Measuring the Impact of Centralized High-Performance Computing with Research Collaboration Networks

Toward the endeavor to develop metrics to assess the impact of technology investment, research collaboration networks can provide a compelling and unique lens. Prior work, assessing networks representing the cumulative collaborations over multiple years, interrogated the distinctions between cohorts of researchers who did and did not engage with centralized computing resources. This work examines yearly networks individually, in the effort to develop a process by which a technology investment event can be contextualized around the evolution of trends in network metrics over time.

The first of two approaches contrasts cohort distributions of node metrics which can be interpreted to correspond to important researcher characteristics, such as collaborativeness, influence and interdisciplinary interaction. The second approach implements an agent-based model on the network to assess idea propagation. Modeling the spreading of an idea as a disease, individual nodes from the two cohorts are assessed by the overall number of researcher nodes in the network who “contract” the idea with infectiousness and recovery time parameters varied. Results are presented from the application of these tools to the Arizona State University (ASU) research network over an eight year period.

Annual networks constructed from ASU publications and award data were employed to assess the impact of investment in compute clusters during this period. Significant differences between cohort distributions were found for collaboration and idea propagation metrics favoring those nodes corresponding to researchers with compute cluster accounts.

Through quantitative measurement of the enhancement of network metrics which correspond to valued research practices, network methods provide effective tools to guide and detect success in cyberinfrastructure investment decisions.