First Principles and Theoretical Chemistry communities have long histories
of advances and expertise in high performance computing. However, higher
performance algorithms, mature community codes, and inter-laboratory
collaborations expose systemic problems and several difficulties to the
successful execution of large-scale projects. For instance,
current collaborations may involve researchers from various institutions,
sharing large spatio-temporal data sets generated from several hundreds of
calculations submitted manually on a variety of physically distributed
parallel platforms. Results are poorly annotated and interfaces are not
standardized. Moreover, a newcomer, even if familiar with the topic in
question, is subjected to a steep learning curve before acquiring the
skills necessary to use the methods, analyze the outcome, and produce
quality results. These difficulties can be eliminated in some cases and
greatly alleviated in many others by a large-scale project manager to
handle interfaces, task automation, scheduling, visualization, and
collaborative tools among the participants of a single project.
We propose to build tools that will allow for cooperative, group-based
work in all stages of a project: preparation, submission, monitoring,
computational and visual steering, and analysis of results. These tools
will be connected to a novel grid system based on a publish/subscribe
architecture. This collection of tools, services, and the interconnecting
framework will constitute the VLab. These tools will not only be fully
exploited by experienced researchers collaborating in large scale
projects, but will also make the VLab an interactive facility for distant
non-specialist users and permit monitoring of their work by the VLab
support staff.
