The Display Wall

The High Performance Distributed Systems (HPDS) group is located at the Department of Computer Science at the University of Tromsø. The display wall laboratory is the cornerstone of the HPDS-group. The display wall laboratory was built in 2004. The group's display wall is a wall-sized, high-resolution tiled display built using 28 projectors driven by a 28-node display cluster for a total resolution of 22 megapixels.

Several other compute resources are used to create and deliver content for the display wall, including two 30-node clusters of workstations, the University of Tromsø Computing Centre's large 5632-core compute cluster and resources available at the University of Copenhagen, Princeton University, and Department of Chemistry, Tromsø.

The display wall uses a camera-based system to enable multi-user, touch- and device-free interaction with applications running on the display wall. Applications range from visualization of experiment results, display and navigation of gigapixel images to games and videos.

The display wall used to visualize the Earth.

The group has published a number of videos demonstrating the display wall in use browsing both large collections of comics, very high-resolution (gigapix-scale) images and more. Take a look at the video page for a complete listing.

Systems Research

The group's research activities are organized into research areas and projects. A project typically includes several of the research areas. The group does systems research, the scientific study, analysis, modelling and engineering of effective software platforms. Systems research produces artefacts in the form of functioning prototypes of systems. Consequently, the group researches and develops ideas, architectures, designs, and implementations. The group's methodology is to design and conduct experiments on the systems it develops. Using results from these experiments, each system's characteristics and actual behaviour is documented, enabling comparisons to other state of the art systems and identification of lessons learned on how to do better.

High-performance visualization at high resolutions

The group has advanced the state of the art with its research on pixel distribution, compression, and caching to increase the visualization performance of typical standard desktop applications enabling them to visualize their output at an order of magnitude better resolution than what a single PC can support. The group has also developed several display-wall aware applications and systems to experimentally document how to do better with regards to both frame-rates and processing than existing architectures and designs for high performance visualizations. The systems developed are used in various applications, including games, in an application supporting interactive viewing and navigation of a tiled 13 gigapixel image over the City of Tromsø, and in a system transforming a standard 1.2 megapixel laptop into a 22 megapixel laptop.

The 22 megapixel laptop.

Collaborative systems and resource sharing

A user typically needs to use many applications distributed over several computers. Today there is weak support for placeshifting the output from the applications to whichever computer the user happens to use at the moment. The group has developed scalable and flexible approaches and systems to do placeshifting and documented their characteristics. The group has also developed several approaches to placeshift user and device input events between computers so that devices across a network flexibly can be used by (typically remote) applications. To support collaboration in meeting room and teaching environments the group developed the Network Accessible Display (NAD) model where displays and projectors are available as network accessible resources to be used by one or several users at the same time through a network.

Systems support for multi-user distributed human-computer interfaces

Using keyboards and mice are impractical when users walk along a wall-sized display. The group has researched and implemented an optical, sensor-based object detection and localisation system that accurately locates objects in 3D in front of a display wall. The system is used to enable multi-user touch- and device-free interaction with applications using the display wall. The system comprises 16 cheap commodity web cameras and nine computers. A second system enables interaction with the display wall through sounds made by users, by identifying the source location of the sound in the room.

The group has developed the Tromsø Touch Table (T3), a horizontal rear projection screen with a multi-touch user interface using four infrared lasers and a camera. The tables enhance and complement the display wall by being movable and horizontal, extending the display wall into the physical third dimension. These tables are also used to distribute the user interface for the display wall around the room and to remote sites.

Painting fire on the display wall using the touch- and device-free Camera-Sense system.

Parallel, multi- and many-core systems

The group conducts research on load-time and run-time orchestration of applications, including instrumentation along the communication paths of a parallel multi-process and multi-threaded application. Research is also conducted on parallel programming models simplifying parallel programming for non-expert programmers from other scientific fields, and for teaching concurrency to students using a high-level language. An implementation, PyCSP, is publicly available and in use for both research and education.

The group does research on novel inter-thread coordination mechanisms for multi- and many-core architectures that improve both parallelism and fault-tolerance without unduly increasing the complexity faced by programmers. The group develops scalable novel inter-core communication mechanisms, and dependable and fault-tolerant novel memory access mechanisms utilizing the high inter-core bandwidth and low inter-core latency of many-core architectures.

An implementation of Boids called Occoids written in occam-π running on the display wall.

Cross-disciplinary Weather forecasting

To get useful level of detail in a weather forecast, the size of the forecasted geographical region is traded off against the available compute resources. Even when using a supercomputer, the level of detail for a large region becomes rather low. In cooperation with the Meteorological Institute of Norway the group has developed a prototype for on-demand production of highly detailed weather forecasts for a user-selectable small region. A user can select any small region from a high-performance high-resolution visualization of the earth on the display wall, and see a short-term high-resolution forecast visualized on the display wall less than three minutes later. A state-of-the-art and freely available numerical weather forecasting model, WRF, is used to do the actual forecast. Because of the small size of the region, the computation can be done using just a few PCs or compute cores.

Weather forecasting on the display wall.

Biological Data Processing Systems Lab

The big data era in molecular biology has created exiting potential for novel biological discoveries, but also exiting challenges for computer scientists in data management, processing, and visualization. In the next decades there will be developed sophisticated bioinformatics methods and framework to analyze and explore the information in the data. However, these will require development of novel infrastructure systems targeted for bioinformatics data and methods.

Our research goal is to build and experimentally evaluate infrastructure systems that support the methods under development by our bioinformatics collaborators. We are designing and implementing systems for big data storage, interactive analysis, and large-scale visualization. We are primarily interested in improving the scalability and interactivity of bioinformatics analysis methods and frameworks.

We combine experimental computer science with real problems, applications, and data obtained from our biology collaborators. We focus on distributed and parallel systems, including high-resolution visualizations.

For additional details, please visit the lab website, or contact the principal investigator.

Gene expression analysis on a display wall.

Cross-disciplinary Opera

The group is researching how to do systems support for a distributed opera with live performances taking place at several geographically distant locations, as part of the Verdione project, funded by the Norwegian Research Council. Partners include both national and international artists, conductors, and composers (under The World Opera organization) in addition to companies and computer science departments and laboratories in Tromsø and Oslo.

The Tromsø Touch Table being used to investigate ways of handling latency in the World Opera context.