Parashar has significantly contributed to catalyzing the transformative impact of high-performance parallel (and distributed) computing (HPC) on science and engineering. Parashar’s pioneering contributions in HPC have enabled new insights through large-scale computations and data in various domains. Specifically, his research has enabled advanced application formulations, such as those based on dynamically adaptive, coupled methods and data-driven workflows, to be implemented on extreme-scale HPC systems. His contributions have included data structure and algorithm innovations, programming abstractions, and runtime systems. He has pioneered using autonomic computing techniques to address application/system complexity and uncertainty. He has also deployed open-source software encapsulating these research innovations, which directly impact a range of applications. Parashar has also demonstrated outstanding leadership and strategic vision at the national, regional, and institutional levels, driving technical and social changes that enable computation/data to have a transformative impact on science and society.
One of Parashar’s early and distinctive contributions is enabling large-scale adaptive applications based on structured adaptive mesh refinement (SAMR). While SAMR techniques can provide attractive cost/accuracy tradeoffs, they present significant challenges. Parashar is one of the earliest researchers to tackle scalable SAMR. His research has included a theoretical framework for locality-preserving distributed and dynamic data structures, programming abstractions that enable distributed, dynamically adaptive formulations, and a family of innovative partitioning algorithms and mechanisms for actively managing SAMR grid hierarchies. Together, these contributions continue to enable scalable SAMR-based applications and have led to realistic simulations of complex phenomena such as, for example, colliding black holes and neutron stars, dynamic response to detonation of energetic materials, forest fire propagation, fluid flows in the subsurface and the human heart, etc. These innovations have been deployed to the broader community through software systems such as DAGH/GrACE, which international application groups use, including very high-profile US DOE and NSF projects. Furthermore, the conceptual framework underlying DAGH/GrACE continues to enable/inspire leading SAMR frameworks/applications and has been incorporated into existing frameworks. For example, they have influenced the codebases used in the research efforts that were part of the 2017 detection of gravitational waves (Physics Nobel Prize).
Complementing this work, Parashar’s more recent research has targeted the dynamic interactions and data-intensive couplings required by application workflows involving multiple interacting processes. Since these processes can run at different spatial and temporal scales, levels of parallelism and locations, expressing required couplings using existing programming systems is extremely challenging. Parashar’s research has developed dynamic, semantically specialized shared-space abstractions, which enables the online indexing of large data streams so that they can be flexibly queried to support runtime coupling, interactions, and coordination. This work has been deployed in software systems including the DataSpaces data staging service, which is being used on some of the largest systems in the world. For example, DataSpaces has enabled large-scale tightly coupled parallel fusion simulations. DataSpaces also supports in-situ/in-transit data processing and analytics and uses machine-learning-based autonomic runtime approaches to manage extreme-heterogeneity and application-level dynamism. DataSpaces, as part of ADIOS, was awarded a 2013 R&D 100 Award.
Parashar’s other contributions include deploying large scale production systems, such as the cyberinfrastructure for the NSF Ocean Observatories Initiative, the largest ocean-observing network in the world. His current research is leveraging streaming data and the computing continuum for urgent, socially important applications, and developing intelligent data-delivery mechanism to democratize data-driven science. He is also driving Translational Computer Science as a paradigm for transdisciplinary research with integrated real-world impact.
