State of the art
The current state of the art can be summarized very simply with the conclusion of the 2010 SciDAC report: "New, scalable, high-performance computational tools are essential" [EXA10]. State of the art simulations in both the biological and materials science domains create new challenges in data mining and exploration. Such computer experiments often lead to a data deluge [BEL09] rendering thorough analysis a difficult and time-consuming task. Furthermore the increase in publicly available experimental data through structural, simulation or experimental databases adds even more valuable information that should be taken into account. Advanced scientific visualization, visual analytics and virtual reality can provide novel and complementary methods to mine such data [FOX11, DES06, KEI10].
Currently, no ready-to-use exascale and visual analysis tools exist for studying nanoscopic molecular structures. Given the numerous reports describing the need for such tools and the grand challenges that are to be overcome [AGE10, CHI10, EXA10, EXA11b, ROB09], we anticipate that ExaVis is a timely proposal in a very competitive field. Even fundamental molecular visualization tools have recently undergone important changes with the advent of ever more powerful GPUs [CHA11a], and we have contributed to scalable high-performance visualization algorithms in this context [CHA10, CHA11b]. The emerging field of visual analytics seeks to provide computer-aided real-time analytic capabilities to improve our understanding of large datasets and render this process more efficient [KEI10]. The European Commission has recently started to support and fund this approach via its VisMaster coordination action. So far, visual analytics has only received little attention in the life and materials sciences. In Europe, the VISUS institute headed by Prof. T Ertl is at the forefront of this research [VISUS], and ExaViz will build upon our existing collaboration that just lead to our first joint publication [CHA11a]. The emergence of exascale simulations also motivates the need for new approaches when coupling simulations and visualization like in situ visualization [YU10].
Although numerous projects have already demonstrated the value of integrating Virtual Reality (VR) techniques to assist researchers in understanding molecular systems of ever-increasing complexity (e.g. [HAA02], [KLO02], [YAN05], [DES06], [KOL08], [DEL09]), the specific question of extending this to the exascale has not been tackled and existing solutions are far from being mature. In France, we find numerous applications of VR in the molecular field, although generally using different angles of approach to that of ExaViz. We can cite as examples various projects, such as Arc Docking [ARC03] focusing on the docking problem in an immersive context, CoRSAIRe [COR05], Amusibio [RED06], Samson [RED08] and FlowVRNano [FVNANO] which deal with the actual generation of primary data via interactive simulations, but not specifically their analysis and post-processing. CollaViz is a visualization-centric collaborative platform project currently under development. It focuses on collaborative and remote applications [COLLAVIZ].
Some of these solutions are very specific and tailored to a single precisely defined application, and some frameworks are more general. With ExaViz, we aim at providing a high level of abstraction with very general tools, specifically addressing molecular applications and driven by state-of-the-art research needs. To the best of our knowledge none of the existing implementations is yet widely used or provides such a level of abstraction. Hence, the problems of scalable visual analysis in a context of VR applied to the molecular field have not yet been dealt with. One of the inherent problems is the lack of clearly defined standards that allow a constructive community-wide approach. In particular most of these projects aim at providing a working application, but don’t allow using a general framework for interfacing one’s own code. This is particularly important as traditionally many post-processing analyses rely on in-house codes.
CAVE VR systems, referred to as “immersive”, allow several users to be immersed at the heart of a common virtual object, providing an ideal context for visual analytics. In particular, adaptive multiuser stereoscopy, a technique available in the recent EVE system from LIMSI, enables each user on site to have its own specific 3D viewpoint of its structure, and to navigate around it in a totally natural manner. This type of system is currently unique in the world if we consider both the functionalities previously described and the large size of the workspace available. EVE will be one of the platforms targeted for ExaViz, in addition to more common desktop use and emerging solutions for mobile devices.
The ExaViz project will be driven by the specific needs of state of the art molecular simulations. Two grand challenge applications were chosen for the life sciences, concerning the visual analysis and post-processing of large-scale molecular dynamics simulations of complex biological systems. In the context of an HPC Europa 2 collaboration with Prof. MSP Sansom's group at Oxford university, we recently setup frontier simulations of an entire influenza virion comprising over 5 million coarse grain particles. This multi-scale model actually represents over 60 million real particles and generated almost a terabyte of compressed data. We have also carried out very long and extensive simulations on the nicotinic receptor homologue GLIC, a major target for the pharmaceutical industry [NUR10, NUR11, BOC09]. The common bottleneck in these projects concerns processing simulation data to make it tractable for analysis and visualization, and in making analysis programs efficient. Analyses of the full systems are extremely difficult, if not impossible, with current tools.
In a more general context, ExaViz combines three key technologies: 1/ human-machine interactions, 2/ modeling, simulation and computation and 3/ virtual and augmented reality and 3D. These technologies were identified as key technologies for 2010 by a report of the French Ministry for the Industry [KEYTECH]. Whereas 1/ has achieved maturity, technologies 2/ and 3/ are still rapidly progressing. A visual analytics framework combining these technologies will allow scientists to keep up to pace with the latest developments and provide applications that correspond to emerging industry-standards.
