Next-Generation Algorithms and High-Performance Computing

• 
  Would you characterize the SBE&S research in your lab as needing and/or using primarily desktop, teraflop, or petaflop computing resources?
  

• 
  Garching campus as a whole: usage of all kinds/scales of computing resources
  
• 
  Desktop for code development
  
• 
  Small clusters for optimisation/testing of parallel code
  
• 
  Teraflop/petaflop for production runs
  
• 
  Max Planck institutes: clear petaflop needs
  
• 
  TUM
  
• 
  Esp. engineering research: also many scenarios, where the model is still in the fore (materials, optimization, …) and where computational power is not the bottleneck
  
• 
  Algorithm development of math and informatics is more desktop- or cluster-related, the big machines primarily for scalability studies etc.
  

• 
  What are the computational bottlenecks in your simulation problems? What solutions exist now and in the near future, either from a hardware perspective, algorithm perspective, or both?
  

• 
  Problems
  

-- Scalability: multi-core and the resulting massive parallelism (examples: hardly any robust multigrid scheme scaling well beyond 1,000 cores; many eigenvalue solvers not scaling beyond 10,000 cores)

-- Communication/synchronisation for MPP, new parallel programming paradigms

-- Data handling, analysis, exploration, rendering

-- Still many open modeling issues (i. e. the “pre-computing stage”)

-Software (maintenance/re-engineering of older codes, version zoo, lack of reliability due to missing testing culture and verification)

• 
  Solutions
  

-- Hierarchy etc. as general paradigms

• 
  What is the state-of-the-art in your community in discrete event simulations (model size, parallelisation or thread limits, special techniques – time warp, rollback etc.)?
  

• 
  DES is used (by computer engineering groups, for traffic simulation (cf. Ph.D. thesis Srihari Narasimhan)), but, as far as we know, no DES-centered research at present
  

• 
  Is there a formal scientific computing or computational science and engineering graduate program at your institution? Or, if you are in a company, are their educational institutions/programs in your country or region that you find prepare students in formal scientific computing or computational science and graduate engineering effectively?
  

• 
  TUM: two international master’s programs Computational Mechanics and Computational Science & Engineering
  
• 
  TUM & Bavaria: Bavarian Graduate School of Computational Engineering (BGCE) as a Bavaria-wide honours program (“elite program”)
  
• 
  TUM: International Graduate School of Science & Engineering (IGSSE) on Ph.D. level
  
• 
  Germany and neighbours: Working Group of CSE programs in Germany/Austria/Switzerland (München, Stuttgart, Darmstadt, Frankfurt, Aachen, Braunschweig, Rostock, Erlangen, Bochum, Bremen, Hannover, Dresden, Zürich, Basel, Graz) established 2005, will get the legal status of an “eingetragener Verein” in 2008
  
• 
  Undergraduate + graduate programs (Darmstadt, Erlangen)
  
• 
  Meer graduate programs (most)
  
• 
  Doctoral programs (Aachen, Darmstadt, Erlangen, München)
  

• 
  What level of computing expertise do your incoming graduate students or employees arrive with? Do you feel that they have the necessary background to conduct research in SBE&S or do they need extensive preparation after arriving before they can be productive researchers?
  

• 
  Very heterogeneous (in particular to the international programs – even a CSE background does not at all ensure a decent computing background)
  

• 
  What kind of training is available in your organisation for simulation and high performance computing? Is it adequate? Do you believe it will be adequate to address computing on multi-core or petascale architectures? What plans are in place for training programs in next-generation computing?
  

• 
  Tailored courses within specialized CSE programs and mainstreaming programs (informatics, mathematics, engineering, …): all kind of lectures, a broad offer of lab courses (CFD lab, scientific computing lab, performance optimized computing lab, visual computing lab, augmented reality lab)
  
• 
  Additional (specialized) courses at Leibniz Supercomputing Centre (LRZ): parallelisation, …
  
• 
  A very solid basis – definitely the most elaborate in Germany, permanently (and currently) updated to meet future requirements (e.g., activities within the Munich Multi-Core Centre, of a lecture series on future architectural trends co-organized by TUM and IBM)
  
• 
  Summer school courses (Simulation – from Models to Software; Simulation Technology; …), European Advanced Courses (ATHENS program: Parallel Numerical Simulation)
  
• 
  Also a lot of TUM activities on the international scale: establishing Computational Engineering in Belgrade, establishing Computational Science in Tashkent/Uzbekistan, establishing Applied and Computational Physics in St. Petersburg/Russia, planned joint CSE program with National University of Singapore
  

• 
  What fraction of students in your institution study/work in the area of SBE&S, and how has this fraction changed over the past 5—7 years?
  

• 
  No general answer possible: a significant part in maths, a minority in informatics, in engineering the majority on a “use some simulation tool” basis, but a minority really doing CSE work
  
• 
  Increasing numbers
  

• 
  What fraction of graduate students/postdocs in SBE&S come from abroad? From which country do these researchers originate? How many students in your country would you estimate go abroad to do their PhDs in SBE&S?
  

• 
  Hard to give special numbers for simulation; however, simulation is a very internationally oriented field, many of TUM’s partnerships have their origin here
  
• 
  Incoming:
  

-- TUM in general: about 20% students from abroad; higher-level staff (faculty, professors): still rather small, but increasing numbers

• 
  Outgoing:
  

-- Long-term goal of the informatics dept., e. g.: offer all students a one-semester-abroad option

• 
  After completing their PhD in an SBE&S related field, what is the route your students take? Do they take postdoctoral positions in your country or abroad? What fraction eventually achieve permanent jobs in SBE&S? Is this a desired path for students? Do they perceive many job opportunities related to their training in SBE&S? How does industry view students with training in SBE&S?=
  

• 
  International students: all variants (returning for a career at home, aiming at an academic career here, trying to get a job in industry)
  
• 
  German students: main part heading for industry, small part staying at university
  

• 
  What are the roles of academic, government and industrial laboratories in SBE&S research in your country?
  

• 
  Academic: education; research depending on university’s focus
  
• 
  Government: topical research centers (FZJ, FZK, DESY, DKRZ, GSI, …) – most of them with are smaller or larger simulation branch; organised in several associations (Helmholtz, Leibniz, Max Planck)
  
• 
  Supercomputers: operated by all types of institutions (HLRS Stuttgart: university; Jülich: government lab; LRZ München: Bavarian Academy of Sciences; RZG Garching: Max Planck Society; …)
  
• 
  Industry: not that much research in companies; industry-funded labs at universities on special topics (examples: TUM&Siemens – Center for Knowledge Interchange; TUM& Audi: Ingolstadt Institut iniTUM)
  

• 
  Who pays for, and/or is responsible for, the development and sustainability of SBE&S infrastructure (includes long term data storage costs, code maintenance, open-source software etc.)?
  

• 
  Funding mainly from state/government (hardware)
  
• 
  Code maintenance via student/postdoc work – actually hardly happening in a professionally organised way at present (one reason: models and algorithms are funded, simulation software is not)
  
• 
  Responsibility according to IT infrastructure plan: DFG requires an IT strategy (cf. DFG Committee on IT Infrastructure, publishing “Recommendations” every 5 years and evaluating all IT-related purchases by public universities in Germany) and recommends a CIO; depending on this strategy, responsibility is centralised (Computing / IT Services Center) or local
  

• 
  What is the funding situation for SBE&S in your country? E. g., what are the major sources of funding? E. g., over the past 5 years, has funding of SBE&S increased, decreased or remained roughly constant relative to funding of all of science and engineering research? E. g., is most SBE&S research funded as single investigator grants, small term grants, or large teams and/or initiatives? E. g., what is the typical duration over which proposals in SBE&S are funded?
  

• 
  Computer systems: a national plan (by National Science Council / Wissenschaftsrat) with a “pyramid” of tiers (0: future European HPC centers; 1: national HPC centers (Stuttgart, Jülich, München); 2: regional; 3: departmental); funding as for all university buildings and equipment as a joint venture of the federal government and the state; situation is rather good in Germany
  
• 
  Network infrastructure: German Research Network (Deutsches Forschungsnetz, DFN) organising the science network nation-wide, with public funds; situation rather good
  
• 
  Projects on all levels (from individual projects to consortia); the bigger ones are primarily application driven (a research consortium “Fast linear solvers” has close-to-zero chances, a consortium on turbulence modeling and simulation has excellent ones; situation is clearly sub-optimal for cross-sectional topics such as simulation
  
• 
  Terms: not topic-dependent, but depending on the funding program (for example DFG: individual grants 3 years, research units 6 years, SFB 12 years, priority programmes 6 years)
  
• 
  Amount for funding: not that much dedicated money, but an increasing part of the funds for engineering, e.g., are simulation-related.
  

Email: rcg28@cam.ac.uk

Dr. Jonathan Goodman, Senior Lecturer, Unilever Centre

Email: jmg11@cam.ac.uk

Dr. Dmitry Nerukh, Senior Research Associate, Unilever Center

Email: dn232@cam.ac.uk

Dr. Maxim Fedorov, Postdoctoral

Email: mvf22@cam.ac.uk

Dr. Volker Thome, Postdoctoral

Email: vt228@cam.ac.uk

Dr. Hamsa Y. Mussa, Postdoctoral

Email: hym21@cam.ac.uk

Mr. James Bell, PhD student

Email: jcb56@cam.ac.uk

The Unilever Centre for Molecular Informatics was established in 2000 by a large grant from Unilever that also supported the construction of a new building and provided up to 15 years of running costs. The Centre has 4 academic staff members, including the Director, Prof. Robert Glen, and roughly 40 additional students, postdocs, and visitors. The overall mission of the Unilever Centre is to develop new ways of extracting insight into molecular processes such as drug binding from the rapidly increasing amount of data of all types that is becoming available.

All research at the center is academic in nature, with the goal of seeking alignment between academic interests and issues of interest to Unilever. Prof. Glen leads a large group in the general area of molecular informatics, with present interests in molecular similarity and docking, complexity analysis, aspects of drug design, new molecular property calculations, and data analysis. This work spans a wide range of computational methods from data mining to physics-based approaches, and it also includes some wet lab activities. Dr. Jonathan Goodman leads research into understanding organic reactivity and structure using both computational methods (electronic structure calculations) and synthetic approaches. Prof. Murray-Rust leads a research group that is also focused on molecular informatics, with a strong emphasis on data management through the development of the Chemical Markup Language, and other activities revolving around open data and data mining directly from the chemical literature. Dr. Mitchell performs research on enzyme-ligand interactions using both knowledge-based methods and physically based methods.

The Unilever Centre exploits high-performance computing resources at the University of Cambridge (recently upgraded to over 2000 processors), and it employs distributed computing for some projects where many independent calculations are involved.

During the course of the meeting, the following additional points arose in connection with the operation of the Unilever Centre and its relationship to general issues concerning the future of simulation-based engineering and science:

• 
  Funding. Roughly one-third of the Centre funding is provided by Unilever under the operating contract, which is currently in its second 5-year period. Roughly equal fractions also come from government research grants, and contracts from other companies that are not direct competitors of Unilever. The group is limited in size more by available space and facilities than by available funding.
  
• 
  Intellectual property. Under the first 5-year contract, all intellectual property developed in the Centre was owned by the University of Cambridge. This was modified in the second (present) 5-year contract so that IP resulting from research that is fully funded by Unilever is owned by Unilever.
  
• 
  Research roadblocks for adoption of high-performance computation in the pharmaceutical industry. It was discussed that accuracy attainable with molecular dynamics and free-energy simulations using existing molecular mechanics force fields is simply not high enough to be considered predictive. This lack of reliability is a key factor discouraging the pharmaceutical industry from investing in HPC.
  
• 
  Other fundamental questions of importance to the chemical and pharmaceutical industry that cannot at present be satisfactorily addressed include the prediction of solubility, the polymorph into which a molecule crystallizes, as well as the reliable prediction of drug-binding affinities, already mentioned above.
  
• 
  Employment outcomes for graduates of the Unilever Centre: There are good employment prospects for graduates of the Unilever Centre. At the moment approximately 50% go on to industry and 50% remain in universities.
  

Dr. Janette Jones, Discovery Platform Director for Systems Biology

Dr. Massimo Noro, Work Group Leader Modelling

Dr. Peter Olmsted, Professor of Physics, School of Physics, Polymers and Complex Fluids Group, University of Leeds (university-based collaborator)

Dr. Ian Stott, Lead Scientist for Informatics; Physical and Chemical Insights

Dr. Patrick Warren, Scientist, Physical and Chemical Insights

Dr. Simon Watson, Scientist, Physical and Chemical Insights

Dr. Jerry Winter, Scientist, Physical and Chemical Insights

BACKGROUND

Unilever is one of the world's most successful consumer goods companies, with focuses in three businesses: food, home care, and personal care products. Unilever leads the home care market in much of the world, which includes cleansing and hygiene products, with brand names such as Cif, Comfort, Domestos, Persil, and Comfort. Within the personal care market, Unilever is among the global leaders in products for skin cleansing, deodorants, and antiperspirants. It has annual sales of over €40 billion and has 174,000 employees in close to 100 countries.

Unilever devotes approximately 2.5% of its yearly turnover to research and development, making it one of Europe's biggest R&D investors in household care. The research center at Port Sunlight employs over 700 scientists researching home and personal care products. Each year research conducted at Port Sunlight results in over 100 patent filings and approximately 140 peer-reviewed papers and conference presentations. In addition to collaborations with other Unilever R&D centers located throughout the world, researchers at Port Sunlight have collaborations with universities and consultants, both within and outside the UK. In fact, Unilever has been successful in leveraging UK and regional government funding, as well as EU funding, to advance its research goals.

The modeling group at Port Sunlight is headed by Dominic Tildesley, a former chaired professor of chemistry at Southampton (until 1996) and Imperial College (1996-98), who joined Unilever as Head of the Physical Sciences Group in 1998. Tildesley is a world-renowned expert on molecular simulation and coauthor with Mike Allen of the University of Warwick of the seminal text book in the field (Allen and Tildesley 1994). The Unilever group is widely admired as one of the most successful modeling groups in industry. This same group was visited in 1999 as part of the WTEC study on molecular modeling, and it is interesting to note the evolution in the effort between 1999 and 2008. Although the size of the group is comparable, the focus has shifted somewhat, with less emphasis on traditional molecular modeling and an increased emphasis on systems biology and chemical informatics. The group’s shifting interests reflect a growing interest in the development of targeted products that reflect customer characteristics. For example, Unilever has announced a collaboration with the systems biology modeling company Genomatica to “to take advantage of Genomatica’s unique metabolic modeling and simulation technologies to accelerate the development of novel ingredients to improve the effectiveness of Unilever products” (http://www.genomatica.com/news.20060809.html).

The WTEC visiting team viewed presentations by several members of the modeling group, followed by discussions of both technical issues and funding mechanisms.

Skin lipids models are essential to understand the interaction of skin treatments with the outer layers of skin. Unilever has been developing such model systems by approaching the same simplified system through theory, simulations, atomic force microscopy (AFM), and light-scattering techniques. Skin lipids differ from the usual lipid bilayers encountered in biology due to the role played by cholesterol and ceramides; thus, the simulation of self-assembled skin lipids requires additional force fields beyond typical lipid bilayer simulations: the complexity of skin lipids makes their simulation computationally more challenging than typical lipid bilayer simulations. The project has leveraged value from external collaborations, especially at the University of Leeds (Peter Olmsted) and through the SoftComp Network of Excellence (http://www.eu-softcomp.net).

The Softcomp network is an EU Framework 6 (http://cordis.europa.eu/fp6) funded project focused on the intelligent design of functional and nanoscale soft-matter composites; Unilever has made use of the SoftComp Linux cluster for some of its simulations. Unilever has also made use of the White Rose Grid (http://www.wrgrid.org.uk), a regional distributed computational facility based at the Universities of York, Leeds, and Sheffield. Additional funding has been secured through UK science council sources, such as the Engineering and Physical Sciences Research Council (http://www.epsrc.ac.uk), to provide training for the next generation of scientists, and also through Yorkshire Forward (http://www.yorkshire-forward.com), a regional development agency based in Leeds and charged with improving the Yorkshire and Humber economy. The main goal of the Yorkshire Forward funded project was to reinvest into the region in order to increase local wealth through product development and to create and safeguard jobs. Unilever has one of the largest deodorant factories in Europe located in the Yorkshire area.

Systems Biology (Janette Jones and Patrick Warren)

Unilever R&D managers strongly believe that for the complex systems for which they are designing products, a systems biology approach is the only way to develop in-depth comprehensive understanding. Unilever is supportive of initiatives that enable its faster adoption of this paradigm shift in ways of working. This presentation was an introduction to some of the application areas where Unilever researchers see the approach having most impact. One immediate example, given the household cleaning products Unilever is known for, is in the development of new “designer” bacteriacides using insights from bacterial genomes. Unilever has made a substantial commitment to this approach. Outlined in the presentation were the outcomes of the in-house systems biology effort to date and expectations for the future. As part of its systems biology strategy applied to microbial systems, Unilever has developed a collaboration with the systems biology modeling company Genomatica (see introduction above) using the approach of constraint-based modeling to understand biochemical pathways within microorganisms. An interesting insight into the nature of constraint modeling methods was published by the Unilever researchers (Warren and Jones 2007), showing that the dual of the constraint modeling problem has a thermodynamic interpretation.

The finite elements research program at Unilever R&D encompasses computational solid mechanics (CSM) and computational fluid dynamics (CFD). These classical engineering approaches to materials modeling encompasses several length and time scales. Most of the work in these areas is performed using commercial software packages running on PCs, multicore processors, and in some cases, small clusters. The group uses consultants when needed, including consultants from the software vendors. The Unilever internal skills have been applied to a wide range of industrial problems, from processing analysis to product mechanical and flow properties, to in-use product properties and consumer interaction. Several examples were discussed, including powders granulation technology, textile mechanics, and hair multiscale physics behavior.

The group is involved in several collaborations, leveraging funding provided by the UK science and technology councils and by the EU. One EU 6th framework project specifically cited was the nanobiotact project (http://www.nanobiotact.org) in which Unilever is a participant. The goal of this project, as stated on the website, is to “design and construct an articulated artificial finger with a biomimetic sensor based on an array of NEMS force transducers that will mimic the spatial resolution, sensitivity and dynamics of human tactile neural sensors.” At Unilever, such a device would have impact in quantifying the sensory impact of products and correlate product structure with consumer impressions of the feel—e.g., of a lotion/cream, or of a product washed/treated by a particular regimen.

The discussion concluded with some philosophical observations of how to conduct research. Simon Watson suggested that any project involving modeling needs to balance the industrial needs for fast, high-impact results, and create new capabilities in the longer term. His group has embraced the Open Innovation way of working (http://www.openinnovation.net) and they innovate through joint projects with academic and industrial partners (e.g., suppliers or small companies).