/CRC, ISBN10: 1439827354, ISBN13: 9781439827352.  O. Schenk, M. Christen, H. Burkhart
Parallel Stencil Computations on Manycore Architectures in Hyperthermia Applications
Scientific Computing with Multicore and Accelerators by D. Bader and J. Dongarra (Editors), Computational Science series from Chapman & Hall / CRC Press, Taylor and Francis Group. pp. 255277, ISBN: 9781439825365, 2011.  O. Schenk, K. Gärtner
Parallel Numerical Linear Algebra
invited book chapter in Encyclopedia of Parallel Computing, D. Padua (Editor), pp. 14581464, 2012, Springer, ISBN 9780387097657.
Refeered Journal Articles
 P. Arbenz, Y. Saad, A. Sameh, O. Schenk
Guest editorial: Special issue on Parallel Matrix Algorithms and Applications (PMAA'10)
Parallel Computing 37 (12): 731732 (2011), doi:10.1016/j.parco.2011.10.011.  M. Christen, O. Schenk, and H. Burkhart
Automatic Code Generation and Tuning for Stencil Kernels on Modern Microarchitecture
Journal Computer Science Research and Development, Proceedings of the International Supercomputing Conference ISC11. Volume 26, pp. 205210, 2011, DOI 10.1007/s0045001101606
Refeered Conference Articles
 F. Curtis, O. Schenk, and W. Waechter
An InteriorPoint Algorithm for LargeScale Nonlinear Optimization with Inexact Step Computations
SIAM J. Sci. Comput. Volume 32, Issue 6, pp. 34473475 (2010)  M. Christen, O. Schenk, and H. Burkhart
Patus: A Code Generation and Autotuning Framework For Parallel Iterative Stencil Computations on Modern Microarchitectures
25th IEEE International Parallel Distributed Processing Symposium, May 1620, 2011. Anchorage (Alaska) USA, in press.
Teaching
Our group is offering core and elective courses within the Bachelor of Informatics, the Master of Computational Science, the Master of Financial Technology and Computing, the Master of Artificial Intelligence, and the Master of Informatics at USI Lugano.
Semester  Course  ECTS  Lecturer (Assistant) 

Spring 2018  Software Atelier: Simulation, Data Science & Supercomputing  6  Schenk (Janalík) 
USICSCS Summer School on Effective High Performance Computing  3  Schenk (Janalík, Kardoš)  
Fall 2017  HighPerformance Computing  6  Schenk (Janalík, Kardoš) 
Software Atelier: Partial Differential Equations  3  Kourounis (Kothari)  
Privatissimum  3  Limongelli, Papadopoulou, Pivkin, Pozzi, Schenk  
Numerical Computing  6  Schenk (Verbosio)  
Spring 2017  Software Atelier: Supercomputing and Simulations  6  Schenk (Janalík) 
USICSCS Summer School on Effective High Performance Computing  3  Schenk (Janalík, Kardoš)  
Fall 2016  HighPerformance Computing  6  Schenk (Kardoš) 
Software Atelier: Partial Differential Equations  3  Kourounis (Kothari)  
Numerical Computing  6  Schenk (Verbosio)  
Spring 2016  Software Atelier: Supercomputing and Simulations  6  Schenk (Janalík) 
USICSCS Summer School on Effective High Performance Computing  3  Schenk (Janalík, Kardoš)  
Numerical Computing  6  Schenk (Verbosio)  
Fall 2015  HighPerformance Computing  6  Schenk (Kardoš) 
Software Atelier: Partial Differential Equations  3  Kourounis (Kothari)  
Numerical Computing  6  Schenk (Verbosio)  
Spring 2015  Software Atelier: Supercomputing and Simulations  6  Schenk 
Introduction to Computational Science  3  Schenk  
Fall 2014  HighPerformance Computing  6  Schenk 
Numerical Computing  6  Schenk (Verbosio)  
PDE Software Lab  3  Kourounis, Krause  
Spring 2014  Special Topics in Informatics and Applied Mathematics and Computational Science  3  Schenk 
Parallel and Distributed Computing Lab  3  Schenk  
Computational Science  6  Schenk  
Fall 2013  Parallel and Distributed Computing  6  Schenk 
Spring 2013  Special Topics in Informatics and Applied Mathematics and Computational Science  3  Schenk 
Parallel and Distributed Computing Lab  3  Schenk  
Computer Simulations in Science and Engineering Summer school  3  Schenk  
Computational Science  6  Schenk  
Fall 2012  Parallel and Distributed Computing  6  Schenk 
Student Projects
If you are a student interested in doing a bachelor's or master's thesis in one of our research areas please do not hesitate to contact us! You may find some suggestions for topics below, but this list is not necessarily exhaustive. The thesis type given can often be changed, so if you are looking for a master's thesis or find a bachelor's thesis topic interesting, do talk to us. Alternatively, if you have a specific interest or have any ideas of your own that are not listed, we are open for discussion. Please note that group work is not offered individually on our website – but can well be arranged on request.
List of open BSc and MSc student projects at the Advanced Computing Laboratory in 2018:
Project  Student  Supervision  Type 

Restricted Maximum LikelihoodMethod for Genomic Prediction 

In genetics, genomic prediction provides a powerful tool for animal and plant breeding. This process aims at the improvement of breed selection, reducing costs. The mathematical models regulating such process are "mixed models," since they include both fixed and random effects, and the equations describing the model are called “mixed model equations” (MMEs). The MMEs are solved in order to compute the maximumlikelihood estimates for the parameters describing the model. What characterizes MMEs in genomic prediction is the presence of a heterogeneous type of coupling structure between the effects. The different environments in which the genome is studied affect the genomic markers in a sparse or dense way, meaning that the matrices describing the MMEs are dense, but present large sparse blocks. This particular structure has led to the development of a sparsedense solver for largescale datasets. This project aims at the investigation of the stateoftheart restricted maximumlikelihood method techniques analyzing the details of the different frameworks currently used. In addition, it will be possible to examine performance, advantages, and limitations of the available software. The computing language used will be MATLAB  while it will be possible to use the statistical language R as well  according to the student preferences. The requirements are knowledge of MATLAB plus the basics of Introduction to Computational Science and Numerical Computing. During the BSc project, you will be working together with the researchers at the ICS and will have the chance to get familiar with the most important libraries used in numerical linear algebra. More information: 
Open 
Fabio Verbosio, Olaf Schenk 

Modern Covariance Matrix Approximation in Finance 

Estimating (inverse) covariance matrices is a ubiquitous task in multivariate analysis that is particularly important in financial applications. An accurate approximation of such matrices becomes notoriously difficult when the number of observations is less than the number of random variables. In such cases, the unbiased sample covariance matrix has a high degree of uncertainty. This research project explores modern techniques that address these challenges. The two research directions we will pursue are "covariance matrix cleaning” and "penalized maximum loglikelihood.” In the first method, we augment the unbiased sample covariance matrix based on some “a priori” (probabilistic) knowledge. The second method approaches the problem by minimizing the L1 regularized likelihood function to recover a sparse approximation of the (inverse) covariance matrix. Using existing frameworks, we will use the approximated (inverse) covariance matrix to solve the optimal meanvariance portfolio of financial assets. The candidate should have working experience with Matlab and have completed coursework in linear algebra and statistics. Additional requirements are knowledge of MATLAB plus the basics of Introduction to Computational Science and Numerical Computing. More information: 
Open 
Aryan Eftekhari, Olaf Schenk 

The Swiss Scientific Social Network 

A social network consists of a set of objects (entities or individuals) connected to each other by social (dyadic) relations. The most common example nowadays is the one given by the social media users on the web. The best way to model social networks is using graphs: such datasets represent the entities as nodes and the relations as edges connecting two different nodes. One very similar structure is the one found in the World Wide Web (WWW), where the web pages represent the nodes of a giant graph having billions of nodes and countless edges—representing the hyperlinks within pages. More than twenty years ago, the PageRank (PR) algorithm was developed in order to rank the different web pages and optimize the search of particular pages (the developers of PR are the founders of Google). PR works purely on the structure of the graph representing the WWW, so its applicability is independent of the nature of the graph itself. The goal of this BSc project is to analyze the social network of the scientific authors belonging to Swiss institutions and to apply the PR algorithm. The first step will be retrieving the information necessary for the construction of the social network based on the relation of coauthorship. After that, it will be the time to model the requested data structures and the implementation and application of the PR algorithm. The results will provide an interesting picture of the different research scenarios in Switzerland and how they interact with each other. The requirements are knowledge of MATLAB plus the basics of Introduction to Computational Science and Numerical Computing. During the BSc project, you will be working together with the researchers at the ICS and will have the chance to get familiar with the most important libraries used in numerical linear algebra. More information: 
Vanessa Braglia 
Fabio Verbosio, Olaf Schenk 

Porting Physical Parameterizations from a Climate Model to Accelerators 

This MSc project involves the new ICON (ICOsahedral Nonhydrostatic) climate and numerical weather prediction model being developed by the Max Planck Institute for Meteorology (MPIM) and the German Weather Service (DWD). The parameterization will first be isolated in a testbed subset of the model, so that subsequent changes can be easily validated. The task will involve porting certain physical parameterizations (which calculate the collective effect of physical phenomena which occur on a scale that is smaller than the underlying numerical grid) to accelerators using the OpenACC programming paradigm. In addition, automatic analysis of loop kernels using the Execution Cache Memory (ECM) model and the Roofline model will be used to validate benchmark results. The finished product would be the validated parameterization running within this testbed framework; however, potentially the work might also include its integration into the overall ICON model. Prerequisites are knowledge of highlevel programming languages and parallel programming. 
Thomas Koester 
Will Sawyer (CSCS), Gerhard Wellein (FAU Erlangen), Olaf Schenk 

AC Power Flow Problem: Search for an Efficient and Robust Solution 

Steadystate power flow (PF) analysis is a basis of power system planning and operation. Since the underlying physical equations that govern flows in power network are nonlinear, a closedform solution of the AC power flow problem is not possible. Therefore, this problem is solved by numerical methods such as NewtonRaphson, which require a good starting point to converge to a solution. This means that if the solution algorithm does not converge, it is generally impossible to tell whether there is indeed no solution or the chosen starting point is not good enough. Such a situation is clearly undesirable because power grids are critical infrastructure and the algorithms used for their analysis and control must be as robust as possible. One potential way of addressing this challenge is the utilization of socalled convex relaxation techniques, which enable the application of fast and reliable solution algorithms to various problems arising in power system operation. The goal of this MSc project is to develop C++ code for solving the PF equations using the standard Newton method. The backbone C++ code will be provided, as well as the complete Matlab implementation as a reference. A series of experiments will be performed to demonstrate the sensitivity of the solution to the initial guess. The second part of the project is to reformulate the PF as a convex problem based on the methods proposed in the literature. The efficiency and robustness of the proposed approach will be analyzed through a number of test cases representing power grids of different sizes. More information: 
Open 
Juraj Kardos, Olaf Schenk 
Master of Science in Computational Science, Master of Science in Informatics 
Covariance Matrix Approximation Using Graphical Models 

Recovering the (inverse) covariance matrix is a fundamental task in the modern multivariate analysis. In this project, we are concerned with a sparse approximation of the (inverse) covariance matrix. This task is challenging due to the inherently high degree of uncertainty in the estimate when the number of samples is limited. A sparse approximation of the covariance matrix is essential in highdimensional settings, and can also have a significant impact on the accuracy of the approximation. In this project, a new method for approximating the sparse structure of the covariance matrix is explored. The algorithm operates by representing the matrix elements as the vertices of an undirected graph. The goal of this project is to implement a parallelized variate of the algorithm which is fit for modern compute architectures. The algorithm will then be coupled with the Sparse Quadratic Inverse Covariance Approximation (SQUIC) code base and tested against synthetic data. Finally, the developed framework is applied to optimization problems in finance (portfolio optimization, regression, etc.). The candidate should have working experience with Matlab, C++, and MPI and have completed coursework in linear algebra and statistics. More information: 
Open 
Juraj Kardos, Olaf Schenk 
Master of Science in Computational Science, Master of Science in Informatics 
GPU implementation of Gadget3 with OpenACC 

The cosmological TreePMMHDSPH Gadget is a highly optimized and fully MPI/OpenMP parallelized code. It is used within various, large scale projects (e.g. Magneticum, www.magneticum.org), typically running on hundreds of thousands of cores. It makes use of the FFTW library to perform the PM part of the gravity solver, where for the local short range part a tree code is used. These routines  as well as the core routines of the hydro solver (based on smoothed particle hydrodynamic method, SPH) – have been started to port to modern systems, which in addition to multicore CPUs, provide GPU accelerators on the individual nodes (like PizDaint at CSCS) of modern architectures. In addition, many of the additional physics modules (like cooling, starformation, chemical networks), which are essential for modern, cosmological applications, would also need to be ported to GPU accelerators. Being purely local processes, they offer the ideal target to the OpenACC programming model. Within this MSc project, the applicant will learn the OpenACC programming model, apply it to a real cosmological application, and will get insight into the complexity of highly parallel, cosmological simulation tools.
The prerequisites for this MSc project are good knowledge of C, some familiarity with GPU programming, and basic knowledge of parallel programming concepts. More information: 
Open 
Claudio Gheller (CSCS), Olaf Schenk 
Master of Science in Computational Science, Master of Science in Informatics 
Scalability of TensorFlow on Piz Daint 

In 2017 the scalability of distributed TensorFlow was investigated at CSCS. The results indicate a good scalability on Daint up to 128 nodes, which seems to be comparable to benchmarks from Google. Starting from 128 nodes there is a huge performance drop. Further, the performance depends strongly on the number of deployed parameter servers. However, there is no insight why the number of parameter servers affects the performance in the way it does. The first goal of the MSc project is to analyze the root cause of these performance related questions. TensorFlow is one of the most popular DL toolkits; however it is one of the least HPCaware toolkit. Other DL toolkits provide better integration of HPC technology such as Caffe2 which uses MPI as a communication layer. The second goal is to study the performance of Caffe2 on PizDaint. A performance analysis is required together with an assessment on the effectiveness of Caffe2 as being more HPC friendly. For instance, does the usage of MPI in Caffe2 justify in term of performance/scalability that we should consider this toolkit a more suitable toolkit than TensorFlow for Piz Daint. A similar performance analysis will be performed as the one already accomplished on TensorFlow. The prerequisites for this MSc project are a good math background, same basic concepts of parallelism, and good knowledge of Python. More information: 
Open 
Claudio Gheller (CSCS), Maxime Martinasso (CSCS), Olaf Schenk 
Master of Science in Computational Science, Master of Science in Informatics 
Security Constrained Optimal Power Flow Problems  A study of Different Optimization Techniques (BSc project, graduated 2018) 

The electrical power grid is a critical infrastructure and, in addition to economic dispatch, the grid operation should be resilient to failures of its components. Increased penetration of renewable energy sources is placing greater stress on the grid, shifting operation of the power grid equipment towards their operational limits. Thus, any unexpected contingency could be critical to the overall operation. Consequently, it is essential to operate the grid with a focus on the security measures. Security constrained optimal power flow imposes additional security constraints on the optimal power flow problem. It aims for minimum adjustments in the base precontingency operating state, such that in the event of any contingency, the postcontingency states will remain secure and within operating limits. For a realistic power network, however, with numerous contingencies considered, the overall problem size becomes intractable for singlecore optimization tools in short time frames for realtime industrial operations, such as rapid resolution of new optimal operating conditions over changing network demands, or realtime electricity market responses to electricity prices. Optimization software becomes a major bottleneck for the energy system modelers. Given that optimization software becomes a major bottleneck, this thesis aims to explore different optimization frameworks; specifically, it will cover the IPOPT and Optizelle optimization frameworks. 
Samuel Adolfo Cruz Alegria (graduated 2018, BSc thesis) 
Juraj Kardos, Olaf Schenk 
Open Positions
Applications
If you are interested in joining our group as a Ph.D. student or postdoc, do not hesitate to contact Olaf Schenk. If possible, please send your application package including the following documents (all as PDFs) to Prof. Olaf Schenk (Email: ):
 Cover letter (including a short explanation of why you want to do your PhD/Postdoc in the Advanced Computing Laboratory)
 CV
 BSc and MSc transcripts
 If applicable, please send a short research statement (write a one page description of a potential project that you would be interested in as a PhD/Postdoc project).
Even if we do not have any imminent openings, some funding possibilities may be under way, or external calls (e.g. for postdoctoral fellowships) may be applicable.
Postdoctoral fellowships
Several calls for postdoctoral/early researcher fellowships are available (bi)annually at national, and EUlevel. In case this is of interest, we encourage you to keep track of these calls and  given mutual interest  coordinate such an endeavor with us well in advance:
 European Union: Funding opportunities
 Swiss National Science Foundation: Careers
All of these fellowships require writing a proposal jointly between applicant and host institution and offer an attractive, independent scheme of funding.
Contact
Address
Prof. Olaf Schenk
Advanced Computing Laboratory
Institute of Computational Science
Universita della Svizzera italina
Via Giuseppe Buffi 13
CH6900 Lugano
Switzerland
Tel.: +41 793682281
Email:
How to find us
Our offices are located on the Central USI campus in the Glass building, second floor. The Advanced Computing Laboratory is part of the Institute of Computational Science which is located at USI Lugano, Via Giuseppe Buffi 13, 6900 Lugano, easily reachable by public transportation.