Views and Opinions to the Roadmap questions

Size: px

Start display at page:

Download "Views and Opinions to the Roadmap questions"

Gyles Cameron
5 years ago
Views:

1 Dr. Michel d.s. Mesquita, Torleif M. Lunde and Dr. Martin King Bjerknes Centre for Climate Research Allégt. 70, N-5007 Bergen, NORWAY / / Views and Opinions to the Roadmap questions - How does the future use of national einfrastructure look like? How will researchers use einfrastructure and what are the most important applications? The scientific challenges in our group are related to computing power and disk space. It is expected that the resolution for the global models will go from low resolution to medium resolution during the next years. This means the current computing facilities in Norway will be to weak to meet the future demands. To provide information about climate on a local scale there is a need for RCMs (Regional Climate Models). This type of experiments require more computational power than GCMs (Global Climate Models). To resolve more physical processes important for tropical regions (where climate change might have large consequences), there is a strong need for faster computing facilities. Increased resolution will also demand more from the storage side. A doubling of resolution means four times more data. There could be a switch to more widespread use of graphics processing unit accelerated computing. For example, some of the current hardware already provide up to teraflops computing with thousands of cores in a desktop-size machine (costing in the order 1e4 US$ per unit compared with a current supercomputer with equivalent capability costing 1e6 US$). For example see the Nvidia Tesla Personal Supercomputer: - What are the core services that einfrastructure must provide in the future? For the climate community there is a need for: more and faster computational nodes Storage, which reflects what is produced. Storage should preferably be disk based, and should be closely linked to the computational facilities.the current lack of storage means a lot of the storage linked to the computational nodes are not being used as intended, and that data analysis is done in an inefficient way. Example: The runs we are doing for future climate with RCMs produce 100TB of data (for a single run), and takes around 120 days of pure computations on 256 nodes. In future the amount of data/need for CPUhours will increase. A good computational node dedicated for post processing. This node should have ample of RAM and best possible bandwidth to the storage facilities Explore the possibilities to switch to more widespread use of graphics processing unit accelerated computing - How does the future national einfrastructure relate to Nordic and European einfrastructure? (Globalisation of einfrastructure) The planning of the future national einfrastructure should take account plans within other Nordic and European countries. This will mean that there will be a need for hardware and software support, which is more scattered, rather than concentrated on centralised machines like today. During the transition to GPGPU computing, researchers may also need help in code development and hardware setup. Even big computers can benefit from GPU: julyaugust-2008/los-alamos-national-lab-and-ibmbring-computing-into-the-petascale-era/print.html In climate research, it makes sense to move to GPGPU. It s the next step of HPC, at least until more powerful hardware are invented.

2 Espen Uleberg Input - About einfrastructure and archaeology. The University museums in Norway have worked towards shared data solutions since 1991/2, first in projects, and today MUSIT (museum IT) continues this work. MUSIT is a collaborative initiative aimed at managing and disseminating digitized museum collections. The initiative is funded by the universities of Bergen, Oslo, Stavanger, Tromsø and Trondheim and their affiliated museums. Bergen Museum started to digitize the archaeological museum archives in the 1990 ies. The NTNU Museum of Natural History and Archaeology has also started to scan its archives. The museums are now planning an einfrastructure project - DDAMA (Digitization and Dissemination of the Archaeological Museum Archives of the University Museums in Norway). Generally the current work in MUSIT and especially the DDAMA project is within area of interest of DARIAH (The Digital Research Infrastructure for the Arts and Humanities). The Norwegian university museums through the Museum Project took part in ARENA (Archaeological Records of Europe - Networked Access). ARENA presented queries in the sites and monuments records and other archival material on the internet in The Archaeological Data Service in York, UK, coordinates now a second phase - ARENA 2 - under the auspices of the DARIAH project. New techniques for analysing high resolution satellite images, LIDAR data and data from GPR (GroundPenetratingRadar) give new possibilities for detecting sites and monuments, and are large datasets. Electronic documentation of sites and monuments and archaeological field work is generating data that should be available through an einfrastructure. During excavations, precise coordinates for finds and structures are registered facilitating the use GIS for analysis and presentation of the data. 3D scanning has been used to document ships during excavation as well as sites and monuments, and the scanning of the Oseberg vikingship is an example of 3D-documentation of large objects in the museum collections. Together with the electronic images taken on site and of objects this is creating quite large datasets. An einfrastructure for the university museum as creates through MUSIT will give new possibilities for nationwide archaeological research. The knowledge in Norway can contribute to the work done in DARIAH. The large artefact collections can also become part of the EUROPEANA project. The two are closely related, but one could say that DARIAH will concentrate on structures, and how to create the einfrastructure while EUROPEANA concentrates on presentations and using einfrastructure when information is made available.

3 Arne Løkketangen Professor Molde University College, Specialized University in Logistics, 9 mars 2011 einfrastructure Use Roadmap The report The scientific case for einfrastructure in Norway is quite impressive in its detailed description of scientific use of computing power. I would like to mention two different application areas that I think are increasingly important in this context, and where I am personally involved. The first area, that is also my main research area, is in developing and applying methodologies for solving hard real-world combinatorial optimization problems, applied to the supply chain. The term combinatorial means that some of the problem variables have discrete values, usually representing a choice. The problems we want to solve are in the class of NP-hard combinatorial optimization problems. Well-known examples are the Travelling Salesman Problem, TSP, and the Knapsack Problem, in many variants. These types of problems are found very frequently in a supply chain, and involve production and distribution among other things. Many of these problems are too difficult for exact solution methods to find optimal solutions in reasonable time, so in practice one often has to settle for good solutions, and not the provable optimum. These methods usually have many parameters and need a lot of tuning. In order to get statistically significant results, many (thousands) of test runs must be made. This level of testing is starting to be a prerequisite for international publications in this area. It should also be mentioned that when the developed methods are moved into the real world user area, they are still as CPU intensive, and the users will have incentive to use them for what-if scenarios in addition to the normal production runs. Even small improvements in the quality of solutions can have a large impact on the bottom line. I would like to mention that the types of problems described above are the focus of many projects supported by NFR. In the evita program, the DOMinant and DOMinant II projects focus on some of these types of combinatorial optimization problems arising in supply chain optimization. The second area that requires large amounts of computing power is the area of automatic generation of program code. Our setting is using the ADATE system to improve existing state-of-the-art optimization code by genetic programming. The initial program is written in ML and subsequently evolved using program transformations. At any one time during program evolution, around 1000 trial programs are kept, at different levels of semantic complexity. Each newly synthesized program is tested against a training set. The total number of executed programs is typically between O(smax2) and O(smax3), where smax is the maximum size of any program in the kingdom. For such complex program synthesis/transformation tasks as optimizing optimization methods, the required running time can easily exceed 105 cpu hours. In our experience we find that ideally ADATE should run much longer, but that again depends on the availability of spare cpu time.

4 Future needs for einfrastructure in Solar Physics Mats Carlsson, Institute of Theoretical Astrophysics, University of Oslo Scientific challenges The interaction of magnetic fields, hydrodynamics and radiation fields offers one of the great challenges in modern physics, astrophysics and engineering. Situations where magnetic fields and radiative energy transport are expected to play a major role range from galactic to laboratory scales. The Sun is extremely useful in this respect, as an astrophysics laboratory and benchmark case, in that it offers a unique opportunity to observe the detailed time evolution of magnetic structures. Solar magnetism also lies at the root of most solar and heliospheric physics. The intricate structure of the solar magnetic field, the activity cycle and the influence of the magnetic field on the heliosphere represent major quests of (astro-) physics which bear directly on the human environment. The Sun s magnetic field, generated by enigmatic dynamo processes in the solar interior, is organized into the highly complex patterns of solar activity observed in the solar photosphere, dominates the structure of the outer solar atmosphere (chromosphere, transition region, corona), regulates the solar wind, and affects the whole extended heliosphere into the Earth s upper atmosphere. An understanding of these complex phenomena requires combining leading edge observations with detailed radiation magneto-hydrodynamic models. Future needs for einfrastructure (from 2015 and onwards) It is not possible to state that we need this or that amount of computing power there is a need to include better descriptions of the physics (brute force implementation of all the physics we know is relevant would require at least 1010 more computing power than we have today) and make simulations of larger volumes at higher spatial resolution. The need for einfrastructure is thus set by what is available internationally: to be competitive we need access to the same type einfrastructure as other leading groups. This means an increase in computing power following Moore s law. Computing time requirements increase faster than storage requirements (because increased resolution leads to shorter timesteps) so we will be limited by computing cycles also in the future. Requirements for einfrastructure The basic challenge is the complex interplay of several processes acting on a wide range of spatial and time scales. The challenge primarily shows up in the computational volume caused by the resolution and the size of the computational domain needed. The need to cover spatial scales from a few tens to thousands of kilometers and temporal scales from a few milliseconds to hours makes the problems very challenging. Typical state-of-the-art simulations today use a grid of 1000x1000x1000 points and need a 2 million CPU hours (90 days on 1000 cores). The bottle-neck is computing cycles on tightly coupled massively parallel machines (Infiniband or better) while storage is less of a problem (one simulation generates about 14 TB of data). Currently there is no need for advanced user-support.

5 esop e- infrastructures in molecular life sciences FUGE Bioinformatics/ELIXIR.NO groups (Inge Jonassen, Kjell Petersen, Eivind Hovig, Finn Drabløs, Nils Peder Willassen, Sigbjørn Lien) There is a growing need for infrastructure for storing and processing data within the area of molecular life science. One of the feasibility studies performed as part of the preparatory project for ELIXIR (European Bioinformatics Infrastructure) assessed the supercomputing facilities for genomic data ( europe.org/bcms/elixir/ Documents/reports/WP13.2.pdf). One f the conclusions in this report is that a tight collaboration between compute providers and domain experts is needed to enable provision of a useful service. In Norway, one of the steps taken so far is through the StoreBioInfo project where the FUGE Bioinformatics platform is working with NorStore to provide a storage solution for users in the molecular life science domain. We think this effort should be extended to also include computational services and work in this direction is already initiated. We believe there is a need for einfrastructure solutions that can be utilized by groups in the molecular life science area that provides storage and compute resources that can be used in combination with the appropriate tools and data base resources. In order to realize this, tools and databases need to be accessible through standardized interfaces and data needs to be transferred using agreed-- upon exchange formats etc. The European ELIXIR project has decided to go for web services to enable programmatic access to tools and databases and various resources are being developed, e.g., to catalog tools (BioCatalogue), for providing exchange formats (BioXSD in addition to more specialized formats such as MAGE-- ML, PSI-- ML etc), and for annotating the tools and data (EDAM and various OBO ontologies). Development of einfrastructures to be utilized within molecular life sciences should be aligned with those of ELIXIR allowing integration of resources and efficient resource utilization. One needs to provide mechanisms for data protection and authentication of users etc. Project data needs to be kept private to the project and also some data is of a sensitive nature such as personal genomic sequences. For the latter, there is an on-- going project between the FUGE platform in Oslo and USIT/NorStore.

National e-infrastructure for Science. Jacko Koster UNINETT Sigma

National e-infrastructure for Science Jacko Koster UNINETT Sigma 0 Norway: evita evita = e-science, Theory and Applications (2006-2015) Research & innovation e-infrastructure 1 escience escience (or Scientific